And such a system will be humongous, costing an arm & a leg and no software that can fully utilize it. It won't sell. That's the same story for the 64-core Threadripper. Every "top-end" hardware is more or less a showpiece, more of "we did it!" rather than being practical, hence why there are no Xeon workstations to follow.
A 4-die M3 will challenge them at the practical level, likely offering more performance & less heat, that's all.
They currently sell such a system in the 2019 Mac Pro. It has room for a 300W CPU, 4 250W GPUs, 2 3.5" HDD, 4 1-wide PCI cards, and over 1 TB of memory. It allows for 10 3.5" HDD drives to be put inside, PCIe storage at all price points, additional ports, capture cards, so on and so forth. The two NAND storage cards are not worth mentioning for a box size argument. So, there is a value in it, even at its $6000 starting price. The modularity is a feature a lot of buyers pay for even if they don't do much with it.
Then, Apple really haven't demonstrated that they will have lower prices with Apple Silicon. They are using all the same price tiers give or take 10%. So this new smaller Mac Pro could start at $6000, and have very little to zero modularity. There's the rumor that Apple will continue on with the 2019 Mac Pro with Xeon W-3300 series processors (up to 38 cores), and presumably new AMD GPUs. That Intel update may result in an even higher price point ($8000 for 12c model?) to make room for an Apple Silicon Mac Pro at $6000. It would be amazing that they would do this as the Xeon W-3300 series uses a different socket compared to the Xeon W-3200 series in the current model. So a whole new Intel motherboard to design and qualify. Odd rumor as they could use all those resources on an Apple Silicon motherboard that could fit in the 2019 Mac Pro.
Lastly, what is a buyer of this smaller Mac Pro really buying if it is a little smaller than a G4 cube? I don't think being quiet and using less energy is high on the list of features for the buyers of this type of machine. They want the most performance, most storage, most flexibility in a box volume that can fit on a desk, and in today's world, in a rack is hugely important too. Intel competitors will be able to match Apple Silicon performance at higher power consumption, and they will have more storage and more modularity, and will be at similar prices. These types of features are more important in this market than being small and quiet. Hard to see how this will not be a repeat of the G4 cube and the 2013 Mac Pro.
Okay, there's no such thing as "the most performance", we all need to make that decision. I find it ironic that people who say that are always the 1st to make that decision, because faster computers always exist until you hit TOP100, well then where are you going to stop? I guess we can define that everything less than TianHe 1 is non-pro? That's nuts.
It's already been said. Desktop computers have a high gate to cross at around 1500 Watts power draw. Most of the USA's power circuits in offices and desks are going to be limited by 1650 Watt power circuits. The 2019 Mac Pro uses a 1400 W power supply. So, design an Apple Silicon desktop around 1200 to 1400 W, and fill in the chips accordingly. Have it idle at less than 50 W, and at full power draw, around 1000 to 1200 W. Don't leave performance off the desk in the workstation Mac.
x86 workstations are basically limited by these circuits, but they all will edge to the line. They will have 1 or 2-socket Xeon boards with support for 3 to 4 GPUs hitting somewhere between 1200 to 1500 W. And, they will have room for lots of HDD storage and other cards.
There’s a point where extra performance doesn’t matter. You will get something twice as expensive and hot with marginal gain. A 4-die is plenty to compete with a 56-core Xeon (assuming 48+16 M3) and Zen4 alike, most likely much better. There’s no law that states a workstation has to match 1,500 watt, which is overkill for an Apple silicon system. Apple is no AMD, they don’t need to build a show piece just to claim a title. We know what the M1 can do.
There are applications that can make use of the compute performance, especially GPU compute applications. Basically any STEM fields using numerical methods (3D, CFD et al) have infinite compute requirements. If what used to be jobs needed to be done on a compute cluster can be done on a desktop, it will increase efficiency and quality of work by giving analysts more time and data to assess their problems. It's a continuum of classes of problems whose analytical fidelity only improves with more compute power.
There aren't any laws about 1500 W, but it's a practical limit for how much power is available to desks in offices, whether at work or at home. Workstation vendors design to it be becausd it's a practical limit. The Mac Pro is Apple's workstation machine whose job "is to challenge what we think a computer can do and do things that no computer has ever done before, be more and more powerful and capable so that we need a desktop because of its capabilities,” says Schiller. “Because if all it’s doing is competing with the notebook and being thinner and lighter, then it doesn’t need to be.” It's interesting that Schiller made this statement before saying the 2019 Mac Pro will be modular. I would have thought they learned that being modular is part of the job of a big desktop. It's not just compute performance. Being able to put 50+ TB of storage into the box is part of it. Being able to run a PCIe card for some job that Apple didn't anticipate is part of it.
There are trades regarding whether a user should just buy some compute hardware in the cloud or have it in a server room and run their jobs from a thin client, but it isn't a binary one way of the other. Some are fine with a thin client with a server room. Some really need all the compute and storage in a single box that they can get. Apple really needs a continuum of machines, more than they have today.
I don't know any STEM fields are so eager to buy a 3990X/5990WX workstation. If not, they're better served with clusters.
That's just a concept as no ITX can serve a TR4. Now we're talking about M3 with 4 dies and 16 RAM chips. Then throw two, huge air coolers. With more than 8 PCIe slots.
That's a minicomputer dude, without the power of a minicomputer.
And such a system will be humongous, costing an arm & a leg and no software that can fully utilize it. It won't sell. That's the same story for the 64-core Threadripper. Every "top-end" hardware is more or less a showpiece, more of "we did it!" rather than being practical, hence why there are no Xeon workstations to follow.
A 4-die M3 will challenge them at the practical level, likely offering more performance & less heat, that's all.
They currently sell such a system in the 2019 Mac Pro. It has room for a 300W CPU, 4 250W GPUs, 2 3.5" HDD, 4 1-wide PCI cards, and over 1 TB of memory. It allows for 10 3.5" HDD drives to be put inside, PCIe storage at all price points, additional ports, capture cards, so on and so forth. The two NAND storage cards are not worth mentioning for a box size argument. So, there is a value in it, even at its $6000 starting price. The modularity is a feature a lot of buyers pay for even if they don't do much with it.
Then, Apple really haven't demonstrated that they will have lower prices with Apple Silicon. They are using all the same price tiers give or take 10%. So this new smaller Mac Pro could start at $6000, and have very little to zero modularity. There's the rumor that Apple will continue on with the 2019 Mac Pro with Xeon W-3300 series processors (up to 38 cores), and presumably new AMD GPUs. That Intel update may result in an even higher price point ($8000 for 12c model?) to make room for an Apple Silicon Mac Pro at $6000. It would be amazing that they would do this as the Xeon W-3300 series uses a different socket compared to the Xeon W-3200 series in the current model. So a whole new Intel motherboard to design and qualify. Odd rumor as they could use all those resources on an Apple Silicon motherboard that could fit in the 2019 Mac Pro.
Lastly, what is a buyer of this smaller Mac Pro really buying if it is a little smaller than a G4 cube? I don't think being quiet and using less energy is high on the list of features for the buyers of this type of machine. They want the most performance, most storage, most flexibility in a box volume that can fit on a desk, and in today's world, in a rack is hugely important too. Intel competitors will be able to match Apple Silicon performance at higher power consumption, and they will have more storage and more modularity, and will be at similar prices. These types of features are more important in this market than being small and quiet. Hard to see how this will not be a repeat of the G4 cube and the 2013 Mac Pro.
Okay, there's no such thing as "the most performance", we all need to make that decision. I find it ironic that people who say that are always the 1st to make that decision, because faster computers always exist until you hit TOP100, well then where are you going to stop? I guess we can define that everything less than TianHe 1 is non-pro? That's nuts.
It's already been said. Desktop computers have a high gate to cross at around 1500 Watts power draw. Most of the USA's power circuits in offices and desks are going to be limited by 1650 Watt power circuits. The 2019 Mac Pro uses a 1400 W power supply. So, design an Apple Silicon desktop around 1200 to 1400 W, and fill in the chips accordingly. Have it idle at less than 50 W, and at full power draw, around 1000 to 1200 W. Don't leave performance off the desk in the workstation Mac.
x86 workstations are basically limited by these circuits, but they all will edge to the line. They will have 1 or 2-socket Xeon boards with support for 3 to 4 GPUs hitting somewhere between 1200 to 1500 W. And, they will have room for lots of HDD storage and other cards.
There’s a point where extra performance doesn’t matter. You will get something twice as expensive and hot with marginal gain. A 4-die is plenty to compete with a 56-core Xeon (assuming 48+16 M3) and Zen4 alike, most likely much better. There’s no law that states a workstation has to match 1,500 watt, which is overkill for an Apple silicon system. Apple is no AMD, they don’t need to build a show piece just to claim a title. We know what the M1 can do.
There are applications that can make use of the compute performance, especially GPU compute applications. Basically any STEM fields using numerical methods (3D, CFD et al) have infinite compute requirements. If what used to be jobs needed to be done on a compute cluster can be done on a desktop, it will increase efficiency and quality of work by giving analysts more time and data to assess their problems. It's a continuum of classes of problems whose analytical fidelity only improves with more compute power.
There aren't any laws about 1500 W, but it's a practical limit for how much power is available to desks in offices, whether at work or at home. Workstation vendors design to it be becausd it's a practical limit. The Mac Pro is Apple's workstation machine whose job "is to challenge what we think a computer can do and do things that no computer has ever done before, be more and more powerful and capable so that we need a desktop because of its capabilities,” says Schiller. “Because if all it’s doing is competing with the notebook and being thinner and lighter, then it doesn’t need to be.” It's interesting that Schiller made this statement before saying the 2019 Mac Pro will be modular. I would have thought they learned that being modular is part of the job of a big desktop. It's not just compute performance. Being able to put 50+ TB of storage into the box is part of it. Being able to run a PCIe card for some job that Apple didn't anticipate is part of it.
There are trades regarding whether a user should just buy some compute hardware in the cloud or have it in a server room and run their jobs from a thin client, but it isn't a binary one way of the other. Some are fine with a thin client with a server room. Some really need all the compute and storage in a single box that they can get. Apple really needs a continuum of machines, more than they have today.
I don't know any STEM fields are so eager to buy a 3990X/5990WX workstation. If not, they're better served with clusters.
All PC vendors, including Apple, sell Xeon and or Threadripper (much less SKUs compared to Xeon SKUs) desktop workstations, with capability to go to 3 or 4 GPUs. They aren't doing it for kicks. The 3990X seems limited by memory performance and capacity so perhaps that's why it has slow uptake relative to Xeons. I know of people who have CAD loads that are limited by their hardware, simulation folks whose light loads can now be done on a workstation versus a server, visualization loads that need lots of GPU compute+memory, and GPU compute loads that are now becoming more and more common. It's a continuum of loads from the cluster, workstation to laptop. It's not a binary choice that once a compute load is above a certain amount, it should be done on a server. People optimize based on what they do, and for some, having big workstation on the desk is the way to go, in addition to access to clusters, and laptops.
There is a software networking effect as well. Having software support sells hardware up and down the line. For this type of software, missing the top end effectively limits the size of the market can sell devices into. They had to know this once they committed to making the 2019 Mac Pro in 2017. They are way behind CUDA with GPU compute optimized software. Lots of work for them to do with Metal to catch up, along with other areas.
And such a system will be humongous, costing an arm & a leg and no software that can fully utilize it. It won't sell. That's the same story for the 64-core Threadripper. Every "top-end" hardware is more or less a showpiece, more of "we did it!" rather than being practical, hence why there are no Xeon workstations to follow.
A 4-die M3 will challenge them at the practical level, likely offering more performance & less heat, that's all.
They currently sell such a system in the 2019 Mac Pro. It has room for a 300W CPU, 4 250W GPUs, 2 3.5" HDD, 4 1-wide PCI cards, and over 1 TB of memory. It allows for 10 3.5" HDD drives to be put inside, PCIe storage at all price points, additional ports, capture cards, so on and so forth. The two NAND storage cards are not worth mentioning for a box size argument. So, there is a value in it, even at its $6000 starting price. The modularity is a feature a lot of buyers pay for even if they don't do much with it.
Then, Apple really haven't demonstrated that they will have lower prices with Apple Silicon. They are using all the same price tiers give or take 10%. So this new smaller Mac Pro could start at $6000, and have very little to zero modularity. There's the rumor that Apple will continue on with the 2019 Mac Pro with Xeon W-3300 series processors (up to 38 cores), and presumably new AMD GPUs. That Intel update may result in an even higher price point ($8000 for 12c model?) to make room for an Apple Silicon Mac Pro at $6000. It would be amazing that they would do this as the Xeon W-3300 series uses a different socket compared to the Xeon W-3200 series in the current model. So a whole new Intel motherboard to design and qualify. Odd rumor as they could use all those resources on an Apple Silicon motherboard that could fit in the 2019 Mac Pro.
Lastly, what is a buyer of this smaller Mac Pro really buying if it is a little smaller than a G4 cube? I don't think being quiet and using less energy is high on the list of features for the buyers of this type of machine. They want the most performance, most storage, most flexibility in a box volume that can fit on a desk, and in today's world, in a rack is hugely important too. Intel competitors will be able to match Apple Silicon performance at higher power consumption, and they will have more storage and more modularity, and will be at similar prices. These types of features are more important in this market than being small and quiet. Hard to see how this will not be a repeat of the G4 cube and the 2013 Mac Pro.
Okay, there's no such thing as "the most performance", we all need to make that decision. I find it ironic that people who say that are always the 1st to make that decision, because faster computers always exist until you hit TOP100, well then where are you going to stop? I guess we can define that everything less than TianHe 1 is non-pro? That's nuts.
It's already been said. Desktop computers have a high gate to cross at around 1500 Watts power draw. Most of the USA's power circuits in offices and desks are going to be limited by 1650 Watt power circuits. The 2019 Mac Pro uses a 1400 W power supply. So, design an Apple Silicon desktop around 1200 to 1400 W, and fill in the chips accordingly. Have it idle at less than 50 W, and at full power draw, around 1000 to 1200 W. Don't leave performance off the desk in the workstation Mac.
x86 workstations are basically limited by these circuits, but they all will edge to the line. They will have 1 or 2-socket Xeon boards with support for 3 to 4 GPUs hitting somewhere between 1200 to 1500 W. And, they will have room for lots of HDD storage and other cards.
There’s a point where extra performance doesn’t matter. You will get something twice as expensive and hot with marginal gain. A 4-die is plenty to compete with a 56-core Xeon (assuming 48+16 M3) and Zen4 alike, most likely much better. There’s no law that states a workstation has to match 1,500 watt, which is overkill for an Apple silicon system. Apple is no AMD, they don’t need to build a show piece just to claim a title. We know what the M1 can do.
There are applications that can make use of the compute performance, especially GPU compute applications. Basically any STEM fields using numerical methods (3D, CFD et al) have infinite compute requirements. If what used to be jobs needed to be done on a compute cluster can be done on a desktop, it will increase efficiency and quality of work by giving analysts more time and data to assess their problems. It's a continuum of classes of problems whose analytical fidelity only improves with more compute power.
There aren't any laws about 1500 W, but it's a practical limit for how much power is available to desks in offices, whether at work or at home. Workstation vendors design to it be becausd it's a practical limit. The Mac Pro is Apple's workstation machine whose job "is to challenge what we think a computer can do and do things that no computer has ever done before, be more and more powerful and capable so that we need a desktop because of its capabilities,” says Schiller. “Because if all it’s doing is competing with the notebook and being thinner and lighter, then it doesn’t need to be.” It's interesting that Schiller made this statement before saying the 2019 Mac Pro will be modular. I would have thought they learned that being modular is part of the job of a big desktop. It's not just compute performance. Being able to put 50+ TB of storage into the box is part of it. Being able to run a PCIe card for some job that Apple didn't anticipate is part of it.
There are trades regarding whether a user should just buy some compute hardware in the cloud or have it in a server room and run their jobs from a thin client, but it isn't a binary one way of the other. Some are fine with a thin client with a server room. Some really need all the compute and storage in a single box that they can get. Apple really needs a continuum of machines, more than they have today.
I don't know any STEM fields are so eager to buy a 3990X/5990WX workstation. If not, they're better served with clusters.
All PC vendors, including Apple, sell Xeon and or Threadripper (much less SKUs compared to Xeon SKUs) desktop workstations, with capability to go to 3 or 4 GPUs. They aren't doing it for kicks. The 3990X seems limited by memory performance and capacity so perhaps that's why it has slow uptake relative to Xeons. I know of people who have CAD loads that are limited by their hardware, simulation folks whose light loads can now be done on a workstation versus a server, visualization loads that need lots of GPU compute+memory, and GPU compute loads that are now becoming more and more common. It's a continuum of loads from the cluster, workstation to laptop. It's not a binary choice that once a compute load is above a certain amount, it should be done on a server. People optimize based on what they do, and for some, having big workstation on the desk is the way to go, in addition to access to clusters, and laptops.
There is a software networking effect as well. Having software support sells hardware up and down the line. For this type of software, missing the top end effectively limits the size of the market can sell devices into. They had to know this once they committed to making the 2019 Mac Pro in 2017. They are way behind CUDA with GPU compute optimized software. Lots of work for them to do with Metal to catch up, along with other areas.
Intel does have 56-core Xeon, never make it to workstations. This isn’t a technical issue but marketing, the truth is a 64-core almost never sales. We can keep set a higher requirement but in reality every workload have that return curve.
This is also why the Mac Pro is down from two CPUs to one. You just don’t have the performance requirements that makes it necessary, 4 GPUs are also sufficient for workstations load, and Apple needs to make a product that sells.
You’ll never be enough if you just chasing the “pure performance”, someday you’re gonna build a supercomputer.
This is also why the Mac Pro is down from two CPUs to one. You just don’t have the performance requirements that makes it necessary, 4 GPUs are also sufficient for workstations load, and Apple needs to make a product that sells.
You’ll never be enough if you just chasing the “pure performance”, someday you’re gonna build a supercomputer.
The race doesn't end. Year over year, Apple has to continually improve the performance of all its devices, and what required a rack 10 years ago is now available in a desktop on your desktop. They always have to design in performance to all practical limits for the Mac Pro. The software will catch up. New software will eventually utilize the performance.
Yes, Apple has to choose a design point for the Apple Silicon Mac Pro, just as they did for the 2019 Mac Pro with using 1 CPU socket versus 2. (I saw that as a problem back then too). The rumors so far sound like the Apple Silicon Mac Pro will be a 500 W SFF workstation like the 2013 Mac Pro or something the size of a G4 cube. Those machines weren't successful. How is the Apple Silicon version going to be any different? Being quiet and small really isn't a big selling feature for buyers of this type of machine. x86 competitors will match or exceed its performance at 1000 to 1500 W.
We will have the same conversation that we did when the 2013 Mac Pro came out, if this Apple Silicon Mac Pro is a SFF workstation. What's going to be different this time? You say Apple needs to make a product that sells. The G4 cube and 2013 Mac Pro didn't sell and were retired after 2 or 3 years. If the Apple Silicon Mac Pro are like them: small, limited capacity, not modular, what's going to be different this time?
This is also why the Mac Pro is down from two CPUs to one. You just don’t have the performance requirements that makes it necessary, 4 GPUs are also sufficient for workstations load, and Apple needs to make a product that sells.
You’ll never be enough if you just chasing the “pure performance”, someday you’re gonna build a supercomputer.
The race doesn't end. Year over year, Apple has to continually improve the performance of all its devices, and what required a rack 10 years ago is now available in a desktop on your desktop. They always have to design in performance to all practical limits for the Mac Pro. The software will catch up. New software will eventually utilize the performance.
Yes, Apple has to choose a design point for the Apple Silicon Mac Pro, just as they did for the 2019 Mac Pro with using 1 CPU socket versus 2. (I saw that as a problem back then too). The rumors so far sound like the Apple Silicon Mac Pro will be a 500 W SFF workstation like the 2013 Mac Pro or something the size of a G4 cube. Those machines weren't successful. How is the Apple Silicon version going to be any different? Being quiet and small really isn't a big selling feature for buyers of this type of machine. x86 competitors will match or exceed its performance at 1000 to 1500 W.
We will have the same conversation that we did when the 2013 Mac Pro came out, if this Apple Silicon Mac Pro is a SFF workstation. What's going to be different this time? You say Apple needs to make a product that sells. The G4 cube and 2013 Mac Pro didn't sell and were retired after 2 or 3 years. If the Apple Silicon Mac Pro are like them: small, limited capacity, not modular, what's going to be different this time?
1.5KW is PSU only, not the system itself. Workstations need headroom, most of the power doesn't just go to CPU/GPU.
A 4-die will be a ~400W chip. This is Apple Silicon we're talking about, able to match & exceed something twice as power-hungry. The last 3275M is about 280W running P95.
Then realized that AMD only offers a 32/64 lineup, there's no in-between. Either you got 32-core or go straight to the best.
The race doesn't end. Year over year, Apple has to continually improve the performance of all its devices, and what required a rack 10 years ago is now available in a desktop on your desktop. They always have to design in performance to all practical limits for the Mac Pro.
The race to improve performance slows down over time in sync with lowered demand. I wouldn't say they have to design a computer to its physical limits but to the needs of the users of each product. Apple could design much larger notebooks that outclass heavy gaming laptops or thicker iMacs but they choose not to.
We will have the same conversation that we did when the 2013 Mac Pro came out, if this Apple Silicon Mac Pro is a SFF workstation. What's going to be different this time? You say Apple needs to make a product that sells. The G4 cube and 2013 Mac Pro didn't sell and were retired after 2 or 3 years. If the Apple Silicon Mac Pro are like them: small, limited capacity, not modular, what's going to be different this time?
The big difference will be performance per dollar. Some people make an assumption that the Cube was wrong, 2013 model was wrong, iMac Pro was wrong and the 2019 model was right based on size and upgrades but there's nothing to indicate the 2019 model was any less of a failure than the others as far as sales performance goes. The market at these price points is minuscule compared to the mainstream by at least a factor of 100.
What always affects users in higher-end models is performance per dollar. Apple's top-end models that can be $40k or more perform worse than Threadripper machines all because of Intel. Apple's own silicon can outperform even AMD at way more competitive price points.
If they stick at half the power level of the competition then in a few years time, the competition can catch or exceed them and for a fraction of a fraction of the users, they will make a choice to switch platform but that's assuming they need 2x the performance in a single workstation. If they only need 2x the performance and their use case can be networked, they can buy 2 or more.
Over the years, computers like the 2013 model, Cube, iMac Pro would run into limitations like handling 4k/8k editing. This isn't going to continue to be a problem, the video editing industry isn't going to move the mainstream beyond the limits of visual acuity for authoring. Once the performance demand in the video space dies down, that cuts out most of the demand for higher-end hardware and a factor of 2x difference won't be a deal-breaker for anyone, especially when Apple's custom media encoders outclass everything by a significant margin.
The race doesn't end. Year over year, Apple has to continually improve the performance of all its devices, and what required a rack 10 years ago is now available in a desktop on your desktop. They always have to design in performance to all practical limits for the Mac Pro.
The race to improve performance slows down over time in sync with lowered demand. I wouldn't say they have to design a computer to its physical limits but to the needs of the users of each product. Apple could design much larger notebooks that outclass heavy gaming laptops or thicker iMacs but they choose not to.
We will have the same conversation that we did when the 2013 Mac Pro came out, if this Apple Silicon Mac Pro is a SFF workstation. What's going to be different this time? You say Apple needs to make a product that sells. The G4 cube and 2013 Mac Pro didn't sell and were retired after 2 or 3 years. If the Apple Silicon Mac Pro are like them: small, limited capacity, not modular, what's going to be different this time?
The big difference will be performance per dollar. Some people make an assumption that the Cube was wrong, 2013 model was wrong, iMac Pro was wrong and the 2019 model was right based on size and upgrades but there's nothing to indicate the 2019 model was any less of a failure than the others as far as sales performance goes. The market at these price points is minuscule compared to the mainstream by at least a factor of 100.
What always affects users in higher-end models is performance per dollar. Apple's top-end models that can be $40k or more perform worse than Threadripper machines all because of Intel. Apple's own silicon can outperform even AMD at way more competitive price points.
If they stick at half the power level of the competition then in a few years time, the competition can catch or exceed them and for a fraction of a fraction of the users, they will make a choice to switch platform but that's assuming they need 2x the performance in a single workstation. If they only need 2x the performance and their use case can be networked, they can buy 2 or more.
Over the years, computers like the 2013 model, Cube, iMac Pro would run into limitations like handling 4k/8k editing. This isn't going to continue to be a problem, the video editing industry isn't going to move the mainstream beyond the limits of visual acuity for authoring. Once the performance demand in the video space dies down, that cuts out most of the demand for higher-end hardware and a factor of 2x difference won't be a deal-breaker for anyone, especially when Apple's custom media encoders outclass everything by a significant margin.
Not unless Apple shoves two Max dies in their notebook ;-)
There’s the feeling that Apple needs to be the best or we’ll be shamed to oblivion. The PC community acts like that, it’s all about the “fastest on the planet” because “it’s better”. There’s really no end of that. There’s a belt curve for everything, which at some point the gain becomes a dread that doesn’t worth to keep it. I don’t recall anyone wants to live in a hassle with little to no return, that’s what exactly happened in workstation chips.
The race doesn't end. Year over year, Apple has to continually improve the performance of all its devices, and what required a rack 10 years ago is now available in a desktop on your desktop. They always have to design in performance to all practical limits for the Mac Pro.
The race to improve performance slows down over time in sync with lowered demand. I wouldn't say they have to design a computer to its physical limits but to the needs of the users of each product. Apple could design much larger notebooks that outclass heavy gaming laptops or thicker iMacs but they choose not to.
The race only stops for a particular type of software. Like maybe a word processor or presentation app, but new types of applications come in and require more performance. This is true even in specific types of software. What required increasing performance may have plateaued but some other aspect of that type of software ended up driving the need for more performance. Like maybe 8K is it for video production. Just dealing with the pixels. But with machine learning evaluating the content of every frame, and making some kind of change to every frame, it starts the race over again.
And yup, Apple has chosen not to address certain markets like "workstation" laptops that basically run 200 W worth of compute performance, or a consumer desktop box, or an MBA15, so on and so forth. We're only discussing where we think they are making mistakes or not. I think they have kept their lineup too simple and could address more of the market. The SFF Mac Pro is something they have tried before, twice depending on how you count the G4 cube, and I'm more skeptical then inviting this time around.
We will have the same conversation that we did when the 2013 Mac Pro came out, if this Apple Silicon Mac Pro is a SFF workstation. What's going to be different this time? You say Apple needs to make a product that sells. The G4 cube and 2013 Mac Pro didn't sell and were retired after 2 or 3 years. If the Apple Silicon Mac Pro are like them: small, limited capacity, not modular, what's going to be different this time?
The big difference will be performance per dollar. Some people make an assumption that the Cube was wrong, 2013 model was wrong, iMac Pro was wrong and the 2019 model was right based on size and upgrades but there's nothing to indicate the 2019 model was any less of a failure than the others as far as sales performance goes. The market at these price points is minuscule compared to the mainstream by at least a factor of 100.
What always affects users in higher-end models is performance per dollar. Apple's top-end models that can be $40k or more perform worse than Threadripper machines all because of Intel. Apple's own silicon can outperform even AMD at way more competitive price points.
If they stick at half the power level of the competition then in a few years time, the competition can catch or exceed them and for a fraction of a fraction of the users, they will make a choice to switch platform but that's assuming they need 2x the performance in a single workstation. If they only need 2x the performance and their use case can be networked, they can buy 2 or more.
Over the years, computers like the 2013 model, Cube, iMac Pro would run into limitations like handling 4k/8k editing. This isn't going to continue to be a problem, the video editing industry isn't going to move the mainstream beyond the limits of visual acuity for authoring. Once the performance demand in the video space dies down, that cuts out most of the demand for higher-end hardware and a factor of 2x difference won't be a deal-breaker for anyone, especially when Apple's custom media encoders outclass everything by a significant margin.
I think Apple loses on performance per dollar in desktop and high end hardware. They win on performance per Watt, but performance per dollar? No. They will maximize their prices so that performance/$ is about the same as x86 competitors, while x86 competitors are much more price elastic. As their usual practice, Apple will likely use their industrial design intangibles to keep prices high even when x86 has better performance/$.
If Apple uses half the power, they won't have a lead in performance, right? x86 can match the performance of M1 Mac Duo and Quad configurations with 16, 32, 64 core machines with dGPUs. They'll be using 2x the power, but the performance will be the same or better. For prices, wait and see. The Apple Silicon machines so far is saying Apple is going to keep their same price tiers, give or take 10%. So, I don't see them having better performance per dollar on the high end. The performance/Watt is great for thin, integrated machines, but in desktops, performance/Watt and integration are not selling features.
The G4 Cube, 2013 Mac Pro and iMac Pro were limited by design, not by market movement, right? They were limitations stemming from a design choice, and choices they made over and over. The G4 cube was a dead end because the G4 CPU was a dead end and the G5 and successor dGPUs was too hot to use passively. That's the same dead end they had with the 2013 Mac Pro, right? I also think they chose not to redesign the interior of 2013 Mac Pro for 1 200W GPU and 1 150W Xeon, and moved forward with the iMac Pro instead. Then, even before they shipped the Mac Pro they knew it wasn't going to be the right product and told the market that the 2019 Mac Pro was coming. They finally arrived at the product that was the most sensible.
It's hard for me to understand how you can make a prediction that we will arrive at enough computing performance with 8K video editing. It supposes that Apple isn't interested in other types of computing which will require a lot of performance for which a workstation would serve, and that we know how video production will be done in the future. Designing a SFF and integration workstation around a perceived future would compound the mistake.
The race doesn't end. Year over year, Apple has to continually improve the performance of all its devices, and what required a rack 10 years ago is now available in a desktop on your desktop. They always have to design in performance to all practical limits for the Mac Pro.
The race to improve performance slows down over time in sync with lowered demand. I wouldn't say they have to design a computer to its physical limits but to the needs of the users of each product. Apple could design much larger notebooks that outclass heavy gaming laptops or thicker iMacs but they choose not to.
The race only stops for a particular type of software. Like maybe a word processor or presentation app, but new types of applications come in and require more performance. This is true even in specific types of software. What required increasing performance may have plateaued but some other aspect of that type of software ended up driving the need for more performance. Like maybe 8K is it for video production. Just dealing with the pixels. But with machine learning evaluating the content of every frame, and making some kind of change to every frame, it starts the race over again.
And yup, Apple has chosen not to address certain markets like "workstation" laptops that basically run 200 W worth of compute performance, or a consumer desktop box, or an MBA15, so on and so forth. We're only discussing where we think they are making mistakes or not. I think they have kept their lineup too simple and could address more of the market. The SFF Mac Pro is something they have tried before, twice depending on how you count the G4 cube, and I'm more skeptical then inviting this time around.
We will have the same conversation that we did when the 2013 Mac Pro came out, if this Apple Silicon Mac Pro is a SFF workstation. What's going to be different this time? You say Apple needs to make a product that sells. The G4 cube and 2013 Mac Pro didn't sell and were retired after 2 or 3 years. If the Apple Silicon Mac Pro are like them: small, limited capacity, not modular, what's going to be different this time?
The big difference will be performance per dollar. Some people make an assumption that the Cube was wrong, 2013 model was wrong, iMac Pro was wrong and the 2019 model was right based on size and upgrades but there's nothing to indicate the 2019 model was any less of a failure than the others as far as sales performance goes. The market at these price points is minuscule compared to the mainstream by at least a factor of 100.
What always affects users in higher-end models is performance per dollar. Apple's top-end models that can be $40k or more perform worse than Threadripper machines all because of Intel. Apple's own silicon can outperform even AMD at way more competitive price points.
If they stick at half the power level of the competition then in a few years time, the competition can catch or exceed them and for a fraction of a fraction of the users, they will make a choice to switch platform but that's assuming they need 2x the performance in a single workstation. If they only need 2x the performance and their use case can be networked, they can buy 2 or more.
Over the years, computers like the 2013 model, Cube, iMac Pro would run into limitations like handling 4k/8k editing. This isn't going to continue to be a problem, the video editing industry isn't going to move the mainstream beyond the limits of visual acuity for authoring. Once the performance demand in the video space dies down, that cuts out most of the demand for higher-end hardware and a factor of 2x difference won't be a deal-breaker for anyone, especially when Apple's custom media encoders outclass everything by a significant margin.
I think Apple loses on performance per dollar in desktop and high end hardware. They win on performance per Watt, but performance per dollar? No. They will maximize their prices so that performance/$ is about the same as x86 competitors, while x86 competitors are much more price elastic. As their usual practice, Apple will likely use their industrial design intangibles to keep prices high even when x86 has better performance/$.
If Apple uses half the power, they won't have a lead in performance, right? x86 can match the performance of M1 Mac Duo and Quad configurations with 16, 32, 64 core machines with dGPUs. They'll be using 2x the power, but the performance will be the same or better. For prices, wait and see. The Apple Silicon machines so far is saying Apple is going to keep their same price tiers, give or take 10%. So, I don't see them having better performance per dollar on the high end. The performance/Watt is great for thin, integrated machines, but in desktops, performance/Watt and integration are not selling features.
The G4 Cube, 2013 Mac Pro and iMac Pro were limited by design, not by market movement, right? They were limitations stemming from a design choice, and choices they made over and over. The G4 cube was a dead end because the G4 CPU was a dead end and the G5 and successor dGPUs was too hot to use passively. That's the same dead end they had with the 2013 Mac Pro, right? I also think they chose not to redesign the interior of 2013 Mac Pro for 1 200W GPU and 1 150W Xeon, and moved forward with the iMac Pro instead. Then, even before they shipped the Mac Pro they knew it wasn't going to be the right product and told the market that the 2019 Mac Pro was coming. They finally arrived at the product that was the most sensible.
It's hard for me to understand how you can make a prediction that we will arrive at enough computing performance with 8K video editing. It supposes that Apple isn't interested in other types of computing which will require a lot of performance for which a workstation would serve, and that we know how video production will be done in the future. Designing a SFF and integration workstation around a perceived future would compound the mistake.
Nah, 4-die CPUs will be around ~172 watts, then no single processor I know is 300 watts by default. That's the limit before it can't be cooled by conventional means.
You can go 600 watts overclocked, somewhat stable, with a chiller for cooling. Must be so productive.
with machine learning evaluating the content of every frame, and making some kind of change to every frame, it starts the race over again.
It's true that machine learning takes a lot of resources but these are offline processing examples that can be scaled in parallel, often in the cloud. Multi-frame machine learning processing at scale is crazy to do locally, even on a high-end PC. It takes days/weeks to finish on any single machine, 2x speedup doesn't make a difference. It needs 100x to be feasible locally and Apple will never be 100x lower by using a lower power level.
A significant portion of this is Intel/AMD pricing (over $17k of the price). Another huge portion is the RAM and SSD upgrades used and Afterburner. Apple charges $700 for M1 Max upgrade. Even adding $300 for the base chip, a quad version can be priced at $4k, which is equivalent to 2x Intel 28-core and just below quad Vega GPUs, which are priced at $17k.
This would be comparable to a Threadripper PC with a next-gen Nvidia 4090, which you would be able to get for around $6k late 2022. They won't beat cheap PCs simply due to margin but they can be very competitive on performance per dollar in the workstation space with their own hardware.
If Apple uses half the power, they won't have a lead in performance, right? x86 can match the performance of M1 Mac Duo and Quad configurations with 16, 32, 64 core machines with dGPUs. They'll be using 2x the power, but the performance will be the same or better.
I'd consider having the same performance at half the power to be leading. It results in less throttling, fan noise. They can also boost clock speeds over standard M1 Max.
The G4 Cube, 2013 Mac Pro and iMac Pro were limited by design, not by market movement, right? They were limitations stemming from a design choice, and choices they made over and over. The G4 cube was a dead end because the G4 CPU was a dead end and the G5 and successor dGPUs was too hot to use passively. That's the same dead end they had with the 2013 Mac Pro, right? I also think they chose not to redesign the interior of 2013 Mac Pro for 1 200W GPU and 1 150W Xeon, and moved forward with the iMac Pro instead. Then, even before they shipped the Mac Pro they knew it wasn't going to be the right product and told the market that the 2019 Mac Pro was coming. They finally arrived at the product that was the most sensible.
The limitations were partly by design but not because of Apple. The fault was always with their suppliers. The choice to use PPC was a flaw because the manufacturers couldn't keep improving, especially on mobile. All the heating problems Apple has ever had came from Intel/IBM/AMD/Nvidia. Apple has never had these problems with their own hardware.
If Intel and AMD had kept pace with TSMC and mobile chips, the Mac Pro would be the same 2013 model today but the suppliers didn't deliver. Intel's performance over the last 10 years has been really poor and probably on purpose to make more profit. You could debate whether not having a design to accommodate low performing suppliers was the fault or the suppliers themselves but regardless, that won't happen again because Apple is doing it themselves now.
Plus they can still accommodate some flexibility. Even back in 2013, I said they can easily put a single PCIe connector on and let people do what they want outside the box. If people want to attach 8x 40TFLOPs GPUs with a 4kW PSU for machine learning in a box costing $30k, that's fine. The workstation itself doesn't have to be designed to accommodate those extreme use cases internally.
with machine learning evaluating the content of every frame, and making some kind of change to every frame, it starts the race over again.
It's true that machine learning takes a lot of resources but these are offline processing examples that can be scaled in parallel, often in the cloud. Multi-frame machine learning processing at scale is crazy to do locally, even on a high-end PC. It takes days/weeks to finish on any single machine, 2x speedup doesn't make a difference. It needs 100x to be feasible locally and Apple will never be 100x lower by using a lower power level.
A significant portion of this is Intel/AMD pricing (over $17k of the price). Another huge portion is the RAM and SSD upgrades used and Afterburner. Apple charges $700 for M1 Max upgrade. Even adding $300 for the base chip, a quad version can be priced at $4k, which is equivalent to 2x Intel 28-core and just below quad Vega GPUs, which are priced at $17k.
This would be comparable to a Threadripper PC with a next-gen Nvidia 4090, which you would be able to get for around $6k late 2022. They won't beat cheap PCs simply due to margin but they can be very competitive on performance per dollar in the workstation space with their own hardware.
If Apple uses half the power, they won't have a lead in performance, right? x86 can match the performance of M1 Mac Duo and Quad configurations with 16, 32, 64 core machines with dGPUs. They'll be using 2x the power, but the performance will be the same or better.
I'd consider having the same performance at half the power to be leading. It results in less throttling, fan noise. They can also boost clock speeds over standard M1 Max.
The G4 Cube, 2013 Mac Pro and iMac Pro were limited by design, not by market movement, right? They were limitations stemming from a design choice, and choices they made over and over. The G4 cube was a dead end because the G4 CPU was a dead end and the G5 and successor dGPUs was too hot to use passively. That's the same dead end they had with the 2013 Mac Pro, right? I also think they chose not to redesign the interior of 2013 Mac Pro for 1 200W GPU and 1 150W Xeon, and moved forward with the iMac Pro instead. Then, even before they shipped the Mac Pro they knew it wasn't going to be the right product and told the market that the 2019 Mac Pro was coming. They finally arrived at the product that was the most sensible.
The limitations were partly by design but not because of Apple. The fault was always with their suppliers. The choice to use PPC was a flaw because the manufacturers couldn't keep improving, especially on mobile. All the heating problems Apple has ever had came from Intel/IBM/AMD/Nvidia. Apple has never had these problems with their own hardware.
If Intel and AMD had kept pace with TSMC and mobile chips, the Mac Pro would be the same 2013 model today but the suppliers didn't deliver. Intel's performance over the last 10 years has been really poor and probably on purpose to make more profit. You could debate whether not having a design to accommodate low performing suppliers was the fault or the suppliers themselves but regardless, that won't happen again because Apple is doing it themselves now.
Plus they can still accommodate some flexibility. Even back in 2013, I said they can easily put a single PCIe connector on and let people do what they want outside the box. If people want to attach 8x 40TFLOPs GPUs with a 4kW PSU for machine learning in a box costing $30k, that's fine. The workstation itself doesn't have to be designed to accommodate those extreme use cases internally.
That's true, consumers always blamed OEM for Intel's failure. That's also why the touch bar performs badly, I assume the design team is overly confident about Intel.
Anyway, this generation of MacBook Pros is future-proof enough, there are fierce competitions going on. An 8+2 M1 Max, despite most believed today, is not enough in both performance & the competition. Intel will definitely be able to keep the pace, and sometimes even kills it. I'm sure Apple has a plan to go even bigger.
A significant portion of this is Intel/AMD pricing (over $17k of the price). Another huge portion is the RAM and SSD upgrades used and Afterburner. Apple charges $700 for M1 Max upgrade. Even adding $300 for the base chip, a quad version can be priced at $4k, which is equivalent to 2x Intel 28-core and just below quad Vega GPUs, which are priced at $17k.
This would be comparable to a Threadripper PC with a next-gen Nvidia 4090, which you would be able to get for around $6k late 2022. They won't beat cheap PCs simply due to margin but they can be very competitive on performance per dollar in the workstation space with their own hardware.
If Apple charges $8k for a M1 Max Quad (as you are thinking), than perhaps they are thinking about making a dent in the HPC market. That's if. My thoughts change if so.
Apple knows that they have competitors in the market. If HP or Dell is offering a 38 core Xeon (Xeon-3375 with a 40 TFLOP GPU - Nvidia 3090 or A8000 whatever) for 20k. Will Apple price a M1 Max Quad, about equivalent to the stated Intel machine, for 10k? I'm really skeptical at 8K. 10K? Not giving me the warm fuzzies either. 15k? That sounds like something Apple would do: having a machine be competitive to others, and try to triangulate a price structure that makes their business profitable while not pricing themselves out of the market. Being way under competitor products? That doesn't sound something like they would do. They will take the money.
So, if x86 competitors are providing a price umbrella for Apple to be under, I think Apple is going to price themselves as close to the umbrella as possible. So 15k to 20k, if competitors are at 20k. Moreover, x86 OEMs are much more price elastic than Apple is. If Apple offers a M1 Max Quad config at $8k, and it is a threat to their x86 sales, they will lower their prices, and lower them to the point of having better performance/$.
If Apple uses half the power, they won't have a lead in performance, right? x86 can match the performance of M1 Mac Duo and Quad configurations with 16, 32, 64 core machines with dGPUs. They'll be using 2x the power, but the performance will be the same or better.
I'd consider having the same performance at half the power to be leading. It results in less throttling, fan noise. They can also boost clock speeds over standard M1 Max.
But is that lead (performance/W) something that will make the machine successful? Is it a feature that will sell the machine if there is performance parity? Keep in mind that they don't have to have a SFF machine to have this performance/Watt lead. They could have it have the internal expansion capabilities of the 2019 Mac Pro and be as power efficient as a SFF model.
If there is performance/price parity, buyers are largely going to make decisions on software compatibility. This isn't a bad situation per se, it allows Apple to defend its content creation share, but gain more share above that? Probably not. Then, being a SFF obviously present issues given the history. So, wait and see what Apple is going to do here.
The G4 Cube, 2013 Mac Pro and iMac Pro were limited by design, not by market movement, right? They were limitations stemming from a design choice, and choices they made over and over. The G4 cube was a dead end because the G4 CPU was a dead end and the G5 and successor dGPUs was too hot to use passively. That's the same dead end they had with the 2013 Mac Pro, right? I also think they chose not to redesign the interior of 2013 Mac Pro for 1 200W GPU and 1 150W Xeon, and moved forward with the iMac Pro instead. Then, even before they shipped the Mac Pro they knew it wasn't going to be the right product and told the market that the 2019 Mac Pro was coming. They finally arrived at the product that was the most sensible.
The limitations were partly by design but not because of Apple. The fault was always with their suppliers. The choice to use PPC was a flaw because the manufacturers couldn't keep improving, especially on mobile. All the heating problems Apple has ever had came from Intel/IBM/AMD/Nvidia. Apple has never had these problems with their own hardware.
If Intel and AMD had kept pace with TSMC and mobile chips, the Mac Pro would be the same 2013 model today but the suppliers didn't deliver. Intel's performance over the last 10 years has been really poor and probably on purpose to make more profit. You could debate whether not having a design to accommodate low performing suppliers was the fault or the suppliers themselves but regardless, that won't happen again because Apple is doing it themselves now.
Plus they can still accommodate some flexibility. Even back in 2013, I said they can easily put a single PCIe connector on and let people do what they want outside the box. If people want to attach 8x 40TFLOPs GPUs with a 4kW PSU for machine learning in a box costing $30k, that's fine. The workstation itself doesn't have to be designed to accommodate those extreme use cases internally.
There isn't any fault nor assignment of blame. Apple was in equal footing to everyone else, with access to all the same components. Apple is responsible for predicting what type of hardware is available and what direction hardware and software will have in the future. They are also responsible for adjusting when they are wrong or something they expect doesn't happen. I'd expect that they should adjust much faster and to learn lessons from mistakes. This is why it is concerning that the rumors of the Apple Silicon Mac Pro is a SFF machine about the size of the G4 cube, and why I keep asking the "what's different this time question".
The Intel transition was their response for PPC fizzling out. Good call. But even there, they didn't bring the cube back with Intel hardware inside. It would have had more sales as the G4 version as the Mac Pro went up in price with Intel Xeons, and a lot of folks would have bought an Intel cube at $2000. But they offered an iMac G5 and iMac Intel models at those price tiers instead. Basically they think the iMac 27 is their mid-tier offering to folks who want what a typical PC desktop could provide in terms of performance, and that's carried through to this day. Nice offering, but it reduced the potential market of buyers. It really doesn't hurt to offer different form factors at the same price points.
For the 2013 Mac Pro, they could have redesigned it internally to have 1 300W GPU instead of 2 150W GPUs. Having it updated with new Xeons and new dGPUs would have kept sales up, and would have made the iMac Pro unnecessary. They could have updated it to TB3, and made accessories for it. But they had to have given up on it by 2015 or so when iMac Pro development started. So, they felt the SFF cylindrical form factor was a bust by 2015, and moved to the iMac Pro as their top end Mac. Obviously, they knew that wasn't enough and went full circle to the 2019 Mac Pro model.
A SFF Apple Silicon Mac Pro could get a lot of unit sales if it was cheap enough, but I'm pretty skeptical it will be cheap. I think it will be the same price tiers as today. It does lose sales to folks who want internal expansion, but perhaps it gains sales from potential iMac 27 buyers getting it instead, if it is cheap enough, like $2500 starting price, but no way, right? Apple still needs to get software transitioned to Metal. They still need to update it on preferably a yearly cadence to show that they are committed to it. And, they need to make accessories for it, including a monitor. It's a long road ahead.
A significant portion of this is Intel/AMD pricing (over $17k of the price). Another huge portion is the RAM and SSD upgrades used and Afterburner. Apple charges $700 for M1 Max upgrade. Even adding $300 for the base chip, a quad version can be priced at $4k, which is equivalent to 2x Intel 28-core and just below quad Vega GPUs, which are priced at $17k.
This would be comparable to a Threadripper PC with a next-gen Nvidia 4090, which you would be able to get for around $6k late 2022. They won't beat cheap PCs simply due to margin but they can be very competitive on performance per dollar in the workstation space with their own hardware.
If Apple charges $8k for a M1 Max Quad (as you are thinking), than perhaps they are thinking about making a dent in the HPC market. That's if. My thoughts change if so.
Apple knows that they have competitors in the market. If HP or Dell is offering a 38 core Xeon (Xeon-3375 with a 40 TFLOP GPU - Nvidia 3090 or A8000 whatever) for 20k. Will Apple price a M1 Max Quad, about equivalent to the stated Intel machine, for 10k? I'm really skeptical at 8K. 10K? Not giving me the warm fuzzies either. 15k? That sounds like something Apple would do: having a machine be competitive to others, and try to triangulate a price structure that makes their business profitable while not pricing themselves out of the market. Being way under competitor products? That doesn't sound something like they would do. They will take the money.
So, if x86 competitors are providing a price umbrella for Apple to be under, I think Apple is going to price themselves as close to the umbrella as possible. So 15k to 20k, if competitors are at 20k. Moreover, x86 OEMs are much more price elastic than Apple is. If Apple offers a M1 Max Quad config at $8k, and it is a threat to their x86 sales, they will lower their prices, and lower them to the point of having better performance/$.
If Apple uses half the power, they won't have a lead in performance, right? x86 can match the performance of M1 Mac Duo and Quad configurations with 16, 32, 64 core machines with dGPUs. They'll be using 2x the power, but the performance will be the same or better.
I'd consider having the same performance at half the power to be leading. It results in less throttling, fan noise. They can also boost clock speeds over standard M1 Max.
But is that lead (performance/W) something that will make the machine successful? Is it a feature that will sell the machine if there is performance parity? Keep in mind that they don't have to have a SFF machine to have this performance/Watt lead. They could have it have the internal expansion capabilities of the 2019 Mac Pro and be as power efficient as a SFF model.
If there is performance/price parity, buyers are largely going to make decisions on software compatibility. This isn't a bad situation per se, it allows Apple to defend its content creation share, but gain more share above that? Probably not. Then, being a SFF obviously present issues given the history. So, wait and see what Apple is going to do here.
The G4 Cube, 2013 Mac Pro and iMac Pro were limited by design, not by market movement, right? They were limitations stemming from a design choice, and choices they made over and over. The G4 cube was a dead end because the G4 CPU was a dead end and the G5 and successor dGPUs was too hot to use passively. That's the same dead end they had with the 2013 Mac Pro, right? I also think they chose not to redesign the interior of 2013 Mac Pro for 1 200W GPU and 1 150W Xeon, and moved forward with the iMac Pro instead. Then, even before they shipped the Mac Pro they knew it wasn't going to be the right product and told the market that the 2019 Mac Pro was coming. They finally arrived at the product that was the most sensible.
The limitations were partly by design but not because of Apple. The fault was always with their suppliers. The choice to use PPC was a flaw because the manufacturers couldn't keep improving, especially on mobile. All the heating problems Apple has ever had came from Intel/IBM/AMD/Nvidia. Apple has never had these problems with their own hardware.
If Intel and AMD had kept pace with TSMC and mobile chips, the Mac Pro would be the same 2013 model today but the suppliers didn't deliver. Intel's performance over the last 10 years has been really poor and probably on purpose to make more profit. You could debate whether not having a design to accommodate low performing suppliers was the fault or the suppliers themselves but regardless, that won't happen again because Apple is doing it themselves now.
Plus they can still accommodate some flexibility. Even back in 2013, I said they can easily put a single PCIe connector on and let people do what they want outside the box. If people want to attach 8x 40TFLOPs GPUs with a 4kW PSU for machine learning in a box costing $30k, that's fine. The workstation itself doesn't have to be designed to accommodate those extreme use cases internally.
There isn't any fault nor assignment of blame. Apple was in equal footing to everyone else, with access to all the same components. Apple is responsible for predicting what type of hardware is available and what direction hardware and software will have in the future. They are also responsible for adjusting when they are wrong or something they expect doesn't happen. I'd expect that they should adjust much faster and to learn lessons from mistakes. This is why it is concerning that the rumors of the Apple Silicon Mac Pro is a SFF machine about the size of the G4 cube, and why I keep asking the "what's different this time question".
The Intel transition was their response for PPC fizzling out. Good call. But even there, they didn't bring the cube back with Intel hardware inside. It would have had more sales as the G4 version as the Mac Pro went up in price with Intel Xeons, and a lot of folks would have bought an Intel cube at $2000. But they offered an iMac G5 and iMac Intel models at those price tiers instead. Basically they think the iMac 27 is their mid-tier offering to folks who want what a typical PC desktop could provide in terms of performance, and that's carried through to this day. Nice offering, but it reduced the potential market of buyers. It really doesn't hurt to offer different form factors at the same price points.
For the 2013 Mac Pro, they could have redesigned it internally to have 1 300W GPU instead of 2 150W GPUs. Having it updated with new Xeons and new dGPUs would have kept sales up, and would have made the iMac Pro unnecessary. They could have updated it to TB3, and made accessories for it. But they had to have given up on it by 2015 or so when iMac Pro development started. So, they felt the SFF cylindrical form factor was a bust by 2015, and moved to the iMac Pro as their top end Mac. Obviously, they knew that wasn't enough and went full circle to the 2019 Mac Pro model.
A SFF Apple Silicon Mac Pro could get a lot of unit sales if it was cheap enough, but I'm pretty skeptical it will be cheap. I think it will be the same price tiers as today. It does lose sales to folks who want internal expansion, but perhaps it gains sales from potential iMac 27 buyers getting it instead, if it is cheap enough, like $2500 starting price, but no way, right? Apple still needs to get software transitioned to Metal. They still need to update it on preferably a yearly cadence to show that they are committed to it. And, they need to make accessories for it, including a monitor. It's a long road ahead.
Sigh, you're still missing the point:
1). Even Intel & AMD have their limits, they can't just make chips infinitely faster, either it'll be impossible to cool or the performance gain makes no sense (diminishing return says you'll be giving 2x more power for less than 20% in return, anyone wants that?). Apple Silicon will match their sweet spot with much less consumption.
2). Nobody says the new Mac Pro will be tiny, there'll be slots, we just don't know how much.
Moreover, x86 OEMs are much more price elastic than Apple is. If Apple offers a M1 Max Quad config at $8k, and it is a threat to their x86 sales, they will lower their prices, and lower them to the point of having better performance/$.
I think they would find that hard to do in a lot of cases. The high-end Xeons are monolithic designs. If they cut prices, they will likely lose money on them. AMD/Intel are constrained by the server market. Using multiple dies is much more cost-effective but it's more to do with the fact that Apple's doing both the CPU and GPU on the same chip. Nobody is making 40TFLOPs integrated graphics, that would be near impossible for 3rd party manufacturers to compete with because GPUs are not used much in the server space. That leaves manufacturers making both high-end CPUs and GPUs separately, which doubles the price.
Normally when Apple maintains 40% margins on every component, that pushes their prices to around double equivalent PCs. Apple Silicon will allow them to maintain 40% margins at less than half the price as they aren't paying a markup to Intel/AMD.
It removes the most expensive profit margins and separate component costs from 3rd parties using components that are dirt cheap because they are used in mobile devices.
But is that lead (performance/W) something that will make the machine successful? Is it a feature that will sell the machine if there is performance parity?
In clusters it's more important than a single workstation. Crypto miners go for the best performance-per-watt over raw power. For a single workstation, it's still worthwhile to have a stable machine that won't be overheating or noisy running for long periods of time. If it's the same performance, there's no benefit to the noisier one.
they felt the SFF cylindrical form factor was a bust by 2015, and moved to the iMac Pro as their top end Mac. Obviously, they knew that wasn't enough and went full circle to the 2019 Mac Pro model.
I think part of this was trying different strategies to boost sales and nothing has been working. People are assuming the 2019 tower is working only because it's the latest iteration. Once they go back to a smaller form factor, people will see that the tower didn't boost sales either. It doesn't matter what they make above $5k, this market is tiny.
As I say though, a single PCIe slot is enough to satisfy everyone while having a small form factor.
One route they could go is to make it flat like behind an XDR display and the internals could be lifted out into a server blade form factor. With a PCIe slot, this could be connected to another blade that has IO cards or GPUs in it. It still won't sell many units but it works for multiple use cases. Musicians and artists who want a compact product would have a single display. Server users get a compact blade. People who want expansion can either buy the blade for the headless form factor and plug in the extra box or buy the iMac and plug in the box. If someone refurbs any of them, they can be resold for the other use cases.
Moreover, x86 OEMs are much more price elastic than Apple is. If Apple offers a M1 Max Quad config at $8k, and it is a threat to their x86 sales, they will lower their prices, and lower them to the point of having better performance/$.
I think they would find that hard to do in a lot of cases. The high-end Xeons are monolithic designs. If they cut prices, they will likely lose money on them. AMD/Intel are constrained by the server market. Using multiple dies is much more cost-effective but it's more to do with the fact that Apple's doing both the CPU and GPU on the same chip. Nobody is making 40TFLOPs integrated graphics, that would be near impossible for 3rd party manufacturers to compete with because GPUs are not used much in the server space. That leaves manufacturers making both high-end CPUs and GPUs separately, which doubles the price.
Normally when Apple maintains 40% margins on every component, that pushes their prices to around double equivalent PCs. Apple Silicon will allow them to maintain 40% margins at less than half the price as they aren't paying a markup to Intel/AMD.
It removes the most expensive profit margins and separate component costs from 3rd parties using components that are dirt cheap because they are used in mobile devices.
But is that lead (performance/W) something that will make the machine successful? Is it a feature that will sell the machine if there is performance parity?
In clusters it's more important than a single workstation. Crypto miners go for the best performance-per-watt over raw power. For a single workstation, it's still worthwhile to have a stable machine that won't be overheating or noisy running for long periods of time. If it's the same performance, there's no benefit to the noisier one.
they felt the SFF cylindrical form factor was a bust by 2015, and moved to the iMac Pro as their top end Mac. Obviously, they knew that wasn't enough and went full circle to the 2019 Mac Pro model.
I think part of this was trying different strategies to boost sales and nothing has been working. People are assuming the 2019 tower is working only because it's the latest iteration. Once they go back to a smaller form factor, people will see that the tower didn't boost sales either. It doesn't matter what they make above $5k, this market is tiny.
As I say though, a single PCIe slot is enough to satisfy everyone while having a small form factor.
One route they could go is to make it flat like behind an XDR display and the internals could be lifted out into a server blade form factor. With a PCIe slot, this could be connected to another blade that has IO cards or GPUs in it. It still won't sell many units but it works for multiple use cases. Musicians and artists who want a compact product would have a single display. Server users get a compact blade. People who want expansion can either buy the blade for the headless form factor and plug in the extra box or buy the iMac and plug in the box. If someone refurbs any of them, they can be resold for the other use cases.
That sounds like an Apple Silicon Xserve, I guess olds are new again
Moreover, x86 OEMs are much more price elastic than Apple is. If Apple offers a M1 Max Quad config at $8k, and it is a threat to their x86 sales, they will lower their prices, and lower them to the point of having better performance/$.
I think they would find that hard to do in a lot of cases. The high-end Xeons are monolithic designs. If they cut prices, they will likely lose money on them. AMD/Intel are constrained by the server market. Using multiple dies is much more cost-effective but it's more to do with the fact that Apple's doing both the CPU and GPU on the same chip. Nobody is making 40TFLOPs integrated graphics, that would be near impossible for 3rd party manufacturers to compete with because GPUs are not used much in the server space. That leaves manufacturers making both high-end CPUs and GPUs separately, which doubles the price.
Normally when Apple maintains 40% margins on every component, that pushes their prices to around double equivalent PCs. Apple Silicon will allow them to maintain 40% margins at less than half the price as they aren't paying a markup to Intel/AMD.
It removes the most expensive profit margins and separate component costs from 3rd parties using components that are dirt cheap because they are used in mobile devices.
High end Xeons are moving to MCM designs with Sapphire Rapids, probably late 2022. 4 dies of about 400 mm each per package, and up to 4/8 sockets per board (the usual Xeon-SP systems architecture). Somewhere between 12 to 60 cores per socket depending on Intel's segmentation. There will be single socket Xeon W and probably Core X variants for workstations. 200 to 400 W TDP per socket. Intel will be afford to go down in price if they are pressured by AMD. Obviously, AMD uses MCM packages already for their servers and workstation CPUs. Neither of them are in hurry to try to undercut the other for server and workstation CPUs, but both have the means to do it. Intel largely uses mature, or trailing, fabs for server CPUs and have margin to spare on them. AMD is seemingly comfortable with lower margins.
Yes, it is the theory that Apple can have reduced retail prices for their Apple Silicon machines because they don't need to pay for Intel, AMD, and graphics memory margins with Apple Silicon. That's the theory. So, why are MBP14/16 prices higher? I think NAND is cheaper now, so there are cheaper prices from cheaper NAND GB/$ costs, but Apple is basically using the same price points for all their Apple Silicon Macs. The M1 Mac Mini is cheaper, but the M1 Pro MBP14/16 base SKUs are about $200 more than prior base SKUs. Where's this price advantage?
An M1 MBA at $800 would sell a lot! An iMac 24 at $1000? Awesome price point. Or, why isn't the base MBP14/16 SKUs at $1800 and $2300?Even you estimate an iMac Pro with M1 Max Duo to cost $5000. Why not $4000? So, wait on see the prices. Like I said, so far, they seem to be sticking with their established price tiers give or take 10%.
GPUs are going into the HPC server space. All codes that solve systems of differential equations (numerical codes in physics/engineering fields) are transitioning to GPU compute. It's crazy not to as it can be a 10x performance lift. Data science and crypto are using GPU compute, at least until the specialist tensor or machine learning chips come. Apple has an advantage here with Apple Silicon as current GPUs don't have a lot of memory, which a lot of numerical codes really like. The window is closing though. One Apple Silicon 128 g-core GPU isn't enough. It's great that it will only need about 250 Watts. Well, it's even better if Apple can put 4 of them in a box.
Nvidia is dominating with their server GPUs. Intel is following with Ponte Vecchio and AMD is also following. It's a lot like the CPU server boards, 4 to 8 GPU sockets in a 3U or 4U rack unit. I can definitely understand that Apple doesn't want to be in the server space, but they really shouldn't short shrift themselves and the potential uses of their hardware by not designing their top end hardware to all practical limits.
But is that lead (performance/W) something that will make the machine successful? Is it a feature that will sell the machine if there is performance parity?
In clusters it's more important than a single workstation. Crypto miners go for the best performance-per-watt over raw power. For a single workstation, it's still worthwhile to have a stable machine that won't be overheating or noisy running for long periods of time. If it's the same performance, there's no benefit to the noisier one.
The workstations won't be overheating because their cooling systems are designed to remove 1000 to 1500 W of heat out of the box, just like the 2019 Mac Pro is. They'll likely be loud, will make the room they are in hotter, but they will work fine and generate their advertised performance.
The ostensible advantage of x86 workstations is being able to run 2 to 4 GPUs in the box, have 20 to 50 TB of storage in the box, support >1 TB of RAM, and software compatibility. Ie, flexibility. It's also cleaner as all the stuff is in the box. What are they gaining with a Mac Pro in a SFF box at the same performance? It uses less power, is quieter. If the user has external storage, PCIe breakout boxes, it becomes a sprawl of boxes and cabling. Not nice. You and I will have to agree to disagree that performance/W is a selling feature in workstations. I think people will choose a big box over a small one because the internal expansion is more valuable in this market than being small, quiet and efficient.
There were long threads on AI over the Mac Pro SFF design in 2013. All the criticisms were basically right. I'm not looking forward to a repeat of those threads if the Apple Silicon Mac Pro is a SFF machine.
they felt the SFF cylindrical form factor was a bust by 2015, and moved to the iMac Pro as their top end Mac. Obviously, they knew that wasn't enough and went full circle to the 2019 Mac Pro model.
I think part of this was trying different strategies to boost sales and nothing has been working. People are assuming the 2019 tower is working only because it's the latest iteration. Once they go back to a smaller form factor, people will see that the tower didn't boost sales either. It doesn't matter what they make above $5k, this market is tiny.
As I say though, a single PCIe slot is enough to satisfy everyone while having a small form factor.
One route they could go is to make it flat like behind an XDR display and the internals could be lifted out into a server blade form factor. With a PCIe slot, this could be connected to another blade that has IO cards or GPUs in it. It still won't sell many units but it works for multiple use cases. Musicians and artists who want a compact product would have a single display. Server users get a compact blade. People who want expansion can either buy the blade for the headless form factor and plug in the extra box or buy the iMac and plug in the box. If someone refurbs any of them, they can be resold for the other use cases.
Yes, the unit sales is tiny compared to consumer machines, but Apple themselves see it as needed. Otherwise, they would not have made the 2019 Mac Pro. Ignoring some market segments can have negative effects on sales of other hardware. No Mac workstations may mean less Macbook Pro sales. There's a negative feedback effect that can result in less software support, which further erodes sales, further decreasing Mac sales. They can't ignore market niches without anticipating the effect of sales of their other products. So at least for the sake of their content creation market share, a workstation Mac has to be available.
Do you think this "Pro Workflow Team" that Ternus trots out in statements and interviews is a kind of performative art? This is in response to your "trying different strategies" comment. For some reason, they think it is comforting for potential buyers to hear it. I think it is downright frightening as it implies, if it is a forthright statement, that they didn't know how this market segment uses computers. It could very well be true, they didn't know how the market segment uses computers, and they thought the 2013 Mac Pro and 2017 iMac Pro were going to sell. It's too hard to believe that a computer company doesn't know how computers are used, in most market segments at least. They must of have had a vision of how this market uses computers, felt that they could get users there, and felt that time would eventually make it true. I keep saying a book has to be written about what happen with Mac hardware from 2012 to 2018 or so, to reveal WTF happened.
They already have a workstation product that meets the needs of most of its market: the 2019 Mac Pro. They sell rack versions of it, and it can be configured in all sorts of ways. Like I said, people in this market aren't looking for a small box. For people who are looking for a SFF box, Apple can just update the Mac mini with a M1 Max. With Ethernet over Thunderbolt, a stack of them can by networked together at something like 30 Gbits/s for people who want a compile farm or run some small scale jobs that can be run over a network. And, OWC miniStacks can used to add storage.
I do think that if they updated the 2019 Mac Pro with Apple Silicon, they can put a M1 Max Quad in an 2-wide MPX Module and put 4 of them in a Mac Pro. I also think that the Mac mini shouldn't have a square footprint. It's just not a great shape for cooling. If it was say 12 x 6 x 1.5 inches, higher performance blower fans or axial fans could be used, and they'd probably be capable of putting anything from an M1 to M1 Max Duo in it.
Moreover, x86 OEMs are much more price elastic than Apple is. If Apple offers a M1 Max Quad config at $8k, and it is a threat to their x86 sales, they will lower their prices, and lower them to the point of having better performance/$.
I think they would find that hard to do in a lot of cases. The high-end Xeons are monolithic designs. If they cut prices, they will likely lose money on them. AMD/Intel are constrained by the server market. Using multiple dies is much more cost-effective but it's more to do with the fact that Apple's doing both the CPU and GPU on the same chip. Nobody is making 40TFLOPs integrated graphics, that would be near impossible for 3rd party manufacturers to compete with because GPUs are not used much in the server space. That leaves manufacturers making both high-end CPUs and GPUs separately, which doubles the price.
Normally when Apple maintains 40% margins on every component, that pushes their prices to around double equivalent PCs. Apple Silicon will allow them to maintain 40% margins at less than half the price as they aren't paying a markup to Intel/AMD.
It removes the most expensive profit margins and separate component costs from 3rd parties using components that are dirt cheap because they are used in mobile devices.
High end Xeons are moving to MCM designs with Sapphire Rapids, probably late 2022. 4 dies of about 400 mm each per package, and up to 4/8 sockets per board (the usual Xeon-SP systems architecture). Somewhere between 12 to 60 cores per socket depending on Intel's segmentation. There will be single socket Xeon W and probably Core X variants for workstations. 200 to 400 W TDP per socket. Intel will be afford to go down in price if they are pressured by AMD. Obviously, AMD uses MCM packages already for their servers and workstation CPUs. Neither of them are in hurry to try to undercut the other for server and workstation CPUs, but both have the means to do it. Intel largely uses mature, or trailing, fabs for server CPUs and have margin to spare on them. AMD is seemingly comfortable with lower margins.
Yes, it is the theory that Apple can have reduced retail prices for their Apple Silicon machines because they don't need to pay for Intel, AMD, and graphics memory margins with Apple Silicon. That's the theory. So, why are MBP14/16 prices higher? I think NAND is cheaper now, so there are cheaper prices from cheaper NAND GB/$ costs, but Apple is basically using the same price points for all their Apple Silicon Macs. The M1 Mac Mini is cheaper, but the M1 Pro MBP14/16 base SKUs are about $200 more than prior base SKUs. Where's this price advantage?
An M1 MBA at $800 would sell a lot! An iMac 24 at $1000? Awesome price point. Or, why isn't the base MBP14/16 SKUs at $1800 and $2300?Even you estimate an iMac Pro with M1 Max Duo to cost $5000. Why not $4000? So, wait on see the prices. Like I said, so far, they seem to be sticking with their established price tiers give or take 10%.
GPUs are going into the HPC server space. All codes that solve systems of differential equations (numerical codes in physics/engineering fields) are transitioning to GPU compute. It's crazy not to as it can be a 10x performance lift. Data science and crypto are using GPU compute, at least until the specialist tensor or machine learning chips come. Apple has an advantage here with Apple Silicon as current GPUs don't have a lot of memory, which a lot of numerical codes really like. The window is closing though. One Apple Silicon 128 g-core GPU isn't enough. It's great that it will only need about 250 Watts. Well, it's even better if Apple can put 4 of them in a box.
Nvidia is dominating with their server GPUs. Intel is following with Ponte Vecchio and AMD is also following. It's a lot like the CPU server boards, 4 to 8 GPU sockets in a 3U or 4U rack unit. I can definitely understand that Apple doesn't want to be in the server space, but they really shouldn't short shrift themselves and the potential uses of their hardware by not designing their top end hardware to all practical limits.
But is that lead (performance/W) something that will make the machine successful? Is it a feature that will sell the machine if there is performance parity?
In clusters it's more important than a single workstation. Crypto miners go for the best performance-per-watt over raw power. For a single workstation, it's still worthwhile to have a stable machine that won't be overheating or noisy running for long periods of time. If it's the same performance, there's no benefit to the noisier one.
The workstations won't be overheating because their cooling systems are designed to remove 1000 to 1500 W of heat out of the box, just like the 2019 Mac Pro is. They'll likely be loud, will make the room they are in hotter, but they will work fine and generate their advertised performance.
The ostensible advantage of x86 workstations is being able to run 2 to 4 GPUs in the box, have 20 to 50 TB of storage in the box, support >1 TB of RAM, and software compatibility. Ie, flexibility. It's also cleaner as all the stuff is in the box. What are they gaining with a Mac Pro in a SFF box at the same performance? It uses less power, is quieter. If the user has external storage, PCIe breakout boxes, it becomes a sprawl of boxes and cabling. Not nice. You and I will have to agree to disagree that performance/W is a selling feature in workstations. I think people will choose a big box over a small one because the internal expansion is more valuable in this market than being small, quiet and efficient.
There were long threads on AI over the Mac Pro SFF design in 2013. All the criticisms were basically right. I'm not looking forward to a repeat of those threads if the Apple Silicon Mac Pro is a SFF machine.
they felt the SFF cylindrical form factor was a bust by 2015, and moved to the iMac Pro as their top end Mac. Obviously, they knew that wasn't enough and went full circle to the 2019 Mac Pro model.
I think part of this was trying different strategies to boost sales and nothing has been working. People are assuming the 2019 tower is working only because it's the latest iteration. Once they go back to a smaller form factor, people will see that the tower didn't boost sales either. It doesn't matter what they make above $5k, this market is tiny.
As I say though, a single PCIe slot is enough to satisfy everyone while having a small form factor.
One route they could go is to make it flat like behind an XDR display and the internals could be lifted out into a server blade form factor. With a PCIe slot, this could be connected to another blade that has IO cards or GPUs in it. It still won't sell many units but it works for multiple use cases. Musicians and artists who want a compact product would have a single display. Server users get a compact blade. People who want expansion can either buy the blade for the headless form factor and plug in the extra box or buy the iMac and plug in the box. If someone refurbs any of them, they can be resold for the other use cases.
Yes, the unit sales is tiny compared to consumer machines, but Apple themselves see it as needed. Otherwise, they would not have made the 2019 Mac Pro. Ignoring some market segments can have negative effects on sales of other hardware. No Mac workstations may mean less Macbook Pro sales. There's a negative feedback effect that can result in less software support, which further erodes sales, further decreasing Mac sales. They can't ignore market niches without anticipating the effect of sales of their other products. So at least for the sake of their content creation market share, a workstation Mac has to be available.
Do you think this "Pro Workflow Team" that Ternus trots out in statements and interviews is a kind of performative art? This is in response to your "trying different strategies" comment. For some reason, they think it is comforting for potential buyers to hear it. I think it is downright frightening as it implies, if it is a forthright statement, that they didn't know how this market segment uses computers. It could very well be true, they didn't know how the market segment uses computers, and they thought the 2013 Mac Pro and 2017 iMac Pro were going to sell. It's too hard to believe that a computer company doesn't know how computers are used, in most market segments at least. They must of have had a vision of how this market uses computers, felt that they could get users there, and felt that time would eventually make it true. I keep saying a book has to be written about what happen with Mac hardware from 2012 to 2018 or so, to reveal WTF happened.
They already have a workstation product that meets the needs of most of its market: the 2019 Mac Pro. They sell rack versions of it, and it can be configured in all sorts of ways. Like I said, people in this market aren't looking for a small box. For people who are looking for a SFF box, Apple can just update the Mac mini with a M1 Max. With Ethernet over Thunderbolt, a stack of them can by networked together at something like 30 Gbits/s for people who want a compile farm or run some small scale jobs that can be run over a network. And, OWC miniStacks can used to add storage.
I do think that if they updated the 2019 Mac Pro with Apple Silicon, they can put a M1 Max Quad in an 2-wide MPX Module and put 4 of them in a Mac Pro. I also think that the Mac mini shouldn't have a square footprint. It's just not a great shape for cooling. If it was say 12 x 6 x 1.5 inches, higher performance blower fans or axial fans could be used, and they'd probably be capable of putting anything from an M1 to M1 Max Duo in it.
"The workstations won't be overheating because their cooling systems are designed to remove 1000 to 1500 W of heat out of the box, just like the 2019 Mac Pro is."
Again, no they don't. The PSU is capable of putting 1.5KW of power, though that doesn’t mean you’ll use all of that (it needs overhead, come on) or all the power will transfer to heat. The most powerful CPU out there tops at ~300W, don't make it sound like cooling a 1000w chip is easy, if not downright impossible. With that in mind, not every workstation needs 4GPUs, depending on what you're doing.
"I think people will choose a big box over a small one because the internal expansion is more valuable in this market than being small, quiet, and efficient."
Okay, I agree. I don't know what Marvin means by "SFF" but it's no trash can. I'd say at least four double-wide slots. On the other hand, you're not using all eight slots in the current design, at least five are occupied (I/O card, MPX modules).
"If it was say 12 x 6 x 1.5 inches, higher performance blower fans or axial fans could be used, and they'd probably be capable of putting anything from an M1 to M1 Max Duo in it."
Or not? The 16" MacBook Pro had a cooling system of at least 120W. I think it can do more, but the point is thickness aren't that restrictive. Apple can just spin up the fan, then mini will be able to 2-die like magic. But why do that on an entry-level system? It won't be much cheaper, might be noisier, and potentially a lot hotter, too. Then, why not just get an all-in-one or a portable? Same performance, better form.
Yes, it is the theory that Apple can have reduced retail prices for their Apple Silicon machines because they don't need to pay for Intel, AMD, and graphics memory margins with Apple Silicon. That's the theory. So, why are MBP14/16 prices higher? I think NAND is cheaper now, so there are cheaper prices from cheaper NAND GB/$ costs, but Apple is basically using the same price points for all their Apple Silicon Macs. The M1 Mac Mini is cheaper, but the M1 Pro MBP14/16 base SKUs are about $200 more than prior base SKUs. Where's this price advantage?
An M1 MBA at $800 would sell a lot! An iMac 24 at $1000? Awesome price point. Or, why isn't the base MBP14/16 SKUs at $1800 and $2300?Even you estimate an iMac Pro with M1 Max Duo to cost $5000. Why not $4000? So, wait on see the prices. Like I said, so far, they seem to be sticking with their established price tiers give or take 10%.
The price advantage doesn't affect entry models so much as there's a cost floor. The market for premium notebooks will also easily bear $300 chips so there's no point in selling them at $50 and making $10 profit, that's just throwing profit away. $10 x 20m units = $200m, $260 x 20m = $5.2b, there's no point in throwing away $5b when the buyer doesn't care that much. Even if unit sales went up 5x at the lower price (which they wouldn't), it's still not worth it.
It's much the same with $800 vs $1000 laptops, people are happy to pay $200 more for premium laptops and it allows for resellers offering discounts. Some of Apple's component prices are a bit excessive. They charge $400 for 16GB RAM when choosing the Max upgrade. The old 5600M was an $800 upgrade and they only charge $400 for the M1 Max chip upgrade but force you to take the $400 RAM upgrade with it so that it's the same $800 upgrade. The good thing is the Pro model is equivalent performance to 5600M and significantly less expensive.
I actually wrote $3,999 for the Duo model iMac Pro originally because that would be closer to the component prices but if they use more expensive memory, $4,999 is a possibility. I'm avoiding being overly optimistic. Another possibility is that they reserve the XDR display panels to higher models and charge a premium. If they have an XDR 27" iMac Pro with Max Duo for $3,999, that would be amazing.
When comparing this to the Mac Pro style upgrades, I can't see them having a $4,999 Max Duo entry point with a selector box to M1 Max Quad priced at over $18,000, that's what they have now with Intel/AMD CPU and GPU upgrades. I reckon that selector box will at most be $3,000, possibly with a memory upgrade on top. Even with a $2k memory upgrade that's $9,999. Even with this much cheaper price, Apple still makes the same kind of profit because the difference is avoiding sending around $10,000 directly to Intel/AMD.
GPUs are going into the HPC server space. All codes that solve systems of differential equations (numerical codes in physics/engineering fields) are transitioning to GPU compute. It's crazy not to as it can be a 10x performance lift. Data science and crypto are using GPU compute, at least until the specialist tensor or machine learning chips come. Apple has an advantage here with Apple Silicon as current GPUs don't have a lot of memory, which a lot of numerical codes really like. The window is closing though. One Apple Silicon 128 g-core GPU isn't enough. It's great that it will only need about 250 Watts. Well, it's even better if Apple can put 4 of them in a box.
It's a lot of engineering effort to make that kind of box with 4 x 4-die tiles and the use case is really specialist. The people doing this kind of work can easily network 4 or more boxes together to achieve the same goal and Apple will still make 4x the sales. It helps reduce expensive refurb inventory. The other thing to consider is what else those users could be buying as an alternative. Best case would be a Threadripper plus dual 4090 around $10k. Apple's $10k option might be slower than this but it's competitive and people will think twice about getting the PC.
Do you think this "Pro Workflow Team" that Ternus trots out in statements and interviews is a kind of performative art? This is in response to your "trying different strategies" comment. For some reason, they think it is comforting for potential buyers to hear it. I think it is downright frightening as it implies, if it is a forthright statement, that they didn't know how this market segment uses computers.
I do think that if they updated the 2019 Mac Pro with Apple Silicon, they can put a M1 Max Quad in an 2-wide MPX Module and put 4 of them in a Mac Pro. I also think that the Mac mini shouldn't have a square footprint. It's just not a great shape for cooling. If it was say 12 x 6 x 1.5 inches, higher performance blower fans or axial fans could be used, and they'd probably be capable of putting anything from an M1 to M1 Max Duo in it.
High-end computing users have different needs and there are segments who respectively loved the 2013 Mac Pro, the iMac Pro and the 2019 tower and currently still own them and are likely different portions of buyers at a similar price point with some overlap. Ideally, it would be possible to have a single form factor for all of them but I don't think that's going to be feasible.
People's use cases are also not permanently tied to the same form factor. For a long time, musicians and video editors were accustomed to using add-in co-processor boards simply because the computers weren't fast enough. Apple even made one with the AfterBurner card. Most of these aren't needed any more, there's no need for Apple to continue selling an AfterBurner card.
High capacity storage, fast IO connection etc are needed too so there's justification for having fast expansion capability. That can be done with a single slot and let people do what they want.
DuhSesame said: I don't know what Marvin means by "SFF" but it's no trash can. I'd say at least four double-wide slots. On the other hand, you're not using all eight slots in the current design, at least five are occupied (I/O card, MPX modules).
I really liked the cylinder form factor, it's a very Apple-like design, it just wasn't powerful enough, wasn't upgraded over time (supplier fault) and had no ability for users to expand on their own. If it had a single PCIe slot at the back, there wouldn't have been any problems with it. People would have dropped new GPUs and IO cards into external boxes.
The iMac Pro probably had the potential to sell more units but it was at an even higher price point than the cylinder which cuts the potential market back down. Unit volume was likely similar to the cylinder but with more revenue. Still not enough to justify making them.
The 2019 Mac Pro was at a higher entry point again without a display so unit volume is even lower. Apple never mentions Pro models in their earnings reports because they don't make enough revenue. If Apple made $1b per year in the Mac Pro line, that's around 100-150k units per year compared to 20-30 million Macs and there's no indication they even sell that much. HP is the biggest in the industry and they make $1.6b in workstations and it's shrinking year over year while notebooks keep growing.
The challenge in making new workstations is figuring out what market is left and what would convince them to buy a new one. If someone owns a 2019 model like 28-core, 128GB RAM, 10TB+ internal storage, 20TFLOPs+ GPUs, what would convince them to buy a new one? It's very hard to do no matter what route they go because it's likely that model satisfies their needs and many can easily migrate down to an iMac Pro. Some video editors are migrating to M1 Max MBPs from iMacs because they are so powerful now.
Every single Mac Pro since 2012 will be a commercial failure because there's not enough users to offset Apple's manufacturing, support and marketing investment. They are probably stuck catering to whatever enthusiast crowd remains but this used to be the GPU enthusiast crowd and there won't be GPU expansion internally. Then there's the co-processor users who don't need cards any more. That leaves people with IO needs. I think a cylinder or cube-like machine with a single slot is enough for the remaining market but I expect that the people placing the orders will be influencing the design of it. The easiest route for Apple by far is to cut the 2019 model in half and leave some PCIe slots for IO expansion.
Comments
https://www.reddit.com/r/pcmasterrace/comments/6r87vn/mini_itx_for_amd_ryzen_threadripper/
That's just a concept as no ITX can serve a TR4. Now we're talking about M3 with 4 dies and 16 RAM chips. Then throw two, huge air coolers. With more than 8 PCIe slots.
That's a minicomputer dude, without the power of a minicomputer.
There is a software networking effect as well. Having software support sells hardware up and down the line. For this type of software, missing the top end effectively limits the size of the market can sell devices into. They had to know this once they committed to making the 2019 Mac Pro in 2017. They are way behind CUDA with GPU compute optimized software. Lots of work for them to do with Metal to catch up, along with other areas.
You’ll never be enough if you just chasing the “pure performance”, someday you’re gonna build a supercomputer.
Yes, Apple has to choose a design point for the Apple Silicon Mac Pro, just as they did for the 2019 Mac Pro with using 1 CPU socket versus 2. (I saw that as a problem back then too). The rumors so far sound like the Apple Silicon Mac Pro will be a 500 W SFF workstation like the 2013 Mac Pro or something the size of a G4 cube. Those machines weren't successful. How is the Apple Silicon version going to be any different? Being quiet and small really isn't a big selling feature for buyers of this type of machine. x86 competitors will match or exceed its performance at 1000 to 1500 W.
We will have the same conversation that we did when the 2013 Mac Pro came out, if this Apple Silicon Mac Pro is a SFF workstation. What's going to be different this time? You say Apple needs to make a product that sells. The G4 cube and 2013 Mac Pro didn't sell and were retired after 2 or 3 years. If the Apple Silicon Mac Pro are like them: small, limited capacity, not modular, what's going to be different this time?
A 4-die will be a ~400W chip. This is Apple Silicon we're talking about, able to match & exceed something twice as power-hungry. The last 3275M is about 280W running P95.
Then realized that AMD only offers a 32/64 lineup, there's no in-between. Either you got 32-core or go straight to the best.
The big difference will be performance per dollar. Some people make an assumption that the Cube was wrong, 2013 model was wrong, iMac Pro was wrong and the 2019 model was right based on size and upgrades but there's nothing to indicate the 2019 model was any less of a failure than the others as far as sales performance goes. The market at these price points is minuscule compared to the mainstream by at least a factor of 100.
What always affects users in higher-end models is performance per dollar. Apple's top-end models that can be $40k or more perform worse than Threadripper machines all because of Intel. Apple's own silicon can outperform even AMD at way more competitive price points.
If they stick at half the power level of the competition then in a few years time, the competition can catch or exceed them and for a fraction of a fraction of the users, they will make a choice to switch platform but that's assuming they need 2x the performance in a single workstation. If they only need 2x the performance and their use case can be networked, they can buy 2 or more.
Over the years, computers like the 2013 model, Cube, iMac Pro would run into limitations like handling 4k/8k editing. This isn't going to continue to be a problem, the video editing industry isn't going to move the mainstream beyond the limits of visual acuity for authoring. Once the performance demand in the video space dies down, that cuts out most of the demand for higher-end hardware and a factor of 2x difference won't be a deal-breaker for anyone, especially when Apple's custom media encoders outclass everything by a significant margin.
There’s the feeling that Apple needs to be the best or we’ll be shamed to oblivion. The PC community acts like that, it’s all about the “fastest on the planet” because “it’s better”. There’s really no end of that. There’s a belt curve for everything, which at some point the gain becomes a dread that doesn’t worth to keep it. I don’t recall anyone wants to live in a hassle with little to no return, that’s what exactly happened in workstation chips.
And yup, Apple has chosen not to address certain markets like "workstation" laptops that basically run 200 W worth of compute performance, or a consumer desktop box, or an MBA15, so on and so forth. We're only discussing where we think they are making mistakes or not. I think they have kept their lineup too simple and could address more of the market. The SFF Mac Pro is something they have tried before, twice depending on how you count the G4 cube, and I'm more skeptical then inviting this time around.
I think Apple loses on performance per dollar in desktop and high end hardware. They win on performance per Watt, but performance per dollar? No. They will maximize their prices so that performance/$ is about the same as x86 competitors, while x86 competitors are much more price elastic. As their usual practice, Apple will likely use their industrial design intangibles to keep prices high even when x86 has better performance/$.
If Apple uses half the power, they won't have a lead in performance, right? x86 can match the performance of M1 Mac Duo and Quad configurations with 16, 32, 64 core machines with dGPUs. They'll be using 2x the power, but the performance will be the same or better. For prices, wait and see. The Apple Silicon machines so far is saying Apple is going to keep their same price tiers, give or take 10%. So, I don't see them having better performance per dollar on the high end. The performance/Watt is great for thin, integrated machines, but in desktops, performance/Watt and integration are not selling features.
The G4 Cube, 2013 Mac Pro and iMac Pro were limited by design, not by market movement, right? They were limitations stemming from a design choice, and choices they made over and over. The G4 cube was a dead end because the G4 CPU was a dead end and the G5 and successor dGPUs was too hot to use passively. That's the same dead end they had with the 2013 Mac Pro, right? I also think they chose not to redesign the interior of 2013 Mac Pro for 1 200W GPU and 1 150W Xeon, and moved forward with the iMac Pro instead. Then, even before they shipped the Mac Pro they knew it wasn't going to be the right product and told the market that the 2019 Mac Pro was coming. They finally arrived at the product that was the most sensible.
It's hard for me to understand how you can make a prediction that we will arrive at enough computing performance with 8K video editing. It supposes that Apple isn't interested in other types of computing which will require a lot of performance for which a workstation would serve, and that we know how video production will be done in the future. Designing a SFF and integration workstation around a perceived future would compound the mistake.
You can go 600 watts overclocked, somewhat stable, with a chiller for cooling. Must be so productive.
Some of the higher-end configs reach over $40k:
https://www.apple.com/shop/product/G0ZCDLL/A/Refurbished-Mac-Pro-25GHz-28-core-Intel-Xeon-W-Two-Radeon-Pro-Vega-II-Duo-Apple-Afterburner
A significant portion of this is Intel/AMD pricing (over $17k of the price). Another huge portion is the RAM and SSD upgrades used and Afterburner. Apple charges $700 for M1 Max upgrade. Even adding $300 for the base chip, a quad version can be priced at $4k, which is equivalent to 2x Intel 28-core and just below quad Vega GPUs, which are priced at $17k.
This would be comparable to a Threadripper PC with a next-gen Nvidia 4090, which you would be able to get for around $6k late 2022. They won't beat cheap PCs simply due to margin but they can be very competitive on performance per dollar in the workstation space with their own hardware.
I'd consider having the same performance at half the power to be leading. It results in less throttling, fan noise. They can also boost clock speeds over standard M1 Max.
The limitations were partly by design but not because of Apple. The fault was always with their suppliers. The choice to use PPC was a flaw because the manufacturers couldn't keep improving, especially on mobile. All the heating problems Apple has ever had came from Intel/IBM/AMD/Nvidia. Apple has never had these problems with their own hardware.
If Intel and AMD had kept pace with TSMC and mobile chips, the Mac Pro would be the same 2013 model today but the suppliers didn't deliver. Intel's performance over the last 10 years has been really poor and probably on purpose to make more profit. You could debate whether not having a design to accommodate low performing suppliers was the fault or the suppliers themselves but regardless, that won't happen again because Apple is doing it themselves now.
Plus they can still accommodate some flexibility. Even back in 2013, I said they can easily put a single PCIe connector on and let people do what they want outside the box. If people want to attach 8x 40TFLOPs GPUs with a 4kW PSU for machine learning in a box costing $30k, that's fine. The workstation itself doesn't have to be designed to accommodate those extreme use cases internally.
Anyway, this generation of MacBook Pros is future-proof enough, there are fierce competitions going on. An 8+2 M1 Max, despite most believed today, is not enough in both performance & the competition. Intel will definitely be able to keep the pace, and sometimes even kills it. I'm sure Apple has a plan to go even bigger.
Apple knows that they have competitors in the market. If HP or Dell is offering a 38 core Xeon (Xeon-3375 with a 40 TFLOP GPU - Nvidia 3090 or A8000 whatever) for 20k. Will Apple price a M1 Max Quad, about equivalent to the stated Intel machine, for 10k? I'm really skeptical at 8K. 10K? Not giving me the warm fuzzies either. 15k? That sounds like something Apple would do: having a machine be competitive to others, and try to triangulate a price structure that makes their business profitable while not pricing themselves out of the market. Being way under competitor products? That doesn't sound something like they would do. They will take the money.
So, if x86 competitors are providing a price umbrella for Apple to be under, I think Apple is going to price themselves as close to the umbrella as possible. So 15k to 20k, if competitors are at 20k. Moreover, x86 OEMs are much more price elastic than Apple is. If Apple offers a M1 Max Quad config at $8k, and it is a threat to their x86 sales, they will lower their prices, and lower them to the point of having better performance/$.
But is that lead (performance/W) something that will make the machine successful? Is it a feature that will sell the machine if there is performance parity? Keep in mind that they don't have to have a SFF machine to have this performance/Watt lead. They could have it have the internal expansion capabilities of the 2019 Mac Pro and be as power efficient as a SFF model.
If there is performance/price parity, buyers are largely going to make decisions on software compatibility. This isn't a bad situation per se, it allows Apple to defend its content creation share, but gain more share above that? Probably not. Then, being a SFF obviously present issues given the history. So, wait and see what Apple is going to do here.
There isn't any fault nor assignment of blame. Apple was in equal footing to everyone else, with access to all the same components. Apple is responsible for predicting what type of hardware is available and what direction hardware and software will have in the future. They are also responsible for adjusting when they are wrong or something they expect doesn't happen. I'd expect that they should adjust much faster and to learn lessons from mistakes. This is why it is concerning that the rumors of the Apple Silicon Mac Pro is a SFF machine about the size of the G4 cube, and why I keep asking the "what's different this time question".
The Intel transition was their response for PPC fizzling out. Good call. But even there, they didn't bring the cube back with Intel hardware inside. It would have had more sales as the G4 version as the Mac Pro went up in price with Intel Xeons, and a lot of folks would have bought an Intel cube at $2000. But they offered an iMac G5 and iMac Intel models at those price tiers instead. Basically they think the iMac 27 is their mid-tier offering to folks who want what a typical PC desktop could provide in terms of performance, and that's carried through to this day. Nice offering, but it reduced the potential market of buyers. It really doesn't hurt to offer different form factors at the same price points.
For the 2013 Mac Pro, they could have redesigned it internally to have 1 300W GPU instead of 2 150W GPUs. Having it updated with new Xeons and new dGPUs would have kept sales up, and would have made the iMac Pro unnecessary. They could have updated it to TB3, and made accessories for it. But they had to have given up on it by 2015 or so when iMac Pro development started. So, they felt the SFF cylindrical form factor was a bust by 2015, and moved to the iMac Pro as their top end Mac. Obviously, they knew that wasn't enough and went full circle to the 2019 Mac Pro model.
A SFF Apple Silicon Mac Pro could get a lot of unit sales if it was cheap enough, but I'm pretty skeptical it will be cheap. I think it will be the same price tiers as today. It does lose sales to folks who want internal expansion, but perhaps it gains sales from potential iMac 27 buyers getting it instead, if it is cheap enough, like $2500 starting price, but no way, right? Apple still needs to get software transitioned to Metal. They still need to update it on preferably a yearly cadence to show that they are committed to it. And, they need to make accessories for it, including a monitor. It's a long road ahead.
1). Even Intel & AMD have their limits, they can't just make chips infinitely faster, either it'll be impossible to cool or the performance gain makes no sense (diminishing return says you'll be giving 2x more power for less than 20% in return, anyone wants that?). Apple Silicon will match their sweet spot with much less consumption.
2). Nobody says the new Mac Pro will be tiny, there'll be slots, we just don't know how much.
Normally when Apple maintains 40% margins on every component, that pushes their prices to around double equivalent PCs. Apple Silicon will allow them to maintain 40% margins at less than half the price as they aren't paying a markup to Intel/AMD.
Retail PC (also previous Mac) = ((Intel + margin) + (AMD/Nvidia + margin) + (extra components + margin)) + retail margin
Retail Apple Silicon Mac = (Apple core components + (extra components + margin)) + retail margin
It removes the most expensive profit margins and separate component costs from 3rd parties using components that are dirt cheap because they are used in mobile devices.
In clusters it's more important than a single workstation. Crypto miners go for the best performance-per-watt over raw power. For a single workstation, it's still worthwhile to have a stable machine that won't be overheating or noisy running for long periods of time. If it's the same performance, there's no benefit to the noisier one.
I think part of this was trying different strategies to boost sales and nothing has been working. People are assuming the 2019 tower is working only because it's the latest iteration. Once they go back to a smaller form factor, people will see that the tower didn't boost sales either. It doesn't matter what they make above $5k, this market is tiny.
As I say though, a single PCIe slot is enough to satisfy everyone while having a small form factor.
One route they could go is to make it flat like behind an XDR display and the internals could be lifted out into a server blade form factor. With a PCIe slot, this could be connected to another blade that has IO cards or GPUs in it. It still won't sell many units but it works for multiple use cases. Musicians and artists who want a compact product would have a single display. Server users get a compact blade. People who want expansion can either buy the blade for the headless form factor and plug in the extra box or buy the iMac and plug in the box. If someone refurbs any of them, they can be resold for the other use cases.
Yes, it is the theory that Apple can have reduced retail prices for their Apple Silicon machines because they don't need to pay for Intel, AMD, and graphics memory margins with Apple Silicon. That's the theory. So, why are MBP14/16 prices higher? I think NAND is cheaper now, so there are cheaper prices from cheaper NAND GB/$ costs, but Apple is basically using the same price points for all their Apple Silicon Macs. The M1 Mac Mini is cheaper, but the M1 Pro MBP14/16 base SKUs are about $200 more than prior base SKUs. Where's this price advantage?
An M1 MBA at $800 would sell a lot! An iMac 24 at $1000? Awesome price point. Or, why isn't the base MBP14/16 SKUs at $1800 and $2300?Even you estimate an iMac Pro with M1 Max Duo to cost $5000. Why not $4000? So, wait on see the prices. Like I said, so far, they seem to be sticking with their established price tiers give or take 10%.
GPUs are going into the HPC server space. All codes that solve systems of differential equations (numerical codes in physics/engineering fields) are transitioning to GPU compute. It's crazy not to as it can be a 10x performance lift. Data science and crypto are using GPU compute, at least until the specialist tensor or machine learning chips come. Apple has an advantage here with Apple Silicon as current GPUs don't have a lot of memory, which a lot of numerical codes really like. The window is closing though. One Apple Silicon 128 g-core GPU isn't enough. It's great that it will only need about 250 Watts. Well, it's even better if Apple can put 4 of them in a box.
Nvidia is dominating with their server GPUs. Intel is following with Ponte Vecchio and AMD is also following. It's a lot like the CPU server boards, 4 to 8 GPU sockets in a 3U or 4U rack unit. I can definitely understand that Apple doesn't want to be in the server space, but they really shouldn't short shrift themselves and the potential uses of their hardware by not designing their top end hardware to all practical limits.
The workstations won't be overheating because their cooling systems are designed to remove 1000 to 1500 W of heat out of the box, just like the 2019 Mac Pro is. They'll likely be loud, will make the room they are in hotter, but they will work fine and generate their advertised performance.
The ostensible advantage of x86 workstations is being able to run 2 to 4 GPUs in the box, have 20 to 50 TB of storage in the box, support >1 TB of RAM, and software compatibility. Ie, flexibility. It's also cleaner as all the stuff is in the box. What are they gaining with a Mac Pro in a SFF box at the same performance? It uses less power, is quieter. If the user has external storage, PCIe breakout boxes, it becomes a sprawl of boxes and cabling. Not nice. You and I will have to agree to disagree that performance/W is a selling feature in workstations. I think people will choose a big box over a small one because the internal expansion is more valuable in this market than being small, quiet and efficient.
There were long threads on AI over the Mac Pro SFF design in 2013. All the criticisms were basically right. I'm not looking forward to a repeat of those threads if the Apple Silicon Mac Pro is a SFF machine.
Yes, the unit sales is tiny compared to consumer machines, but Apple themselves see it as needed. Otherwise, they would not have made the 2019 Mac Pro. Ignoring some market segments can have negative effects on sales of other hardware. No Mac workstations may mean less Macbook Pro sales. There's a negative feedback effect that can result in less software support, which further erodes sales, further decreasing Mac sales. They can't ignore market niches without anticipating the effect of sales of their other products. So at least for the sake of their content creation market share, a workstation Mac has to be available.
Do you think this "Pro Workflow Team" that Ternus trots out in statements and interviews is a kind of performative art? This is in response to your "trying different strategies" comment. For some reason, they think it is comforting for potential buyers to hear it. I think it is downright frightening as it implies, if it is a forthright statement, that they didn't know how this market segment uses computers. It could very well be true, they didn't know how the market segment uses computers, and they thought the 2013 Mac Pro and 2017 iMac Pro were going to sell. It's too hard to believe that a computer company doesn't know how computers are used, in most market segments at least. They must of have had a vision of how this market uses computers, felt that they could get users there, and felt that time would eventually make it true. I keep saying a book has to be written about what happen with Mac hardware from 2012 to 2018 or so, to reveal WTF happened.
They already have a workstation product that meets the needs of most of its market: the 2019 Mac Pro. They sell rack versions of it, and it can be configured in all sorts of ways. Like I said, people in this market aren't looking for a small box. For people who are looking for a SFF box, Apple can just update the Mac mini with a M1 Max. With Ethernet over Thunderbolt, a stack of them can by networked together at something like 30 Gbits/s for people who want a compile farm or run some small scale jobs that can be run over a network. And, OWC miniStacks can used to add storage.
I do think that if they updated the 2019 Mac Pro with Apple Silicon, they can put a M1 Max Quad in an 2-wide MPX Module and put 4 of them in a Mac Pro. I also think that the Mac mini shouldn't have a square footprint. It's just not a great shape for cooling. If it was say 12 x 6 x 1.5 inches, higher performance blower fans or axial fans could be used, and they'd probably be capable of putting anything from an M1 to M1 Max Duo in it.
Again, no they don't. The PSU is capable of putting 1.5KW of power, though that doesn’t mean you’ll use all of that (it needs overhead, come on) or all the power will transfer to heat. The most powerful CPU out there tops at ~300W, don't make it sound like cooling a 1000w chip is easy, if not downright impossible. With that in mind, not every workstation needs 4GPUs, depending on what you're doing.
"I think people will choose a big box over a small one because the internal expansion is more valuable in this market than being small, quiet, and efficient."
Okay, I agree. I don't know what Marvin means by "SFF" but it's no trash can. I'd say at least four double-wide slots. On the other hand, you're not using all eight slots in the current design, at least five are occupied (I/O card, MPX modules).
"If it was say 12 x 6 x 1.5 inches, higher performance blower fans or axial fans could be used, and they'd probably be capable of putting anything from an M1 to M1 Max Duo in it."
Or not? The 16" MacBook Pro had a cooling system of at least 120W. I think it can do more, but the point is thickness aren't that restrictive. Apple can just spin up the fan, then mini will be able to 2-die like magic. But why do that on an entry-level system? It won't be much cheaper, might be noisier, and potentially a lot hotter, too. Then, why not just get an all-in-one or a portable? Same performance, better form.
It's much the same with $800 vs $1000 laptops, people are happy to pay $200 more for premium laptops and it allows for resellers offering discounts. Some of Apple's component prices are a bit excessive. They charge $400 for 16GB RAM when choosing the Max upgrade. The old 5600M was an $800 upgrade and they only charge $400 for the M1 Max chip upgrade but force you to take the $400 RAM upgrade with it so that it's the same $800 upgrade. The good thing is the Pro model is equivalent performance to 5600M and significantly less expensive.
I actually wrote $3,999 for the Duo model iMac Pro originally because that would be closer to the component prices but if they use more expensive memory, $4,999 is a possibility. I'm avoiding being overly optimistic. Another possibility is that they reserve the XDR display panels to higher models and charge a premium. If they have an XDR 27" iMac Pro with Max Duo for $3,999, that would be amazing.
When comparing this to the Mac Pro style upgrades, I can't see them having a $4,999 Max Duo entry point with a selector box to M1 Max Quad priced at over $18,000, that's what they have now with Intel/AMD CPU and GPU upgrades. I reckon that selector box will at most be $3,000, possibly with a memory upgrade on top. Even with a $2k memory upgrade that's $9,999. Even with this much cheaper price, Apple still makes the same kind of profit because the difference is avoiding sending around $10,000 directly to Intel/AMD.
It's a lot of engineering effort to make that kind of box with 4 x 4-die tiles and the use case is really specialist. The people doing this kind of work can easily network 4 or more boxes together to achieve the same goal and Apple will still make 4x the sales. It helps reduce expensive refurb inventory. The other thing to consider is what else those users could be buying as an alternative. Best case would be a Threadripper plus dual 4090 around $10k. Apple's $10k option might be slower than this but it's competitive and people will think twice about getting the PC.
High-end computing users have different needs and there are segments who respectively loved the 2013 Mac Pro, the iMac Pro and the 2019 tower and currently still own them and are likely different portions of buyers at a similar price point with some overlap. Ideally, it would be possible to have a single form factor for all of them but I don't think that's going to be feasible.
People's use cases are also not permanently tied to the same form factor. For a long time, musicians and video editors were accustomed to using add-in co-processor boards simply because the computers weren't fast enough. Apple even made one with the AfterBurner card. Most of these aren't needed any more, there's no need for Apple to continue selling an AfterBurner card.
High capacity storage, fast IO connection etc are needed too so there's justification for having fast expansion capability. That can be done with a single slot and let people do what they want.
I really liked the cylinder form factor, it's a very Apple-like design, it just wasn't powerful enough, wasn't upgraded over time (supplier fault) and had no ability for users to expand on their own. If it had a single PCIe slot at the back, there wouldn't have been any problems with it. People would have dropped new GPUs and IO cards into external boxes.
The iMac Pro probably had the potential to sell more units but it was at an even higher price point than the cylinder which cuts the potential market back down. Unit volume was likely similar to the cylinder but with more revenue. Still not enough to justify making them.
The 2019 Mac Pro was at a higher entry point again without a display so unit volume is even lower. Apple never mentions Pro models in their earnings reports because they don't make enough revenue. If Apple made $1b per year in the Mac Pro line, that's around 100-150k units per year compared to 20-30 million Macs and there's no indication they even sell that much. HP is the biggest in the industry and they make $1.6b in workstations and it's shrinking year over year while notebooks keep growing.
The challenge in making new workstations is figuring out what market is left and what would convince them to buy a new one. If someone owns a 2019 model like 28-core, 128GB RAM, 10TB+ internal storage, 20TFLOPs+ GPUs, what would convince them to buy a new one? It's very hard to do no matter what route they go because it's likely that model satisfies their needs and many can easily migrate down to an iMac Pro. Some video editors are migrating to M1 Max MBPs from iMacs because they are so powerful now.
Every single Mac Pro since 2012 will be a commercial failure because there's not enough users to offset Apple's manufacturing, support and marketing investment. They are probably stuck catering to whatever enthusiast crowd remains but this used to be the GPU enthusiast crowd and there won't be GPU expansion internally. Then there's the co-processor users who don't need cards any more. That leaves people with IO needs. I think a cylinder or cube-like machine with a single slot is enough for the remaining market but I expect that the people placing the orders will be influencing the design of it. The easiest route for Apple by far is to cut the 2019 model in half and leave some PCIe slots for IO expansion.