Also, remember that we had COVID pandemic shutdown causing a significant turnover with employees in many tech companies during that time - the effects of shutdowns are often not felt until a couple of years later. And now we're already seeing the effects.
Another consideration is that we have reduced our dependence on China and that means the age of cheap and quick consumerism is fading.
Also, why do we really need Apple computers for high end computing workstations which often have only a specific purpose? It just makes no economic sense. If consumers need 512GB of RAM, then they would just use a customized PC and that's gonna work just fine - pretty sure for every Mac app built for that, there is also a Windows app. It's not like they cannot do that. They can. They're called pros for a reason.
The iMac Pro doesn’t have the power and expandability of a Mac Pro. iMacs are still thermally constrained compared to Mac Pros. iMacs are not a headless form form, and you want a headless form for rack mounting.
COVID could be the problem as it affected everything.
We have not reduced our dependence on China, however businesses have started to reduce their dependence on China.
We have high end Macs because Pros demand high end Macs for their workflows, because they need high end Macs. Xcode doesn’t run in Windows. You want high end Macs to deal with large software development, compilation and testing. You want high end Macs because your workflow is Mac based. Yes a lot of people can just flip to Windows, however high end software is very expensive and so is retraining everyone. Regular CPUs don’t have the built in specialized MV, video processing, and GPU. Regular CPUs are not as power efficient. This power efficiency adds up in racks of equipment, large businesses. and across the world as more Macs are installed. That power efficiency, reduces air conditioning load. It is cheaper to install more efficient equipment than to install solar to make up for that inefficiency. Companies are being pushed to be more eco friendly.
Still, what is the state of professional ASi native apps and plugins? Yes there is Rosetta but full speed won’t be there.
The only way an M2 Ultra goes in a Mac Pro is if Apple developed an external-to-SOC traffic controller that mimics how their Fabric works - and then add multiple M2Ultra packages in a “modular” config.
This seems the most logical route for them to go. The Mac Pro is such a small shipment volume (<50k units per year), it doesn't make sense to manufacture a custom chip that would only go in the highest-end model where the shipment volume drops to under 10k units/year and will just continue to fall as upgrade cycles lengthen and people migrate to cheaper models.
They can sit a large heatsink on two ultras next to each other and join them with a slower interconnect.
If the interconnect causes a bottleneck for some tasks, they can make it optional to use both together. In many cases, pro apps will scale across both. Video apps like After Effects, Da Vinci, Final Cut rendering will double in performance as they can render separate frames on each.
256GB RAM support would be enough but an option for 512GB would probably be useful for working with 8K frames. 8K60 10-bit uncompressed frames = 10GB RAM/second. Fast SSDs negate the need for lots of RAM in many scenarios now.
Add in a single x16 PCIe 5/6 connector and people can connect whatever else they want externally.
On 3nm, a dual Ultra would be 80TFLOPs and would be roughly equivalent to an Nvidia 4090 combined with a dual i9-13900k and the price would be under $10k, which is the price point most Mac Pro buyers are at. A 5nm model would perform ok too, it would perform like the higher-end 2019 Mac Pro at the mid-range price point and much smaller, cooler, quieter under full load.
Why would 2019 Mac Pro performance be acceptable in 2023-2024? Where the Xeons in the 2019 Macs were 14nm, Intel releases 10nm Xeons in 2023 and 6nm Xeons in 2024. The 14nm Xeons in Macs topped out at 56 threads. As a mere i9-13900K has 32 threads on 10nm, who knows how many threads will be on the 6nm Xeon.
Yes there is real value to having the best hardware available to run proprietary Apple software. But 2019 Mac Pro performance will have people who use third party software jumping ship to 2024 AMD Threadripper and Xeon machines. Yes they will be at the mid-range price point but the 2019 performance means that you will need two of them.
Apple is going to have to face the same challenge that ARM server manufacturers currently have, which is to approximate Xeon and Epyc performance. Nvidia's ARM server CPU - arriving 1Q2023 - does this by starting at 72 cores.
Also, remember that we had COVID pandemic shutdown causing a significant turnover with employees in many tech companies during that time - the effects of shutdowns are often not felt until a couple of years later. And now we're already seeing the effects.
Another consideration is that we have reduced our dependence on China and that means the age of cheap and quick consumerism is fading.
Also, why do we really need Apple computers for high end computing workstations which often have only a specific purpose? It just makes no economic sense. If consumers need 512GB of RAM, then they would just use a customized PC and that's gonna work just fine - pretty sure for every Mac app built for that, there is also a Windows app. It's not like they cannot do that. They can. They're called pros for a reason.
It wouldn’t be a Mac.
The whole reason people buy Macs to begin with is the far superior experience.
So, no. A “chesp” pc (which will actually cost similar or more to a Mac at the same high end specs) won’t cut it.
Mac Pro vs jimmy rig… yeah I’ll get the Mac Pro.
Why buy an “expensive” Mac when you can buy a “cheap” pc? Oh I don’t know, probably the same reason people buy “expensive” iPhones when they can have a “cheap” subsidized android instead?
Come on dude. You know the answer.
If you think that workstations and servers running Red Hat Linux or Windows Server are "jimmy rigs" comparable to Android phones, you are really on the wrong thread. Seriously. These people are talking about needing 1.5 TB of RAM so their rendering jobs will get finished in 5 hours instead of 2 days and you are talking about "cheap PCs"? What PC supports 1.5 terabytes of RAM? Name it! And "user experience". Do you honestly think that engineers doing 3D CAD simulations care about this? Seriously you are on the wrong thread.
Yeah, that's just not going to be enough. However...it's a system on a chip. Why can't we have the current Mac Pro cheesegrater tower with a way to add multiples of the Ultra SOC? Add-on SOC boards? GPU render times have to start catching up with the latest GPU offerings from Nvidia and AMD. My Mac Studio Ultra is an incredible box, but it still doesn't have the raw graphics horsepower of the dual Vega II Duos I have in my 2019 Mac Pro.
Not really surprising. Just read an article that the area reduction of the new smaller nm processes is very small for SRAM this time around. So getting more memory using the SOC design of the Apple Silicon chips is not that easy (the bigger the area surface the higher is the production loss), which would make that think probably extremely expensive to manufacture (especially compared to high end competition on the Intel/AMD side). The power consumption advantage of Apple Silicon is not that big a deal breaker in these kind of boxes as it is in most other Apple HW.
So my guess. We'll probably see an upgraded (M3?) CPU that supports 'external' memory expansion and probably some smart way to handle that extended memory efficiently and at speed. Which probably needs changes to the silicon itself as well as to the kernel memory management. Kind of feels like the MS-DOS days with the 640kb limit and that highmem BS. B/c no matter what memory outside of the builtin SOC memory will be slower than SOC builtin memory.
Don't forget: there is a need on the market for these boxes, but it is comparatively very small market size. These people are willing to spend a ton of money on these boxes and need the high speed/bandwitch/extensability and have been poorly served for years by Apple in the last few years (very few Hw updates rather late in the market plus the Trashcan Mac debacle). Makes me wonder how many users still bet in Apple for these use cases.
Lot's of these very special applications are probably not (yet?) optimized for Apple Silicon (but rather for Intel CPUs). Now that Apple moved away from Intel the manufacturers need to put quite a bit of effort into Apple Silicon optimization. So that takes longer and/or might never happen, since Apple is still a rather niche market and most of thes Apps are running on Intel boxes as well. So the transition cost for SW-Manufacturers is high as well.
One reason why I still believe that the Mac Pro (even though it's not exactly cheap) is probably barely able to be cost effective for Apple to produce. This product category is (financially) probably more about bragging rights and less about actually earning real money. So low-priority update from Apple's point of view.
Last but not least: Remember what happened with the last transition? Apple announce the Intel transition is over even though there were no Apple XServer available with Intel CPUs (they just stopped selling servers for months) .. and in the end the XServe category died (in Hardware) and the Server Software as well in the end. Not sure if Apple intends to continue these high-end pro boxes much longer.
Regarding multi-processor systems— the Ultra is a dual Max system with a very high bandwidth interconnect. The ‘extreme’ was purported to be four Max with a very high bandwidth interconnect. I’m inferring that the ‘issue’ with the Extreme must be an issue with the interconnect. Apple has been consistent, detailed, and enthusiastic in touting the advantages of unified, high speed memory access. Their software model is built on that assumption.
So… there will not be a multi-socket Mac Pro. There will not be NUMA. There will not be DIMMs. There will not be compromises in terms of memory bandwidth or latency.
If they can’t interconnect four Max dies now, maybe they will later. Or maybe at 3nm they’ll just increase the number of cores on a single Max die.
In the meantime, I hope they sell an M2 Pro Mac mini — maybe I’ll buy four of them.
Grid computing is where Apple should be focusing. Leverage processing power from every device on the local network transparently.
The 'pros' who would benefit most already know how to do that, so there's probably not too much more that Apple needs to do for them.
For example, I've been waiting for my dream Mac Pro for a long time. The 2019 version came close, but it was held back by the choice of a crappy Intel Xeon (I bought a Linux ThreadRipper system instead).
I'm getting my first Apple Silicon computer (an M2 MBA) this week, so I can test for myself how the M2 does with my particular workloads. If the performance is what I'm hoping it will be, then I could see buying four M2Pro Mac minis (if such a thing is ever released and if the price is reasonable). I know exactly how to use such a configuration to meet my needs.
Why M2 rather than a new ThreadRipper? Three reasons:
1. The price of Threadripper has gone through the roof 2. AMD is now lagging Threaderipper updates far behind the mainstream desktop or EPYC 3. performance/watt of x86 sucks
Apple didn’t delay the Mac Pro two plus years in order to give us a Mac Studio with a different name.
The new Mac Pro won’t arrive until apple is ready to blow the doors off everything else - even if that means waiting for M3 Extreme.
The M2 just isn’t the destroyer hoped for. It’s great, but not something that will meet expectations of the delayed pro.
M3 has been for a long time where the convergence of all the good things was headed. It may mean Apple breaks a promise, but it’s better than releasing something prematurely just because an ambitious project didn’t work out in time.
The only way an M2 Ultra goes in a Mac Pro is if Apple developed an external-to-SOC traffic controller that mimics how their Fabric works - and then add multiple M2Ultra packages in a “modular” config.
We don't know. It's very much possible that an 'M3 extreme' is also uneconomical.
If the family car dies, the owner does not have the luxury of waiting for the next model, he/she buys what is available on the deanship lot.
My 2013 MacPro was upgraded over time to 128GB of OWC Ram and a 2TB SSD from OWC as these became available There were also two Thuderboltv 27" displays on the wall. First one display died after nine plus years and there were upgrade issues for the MacPro requiring the OEM 1TB SSD be reinstalled so Apple could lock the firmware so no more upgrades could happen.
I ordered my Mac Studio with the 64 core GPU, 128GB shared memory and 8TB SSD along with two Apple Studio displays with VESA mounts that were to reuse the existing wall mounts. The $10,000 system was here and operational in September. I have no regrets as it does everything I want and will last me for many years (age 77).
I delayed ordering the system until the initial bugs were cured in the Studio Displays. There were no issues in upgrading the display MacOS software when they arrived. The Mac Studio arrived earlier than projected as production speeded up in China.
Now there are "spares" in the Apple system so I am no longer unsupported if there were a mechanical issue in the Mac Studio. The Studio sits on my desk below a Studio Display. There is no sound coming out of that enclosure and it barely gets above room temperature to the touch.
This is the best Apple computer system I have owned since 1990.
Apple didn’t delay the Mac Pro two plus years in order to give us a Mac Studio with a different name.
The new Mac Pro won’t arrive until apple is ready to blow the doors off everything else - even if that means waiting for M3 Extreme.
The M2 just isn’t the destroyer hoped for. It’s great, but not something that will meet expectations of the delayed pro.
M3 has been for a long time where the convergence of all the good things was headed. It may mean Apple breaks a promise, but it’s better than releasing something prematurely just because an ambitious project didn’t work out in time.
The only way an M2 Ultra goes in a Mac Pro is if Apple developed an external-to-SOC traffic controller that mimics how their Fabric works - and then add multiple M2Ultra packages in a “modular” config.
We don't know. It's very much possible that an 'M3 extreme' is also uneconomical.
It is not that it is "possible." It is that it is the entire case. Threadripper, Epyc, Xeon, even ARM servers benefit from a combination of economies of scale and data center customers willing (needing) to pay whatever they cost. So it makes sense for the chipmakers (chip designers?) to pay the foundries to stand up the manufacturing lines. But it doesn't make economic sense for Apple to do the same for "Extreme" SOCs to be used in Mac Pros and iMac Pros that are only going to move maybe 10,000 units a year and still need to have a starting price of around $5000. This problem didn't exist with the Mac Pro because it used Xeon CPUs that Intel sells to everybody else, allowing them to be made and sold to Apple relatively cheaply.
The original 2020 M1 merely required adding 2 cores to what was basically a pre-existing smartphone chip at a manufacturer that had already been making octacore ARM chips for Android devices for years. By contrast M2 Extreme would have required horizontal combining 4 M2 chips to make by far the largest PC chip (in surface area) in history. Please note that Intel and AMD have abandoned the horizontal thing with 3-D scaling (vertical and horizontal) AND are atomizing the CPU parts into components - AMD calls them chiplets, Intel calls them tiles - because it is less expensive to manufacture and package in a motherboard. But even there, it is only financially feasible because Intel and AMD are going to sell a lot more Xeon and Epyc CPUs than Apple will Extremes.
I still maintain that Apple should go into the ARM server market. It would require them to emulate what Microsoft did with Windows Server and come out with a bona fide server version of macOS. Not only would they make a lot of money directly, but the byproduct would mean making workstation CPUs that would otherwise need to be manufactured in very small numbers for a few niche customers financially viable. So Apple, go ahead and make a competitor to Nvidia Grace https://www.nextplatform.com/2022/08/29/details-emerge-on-nvidias-grace-arm-cpu/ except limit it to small and medium sized servers!
It is not that it is "possible." It is that it is the entire case. Threadripper, Epyc, Xeon, even ARM servers benefit from a combination of economies of scale and data center customers willing (needing) to pay whatever they cost. So it makes sense for the chipmakers (chip designers?) to pay the foundries to stand up the manufacturing lines. But it doesn't make economic sense for Apple to do the same for "Extreme" SOCs to be used in Mac Pros and iMac Pros that are only going to move maybe 10,000 units a year and still need to have a starting price of around $5000. This problem didn't exist with the Mac Pro because it used Xeon CPUs that Intel sells to everybody else, allowing them to be made and sold to Apple relatively cheaply.
The original 2020 M1 merely required adding 2 cores to what was basically a pre-existing smartphone chip at a manufacturer that had already been making octacore ARM chips for Android devices for years. By contrast M2 Extreme would have required horizontal combining 4 M2 chips to make by far the largest PC chip (in surface area) in history. Please note that Intel and AMD have abandoned the horizontal thing with 3-D scaling (vertical and horizontal) AND are atomizing the CPU parts into components - AMD calls them chiplets, Intel calls them tiles - because it is less expensive to manufacture and package in a motherboard. But even there, it is only financially feasible because Intel and AMD are going to sell a lot more Xeon and Epyc CPUs than Apple will Extremes.
I still maintain that Apple should go into the ARM server market. It would require them to emulate what Microsoft did with Windows Server and come out with a bona fide server version of macOS. Not only would they make a lot of money directly, but the byproduct would mean making workstation CPUs that would otherwise need to be manufactured in very small numbers for a few niche customers financially viable. So Apple, go ahead and make a competitor to Nvidia Grace https://www.nextplatform.com/2022/08/29/details-emerge-on-nvidias-grace-arm-cpu/ except limit it to small and medium sized servers!
AMD made a bet on chiplet architecture, and it has paid off in a big way, mainly in the lucrative server market. This modular approach has allowed them to produce chips from 4 cores all the way up to 96 core monsters, all using the same basic core technology. But where they are really saving money is that they can use the advanced manufacturing processes on their cpu chiplets, which benefit from the advanced manufacturing process, but manufacture other parts of the chips at a less advanced process. They can mix and match various size chips, while getting higher yields vs a monolithic chip architecture.
If Apple wants to get back into the HEDT market, I think they either laser focus on specific segments(video, audio) and build chips around encoders that can handle massive workloads, or they have to move to a more modular design if they want to be more general purposes. As their chips grow larger and the manufacturing gets more expensive, each defect on the wafer will get more and more costly vs the chiplet/tile designs. That being said, Apple may have shown a bit of their hand with the silicon interposer in the M1 Ultra. If they can link two chips at such a high speed, whats to stop them from doing the same thing with modular chiplets in the future? That would allow them to scale without the issues of the monolithic chip designs. The other option would be to scale up as you mentioned and sell linux/unix based servers to operate in more efficient server stacks.
Disappointing rumors and hopefully it is not true. These rumors once again are showing that Apple has very little interest in high end PC hardware, and probably doesn't even know what its users want even with a "Pro Workflow" team. IOW, the Pro Workflow team is too niche with a bunch of Youtube videographers, and doesn't have enough engineering, 3D, information, and research oriented folks.
They should have shipped the M1 Ultra Mac Pro. Used the 2019 Mac Pro form factor. Added a PCIe controller. Kept the slots.
Apple didn’t delay the Mac Pro two plus years in order to give us a Mac Studio with a different name.
The new Mac Pro won’t arrive until apple is ready to blow the doors off everything else - even if that means waiting for M3 Extreme.
The M2 just isn’t the destroyer hoped for. It’s great, but not something that will meet expectations of the delayed pro.
M3 has been for a long time where the convergence of all the good things was headed. It may mean Apple breaks a promise, but it’s better than releasing something prematurely just because an ambitious project didn’t work out in time.
The only way an M2 Ultra goes in a Mac Pro is if Apple developed an external-to-SOC traffic controller that mimics how their Fabric works - and then add multiple M2Ultra packages in a “modular” config.
We don't know. It's very much possible that an 'M3 extreme' is also uneconomical.
It is not that it is "possible." It is that it is the entire case. Threadripper, Epyc, Xeon, even ARM servers benefit from a combination of economies of scale and data center customers willing (needing) to pay whatever they cost. So it makes sense for the chipmakers (chip designers?) to pay the foundries to stand up the manufacturing lines. But it doesn't make economic sense for Apple to do the same for "Extreme" SOCs to be used in Mac Pros and iMac Pros that are only going to move maybe 10,000 units a year and still need to have a starting price of around $5000. This problem didn't exist with the Mac Pro because it used Xeon CPUs that Intel sells to everybody else, allowing them to be made and sold to Apple relatively cheaply.
The original 2020 M1 merely required adding 2 cores to what was basically a pre-existing smartphone chip at a manufacturer that had already been making octacore ARM chips for Android devices for years. By contrast M2 Extreme would have required horizontal combining 4 M2 chips to make by far the largest PC chip (in surface area) in history. Please note that Intel and AMD have abandoned the horizontal thing with 3-D scaling (vertical and horizontal) AND are atomizing the CPU parts into components - AMD calls them chiplets, Intel calls them tiles - because it is less expensive to manufacture and package in a motherboard. But even there, it is only financially feasible because Intel and AMD are going to sell a lot more Xeon and Epyc CPUs than Apple will Extremes.
I still maintain that Apple should go into the ARM server market. It would require them to emulate what Microsoft did with Windows Server and come out with a bona fide server version of macOS. Not only would they make a lot of money directly, but the byproduct would mean making workstation CPUs that would otherwise need to be manufactured in very small numbers for a few niche customers financially viable. So Apple, go ahead and make a competitor to Nvidia Grace https://www.nextplatform.com/2022/08/29/details-emerge-on-nvidias-grace-arm-cpu/ except limit it to small and medium sized servers!
I'm not sure an Apple server would succeed. They'll be competing with HP, Dell and other large server's OEM's that have been doing business with SMB and enterprises for years. Most of those business have devices, services and applications from different vendors, and Apple is terrible integrating with other vendors. Another thing Apple lacks is the ecosystem companies as Microsoft have. They also would be competing with public cloud providers, considering a lot of business and enterprises are moving some or all workloads to AWS, Azure / MS 365 and GCP / Google Workspace.
I don't think creating an additional niche product would work at all.
A thought; how about an M2-Ultra, but multiple processors. Yes, the ultra, but 2, 4, or 8 of them?
Then maybe having the option to upgrade the processor later?
You do know that the Ultra is already two Max CPUs fused together, don't you? It also requires a 2 lb. heat sink for cooling. How do you plan on cooling multiple Ultra CPUs, which is already multiple Max CPUs? Apple doesn't know how to build the thing that would do better than the current Mac Pro. The problem is GPU performance. Apple Silicon can't compete in GPU with dedicated high performance cards that are offered in the Mac Pro. It would be an embarrassment to release a Mac Pro that caps out at 128GB of Memory when the Intel Mac Pro supports 1.5TB of Memory. And they are not going to release a Mac Pro with lower GPU performance due to Apple Silicon.
That is why Apple hasn't replaced it, and it is going on year 3 of their 2-year transition. They bit off more than they can chew.
Apple didn’t delay the Mac Pro two plus years in order to give us a Mac Studio with a different name.
The new Mac Pro won’t arrive until apple is ready to blow the doors off everything else - even if that means waiting for M3 Extreme.
The M2 just isn’t the destroyer hoped for. It’s great, but not something that will meet expectations of the delayed pro.
M3 has been for a long time where the convergence of all the good things was headed. It may mean Apple breaks a promise, but it’s better than releasing something prematurely just because an ambitious project didn’t work out in time.
The only way an M2 Ultra goes in a Mac Pro is if Apple developed an external-to-SOC traffic controller that mimics how their Fabric works - and then add multiple M2Ultra packages in a “modular” config.
We don't know. It's very much possible that an 'M3 extreme' is also uneconomical.
It is not that it is "possible." It is that it is the entire case. Threadripper, Epyc, Xeon, even ARM servers benefit from a combination of economies of scale and data center customers willing (needing) to pay whatever they cost. So it makes sense for the chipmakers (chip designers?) to pay the foundries to stand up the manufacturing lines. But it doesn't make economic sense for Apple to do the same for "Extreme" SOCs to be used in Mac Pros and iMac Pros that are only going to move maybe 10,000 units a year and still need to have a starting price of around $5000. This problem didn't exist with the Mac Pro because it used Xeon CPUs that Intel sells to everybody else, allowing them to be made and sold to Apple relatively cheaply.
The original 2020 M1 merely required adding 2 cores to what was basically a pre-existing smartphone chip at a manufacturer that had already been making octacore ARM chips for Android devices for years. By contrast M2 Extreme would have required horizontal combining 4 M2 chips to make by far the largest PC chip (in surface area) in history. Please note that Intel and AMD have abandoned the horizontal thing with 3-D scaling (vertical and horizontal) AND are atomizing the CPU parts into components - AMD calls them chiplets, Intel calls them tiles - because it is less expensive to manufacture and package in a motherboard. But even there, it is only financially feasible because Intel and AMD are going to sell a lot more Xeon and Epyc CPUs than Apple will Extremes.
I still maintain that Apple should go into the ARM server market. It would require them to emulate what Microsoft did with Windows Server and come out with a bona fide server version of macOS. Not only would they make a lot of money directly, but the byproduct would mean making workstation CPUs that would otherwise need to be manufactured in very small numbers for a few niche customers financially viable. So Apple, go ahead and make a competitor to Nvidia Grace https://www.nextplatform.com/2022/08/29/details-emerge-on-nvidias-grace-arm-cpu/ except limit it to small and medium sized servers!
I'm not sure an Apple server would succeed. They'll be competing with HP, Dell and other large server's OEM's that have been doing business with SMB and enterprises for years. Most of those business have devices, services and applications from different vendors, and Apple is terrible integrating with other vendors. Another thing Apple lacks is the ecosystem companies as Microsoft have. They also would be competing with public cloud providers, considering a lot of business and enterprises are moving some or all workloads to AWS, Azure / MS 365 and GCP / Google Workspace.
I don't think creating an additional niche product would work at all.
If absolutely nothing else, they're going to make a server for themselves. Right now, Xcode Cloud runs on amd64 processors. The VMs have four cores and 16 GB of RAM. A few WWDC 2021 sessions were filmed in front of multiple racks filled top-to-bottom with rackmount macpro7,1 (2019) units. They (and many, many more racks like them) are almost certainly being used for the Xcode Cloud VMs.
It's more than a little embarrassing for Apple's shiny new developer-focused cloud offering to use a processor architecture they barely even sell anymore. I would be fairly surprised if WWDC 2023 doesn't include an announcement that Xcode Cloud VMs are moving to aarch64 with amd64 as a non-default option.
Apple didn’t delay the Mac Pro two plus years in order to give us a Mac Studio with a different name.
The new Mac Pro won’t arrive until apple is ready to blow the doors off everything else - even if that means waiting for M3 Extreme.
The M2 just isn’t the destroyer hoped for. It’s great, but not something that will meet expectations of the delayed pro.
M3 has been for a long time where the convergence of all the good things was headed. It may mean Apple breaks a promise, but it’s better than releasing something prematurely just because an ambitious project didn’t work out in time.
The only way an M2 Ultra goes in a Mac Pro is if Apple developed an external-to-SOC traffic controller that mimics how their Fabric works - and then add multiple M2Ultra packages in a “modular” config.
We don't know. It's very much possible that an 'M3 extreme' is also uneconomical.
It is not that it is "possible." It is that it is the entire case. Threadripper, Epyc, Xeon, even ARM servers benefit from a combination of economies of scale and data center customers willing (needing) to pay whatever they cost. So it makes sense for the chipmakers (chip designers?) to pay the foundries to stand up the manufacturing lines. But it doesn't make economic sense for Apple to do the same for "Extreme" SOCs to be used in Mac Pros and iMac Pros that are only going to move maybe 10,000 units a year and still need to have a starting price of around $5000. This problem didn't exist with the Mac Pro because it used Xeon CPUs that Intel sells to everybody else, allowing them to be made and sold to Apple relatively cheaply.
The original 2020 M1 merely required adding 2 cores to what was basically a pre-existing smartphone chip at a manufacturer that had already been making octacore ARM chips for Android devices for years. By contrast M2 Extreme would have required horizontal combining 4 M2 chips to make by far the largest PC chip (in surface area) in history. Please note that Intel and AMD have abandoned the horizontal thing with 3-D scaling (vertical and horizontal) AND are atomizing the CPU parts into components - AMD calls them chiplets, Intel calls them tiles - because it is less expensive to manufacture and package in a motherboard. But even there, it is only financially feasible because Intel and AMD are going to sell a lot more Xeon and Epyc CPUs than Apple will Extremes.
I still maintain that Apple should go into the ARM server market. It would require them to emulate what Microsoft did with Windows Server and come out with a bona fide server version of macOS. Not only would they make a lot of money directly, but the byproduct would mean making workstation CPUs that would otherwise need to be manufactured in very small numbers for a few niche customers financially viable. So Apple, go ahead and make a competitor to Nvidia Grace https://www.nextplatform.com/2022/08/29/details-emerge-on-nvidias-grace-arm-cpu/ except limit it to small and medium sized servers!
I'm not sure an Apple server would succeed. They'll be competing with HP, Dell and other large server's OEM's that have been doing business with SMB and enterprises for years. Most of those business have devices, services and applications from different vendors, and Apple is terrible integrating with other vendors. Another thing Apple lacks is the ecosystem companies as Microsoft have. They also would be competing with public cloud providers, considering a lot of business and enterprises are moving some or all workloads to AWS, Azure / MS 365 and GCP / Google Workspace.
I don't think creating an additional niche product would work at all.
If absolutely nothing else, they're going to make a server for themselves. Right now, Xcode Cloud runs on amd64 processors. The VMs have four cores and 16 GB of RAM. A few WWDC 2021 sessions were filmed in front of multiple racks filled top-to-bottom with rackmount macpro7,1 (2019) units. They (and many, many more racks like them) are almost certainly being used for the Xcode Cloud VMs.
It's more than a little embarrassing for Apple's shiny new developer-focused cloud offering to use a processor architecture they barely even sell anymore. I would be fairly surprised if WWDC 2023 doesn't include an announcement that Xcode Cloud VMs are moving to aarch64 with amd64 as a non-default option.
Yup — I’ve always thought it could make sense for apple to make servers for their own use.
A rumored processor is rumored to be cancelled. While it's fun reading this kind of articles, they should taken for what it's worth. A rumor!
Yeah, kind of strange to see the angst registered against compounding rumors, especially when the definition of what "Extreme" really means is known to nobody outside of Apple, and possibly nobody inside of Apple.
I have no doubt that Apple will release an Apple Silicon based Mac Pro at some point. Why it's lagging behind the Apple Silicon transition target proposed by Tim Cook is anyone's guess. My guess is that unlike the already transitioned products, the Mac Pro has a much tinier but more functionally demanding customer base. The Apple Silicon based iMac, 24, Mac mini, MacBook Pro, and MacBook Air are incremental improvements over their Intel predecessors. The Mac Studio is a Maxi-Mini, but still something that Apple could have done in Intel Silicon with a more robust cooling strategy.
Yes, Apple Silicon brings some obvious benefits to all of these product but it also imposes few sacrifices. My biggest beef with Apple Silicon is the loss of first class virtualization support, i.e., super easy support of x86 VMs like you got with Intel chips. That's basically the only thing I miss. I'm sure that others feel that the lack of support for discrete GPUs, external GPUs, and multiple monitor support on some models are serious sacrifices. But still, the "cons" list is very short.
A Mac Pro with Apple Silicon really has to take everything up to a new level performance wise but they also have to be very careful not to impose too many sacrifices. The loss potential for Mac Pro is much higher than anything seen on the non-Pro machines. Apple can't deliver a compromise solution with a Pro label on it. They tried that once already and had to revert back to the same fundamentals that existed before they rolled out their New Coke Pro that looked like a fancy trash can. The relative ease that Apple experienced with transitioning the non-Pro Macs to Apple Silicon means nothing when it comes to the Pro. Many of these things have less to do with core counts and memory capabilities and more to do with other functional aspects of the Pro model. After all, Apple could (and probably will) beef up the Studio with newer & faster silicon, more cores, more memory, and more storage but it will still never be functionally equivalent to what the Pro has to be for the most demanding customers.
Comments
Yes there is real value to having the best hardware available to run proprietary Apple software. But 2019 Mac Pro performance will have people who use third party software jumping ship to 2024 AMD Threadripper and Xeon machines. Yes they will be at the mid-range price point but the 2019 performance means that you will need two of them.
Apple is going to have to face the same challenge that ARM server manufacturers currently have, which is to approximate Xeon and Epyc performance. Nvidia's ARM server CPU - arriving 1Q2023 - does this by starting at 72 cores.
So my guess. We'll probably see an upgraded (M3?) CPU that supports 'external' memory expansion and probably some smart way to handle that extended memory efficiently and at speed. Which probably needs changes to the silicon itself as well as to the kernel memory management. Kind of feels like the MS-DOS days with the 640kb limit and that highmem BS. B/c no matter what memory outside of the builtin SOC memory will be slower than SOC builtin memory.
Don't forget: there is a need on the market for these boxes, but it is comparatively very small market size. These people are willing to spend a ton of money on these boxes and need the high speed/bandwitch/extensability and have been poorly served for years by Apple in the last few years (very few Hw updates rather late in the market plus the Trashcan Mac debacle). Makes me wonder how many users still bet in Apple for these use cases.
Lot's of these very special applications are probably not (yet?) optimized for Apple Silicon (but rather for Intel CPUs). Now that Apple moved away from Intel the manufacturers need to put quite a bit of effort into Apple Silicon optimization. So that takes longer and/or might never happen, since Apple is still a rather niche market and most of thes Apps are running on Intel boxes as well. So the transition cost for SW-Manufacturers is high as well.
One reason why I still believe that the Mac Pro (even though it's not exactly cheap) is probably barely able to be cost effective for Apple to produce. This product category is (financially) probably more about bragging rights and less about actually earning real money. So low-priority update from Apple's point of view.
Last but not least: Remember what happened with the last transition? Apple announce the Intel transition is over even though there were no Apple XServer available with Intel CPUs (they just stopped selling servers for months) .. and in the end the XServe category died (in Hardware) and the Server Software as well in the end. Not sure if Apple intends to continue these high-end pro boxes much longer.
For example, I've been waiting for my dream Mac Pro for a long time. The 2019 version came close, but it was held back by the choice of a crappy Intel Xeon (I bought a Linux ThreadRipper system instead).
I'm getting my first Apple Silicon computer (an M2 MBA) this week, so I can test for myself how the M2 does with my particular workloads. If the performance is what I'm hoping it will be, then I could see buying four M2Pro Mac minis (if such a thing is ever released and if the price is reasonable). I know exactly how to use such a configuration to meet my needs.
Why M2 rather than a new ThreadRipper? Three reasons:
1. The price of Threadripper has gone through the roof
2. AMD is now lagging Threaderipper updates far behind the mainstream desktop or EPYC
3. performance/watt of x86 sucks
My 2013 MacPro was upgraded over time to 128GB of OWC Ram and a 2TB SSD from OWC as these became available There were also two Thuderboltv 27" displays on the wall. First one display died after nine plus years and there were upgrade issues for the MacPro requiring the OEM 1TB SSD be reinstalled so Apple could lock the firmware so no more upgrades could happen.
I ordered my Mac Studio with the 64 core GPU, 128GB shared memory and 8TB SSD along with two Apple Studio displays with VESA mounts that were to reuse the existing wall mounts. The $10,000 system was here and operational in September. I have no regrets as it does everything I want and will last me for many years (age 77).
I delayed ordering the system until the initial bugs were cured in the Studio Displays. There were no issues in upgrading the display MacOS software when they arrived. The Mac Studio arrived earlier than projected as production speeded up in China.
Now there are "spares" in the Apple system so I am no longer unsupported if there were a mechanical issue in the Mac Studio. The Studio sits on my desk below a Studio Display. There is no sound coming out of that enclosure and it barely gets above room temperature to the touch.
This is the best Apple computer system I have owned since 1990.
YMMV
The original 2020 M1 merely required adding 2 cores to what was basically a pre-existing smartphone chip at a manufacturer that had already been making octacore ARM chips for Android devices for years. By contrast M2 Extreme would have required horizontal combining 4 M2 chips to make by far the largest PC chip (in surface area) in history. Please note that Intel and AMD have abandoned the horizontal thing with 3-D scaling (vertical and horizontal) AND are atomizing the CPU parts into components - AMD calls them chiplets, Intel calls them tiles - because it is less expensive to manufacture and package in a motherboard. But even there, it is only financially feasible because Intel and AMD are going to sell a lot more Xeon and Epyc CPUs than Apple will Extremes.
I still maintain that Apple should go into the ARM server market. It would require them to emulate what Microsoft did with Windows Server and come out with a bona fide server version of macOS. Not only would they make a lot of money directly, but the byproduct would mean making workstation CPUs that would otherwise need to be manufactured in very small numbers for a few niche customers financially viable. So Apple, go ahead and make a competitor to Nvidia Grace https://www.nextplatform.com/2022/08/29/details-emerge-on-nvidias-grace-arm-cpu/ except limit it to small and medium sized servers!
If Apple wants to get back into the HEDT market, I think they either laser focus on specific segments(video, audio) and build chips around encoders that can handle massive workloads, or they have to move to a more modular design if they want to be more general purposes. As their chips grow larger and the manufacturing gets more expensive, each defect on the wafer will get more and more costly vs the chiplet/tile designs. That being said, Apple may have shown a bit of their hand with the silicon interposer in the M1 Ultra. If they can link two chips at such a high speed, whats to stop them from doing the same thing with modular chiplets in the future? That would allow them to scale without the issues of the monolithic chip designs. The other option would be to scale up as you mentioned and sell linux/unix based servers to operate in more efficient server stacks.
They should have shipped the M1 Ultra Mac Pro. Used the 2019 Mac Pro form factor. Added a PCIe controller. Kept the slots.
I don't think creating an additional niche product would work at all.
That is why Apple hasn't replaced it, and it is going on year 3 of their 2-year transition. They bit off more than they can chew.
It's more than a little embarrassing for Apple's shiny new developer-focused cloud offering to use a processor architecture they barely even sell anymore. I would be fairly surprised if WWDC 2023 doesn't include an announcement that Xcode Cloud VMs are moving to aarch64 with amd64 as a non-default option.
I have no doubt that Apple will release an Apple Silicon based Mac Pro at some point. Why it's lagging behind the Apple Silicon transition target proposed by Tim Cook is anyone's guess. My guess is that unlike the already transitioned products, the Mac Pro has a much tinier but more functionally demanding customer base. The Apple Silicon based iMac, 24, Mac mini, MacBook Pro, and MacBook Air are incremental improvements over their Intel predecessors. The Mac Studio is a Maxi-Mini, but still something that Apple could have done in Intel Silicon with a more robust cooling strategy.
Yes, Apple Silicon brings some obvious benefits to all of these product but it also imposes few sacrifices. My biggest beef with Apple Silicon is the loss of first class virtualization support, i.e., super easy support of x86 VMs like you got with Intel chips. That's basically the only thing I miss. I'm sure that others feel that the lack of support for discrete GPUs, external GPUs, and multiple monitor support on some models are serious sacrifices. But still, the "cons" list is very short.
A Mac Pro with Apple Silicon really has to take everything up to a new level performance wise but they also have to be very careful not to impose too many sacrifices. The loss potential for Mac Pro is much higher than anything seen on the non-Pro machines. Apple can't deliver a compromise solution with a Pro label on it. They tried that once already and had to revert back to the same fundamentals that existed before they rolled out their New Coke Pro that looked like a fancy trash can. The relative ease that Apple experienced with transitioning the non-Pro Macs to Apple Silicon means nothing when it comes to the Pro. Many of these things have less to do with core counts and memory capabilities and more to do with other functional aspects of the Pro model. After all, Apple could (and probably will) beef up the Studio with newer & faster silicon, more cores, more memory, and more storage but it will still never be functionally equivalent to what the Pro has to be for the most demanding customers.