Could you explain the statement "The reasons why for this are complex. It's not just about drivers --
Apple has decided that it didn't need a way for the Apple Silicon
processor to talk to an external graphics card, at all, under any
circumstances."? If my 6-year-old Intel iMac Pro can talk to a graphics card over Thunderbolt 3, why can't an Apple Silicon-based Mac do the same? I think it IS about drivers, and the fact that Apple doesn't want to work with either AMD or nVidia ever again. https://machow2.com/m1-mac-egpus/
My wild guess is that the GPU cannot access an external bus. If all the connections between the GPU, CPU and RAM are all on chip or substrate there is no reason to connect them to external pins. Finding a way to fan out all the signals to external pins may sound trivial but it is not so I doubt it those signals are run out to external pins if they don’t have to be. My guess is Apple doesn’t publish the pinouts of their custom chips but has anyone reversed engineered their functions yet?
I also don’t think 2023 Mac Pro will sell nearly as well as the 2019 Pro. I hope that dip in sales prompts them to double down on improving future Mac Pros, rather than abandoning it altogether.
The 2019 model didn't sell well either, the Mac Studio outsells it by at least 10:1. Apple can see what their sales are and how their customers use them. They have said that most Mac Pro buyers bought mid-range models like this one:
That was a 12-core Intel with Radeon 580X (5TFLOPs), $7k. A single Radeon W6900X (22TFLOPs) cost $5600.
By comparison, the M2 Ultra is faster than the 28-core Intel chip and faster (27TFLOPs, assuming both GPUs fully used) than the higher-end Radeon GPUs.
Apple's new Mac Pro is designed exactly for the majority of their target audience with M2 Ultra, 64GB RAM, 1TB SSD for $7k.
The PCIe slots are primarily used for networking, IO and processing AV streams, not additional GPUs.
There will be no noticeable sales dip with this model or sales jump with another model that does anything differently. This is complete fiction. The high performance computing pro market is a rounding error in the computer industry and has been for a long time.
The only market worth targeting is the enthusiast market that would today buy an i9-13900K + Nvidia 4090 for around $4k. The M2 Ultra is within 30% of the CPU and probably 1/3-1/4 the GPU (lower when using hardware raytracing) for $7k. All they need to improve is GPU performance to be competitive here and M3 Ultra next year will close that gap a bit and maybe offer an opportunity to make a higher-end chip like (M3 Max -> M3 Extreme GPU <- M3 Max = 100TFLOPs) but they'd charge an extra $3k+ for it.
The RAM limit will sort itself over time. That limit doubles every 3 years or so, one day Apple will offer 384GB, then 768GB.
Apple Silicon has been great for Apple, and the Mac faithful. It's got industry-leading computing grunt to power consumption, and it does it all quietly and in a very cool package. No more lap-singeing MacBook Pros!
It's restored pricing at the low-end that hasn't been seen in a long time. It's also broken the shackles of promises Intel made for years that led to engineering choices that couldn't handle the heat as well as they should have been able to when the chips that fell short of those lofty goals were delivered.
Honestly guys this isn't true. At all. The consensus - see Tom's Hardware and Wccftech is that the M2 Ultra's Geekbench scores are essentially the same as the Intel Core i9-13900K (Geekbench 5 has the i9 slightly higher, Geekbench 6 the Ultra slightly higher). Now the Core i9 is a chip primarily for desktop gamers. A hypothetical Mac Pro refresh would have had a Xeon W9-3495X that crushes the Core i9 and should have had an AMD Threadripper 7000 that when it launches in September crushes the Xeon W9. Claiming that an Intel Core i9 system with 256 GB RAM, 2 TB SSD and an Nvidia RTX A5000 (or AMD Radeon Pro W6800X) will set you back $10,000 just isn't being honest. Apple Silicon is only a better value than AMD systems in particular when you don't have the added expense of a discrete GPU, like the $599 Mac Mini whose integrated GPU is better than the entry level AMD, Nvidia and Intel discrete options. Otherwise x86 systems perform better and are cheaper.
When the M3 Extreme arrives next year, maybe. But an M2 Ultra workstation with CPU performance no better than a desktop PC chip and graphics performance akin to a midlevel pro dGPU? Not a chance.
The high performance computing pro market is a rounding error in the computer industry and has been for a long time.
Not even close to being true. The workstation, small server and HEDT markets are booming. There was actually a Threadripper shortage last year. Intel launched Sapphire Rapids (Xeon W-2400 and W-3400) this year to compete with Threadripper in the desktop workstation market. AMD is going to launch the Threadripper 7900X this year along with the usual Pro 7900WX in order to compete with Intel in HEDTs. I am still wondering what is keeping Nvidia from getting in on the act with a stripped down version of their Grace chips targeting this market.
Engineers, architects, researchers etc. have always needed workstations. Do you think that these entire professions stopped existing? Moved to the cloud? Or can get their work done on MacBooks now? And now they have company. Thanks to YouTube and all that, the number of people into serious video editing and computer animation has gone through the roof. Add to those the AI/ML boom the past few years and now the LLM types is going to mean still more. Yet rounding error you say.
Apple not being able to do the Extreme until TSMC's 2nd gen 3nm process is ready until 2024, which forced them to just stick an M2 Ultra in a cheese grater and call it a day because needing to move forward with Sonoma meant that they couldn't wait any longer, is no reason to just go and make up stuff OK?
With the introduction of Apple Vision Pro, the redesign on a lot of their in-house software and hardware across the Mac, iPhone, iPad, and many other things they’ve done in recent times with Apple Silicon, it’s obvious Apple is on a different path, and it doesn’t include Intel, AMD or Nvidia whatsoever, after all, could the Apple Vision Pro be made being tied to those three companies no it cannot.
It’s obvious? I think those who see that it’s obvious that Apple never intends to deliver solutions to those asking now for multi GPU support are to quick in doing presumptions.
When thinking about this stuff it is always important to look at the easiest explanations for things being the way they are.
Since the introduction of Apple Silicon much of the things are now part of chip design that formerly were not. Even if Apple Silicon did impress in lots of ways there still were parts that were not fully scaled out. Think of the very limited display support of M1. Think of the growing but still limited RAM support. Think of the number of Thunderbolt controllers. Think of the video encoding units on Chip on M1 vs M2.
I think expectations from most would have lead to an availability of M3 at WWDC 2023. we know now that this „next generation“ will come later and that we will first have a refined version of M2 chips now. Though there is limits to what Apple can put in this refinements… a bit more power using more cores, a bit more RAM and such things… though not so much completely new stuff.
M2 did not have any on-die capability to address dedicated off chip GPU address spaces from Apple Silicon CPU cores. This isn’t something to just add by refining M2.
Still Apple was late in giving its answer to an Apple Silicon Mac Pro. If it hadn’t come - most people would have speculated the death of the Mac Pro. Now that it’s there many people still speculate that it means the death of the Mac Pro because of the missing multi GPU support. This is from a thinking that Apple just needs to say »oh let’s „allow“ multiple GPUs« and then it is finished. It completely misses the fact that there is at least one further chip generation necessary to reach that point.
So speculation about the death of the Mac Pro after the new introduction of a new variant of that product is very very premature. It’s a widely nonsensical assumption. We could see this resolved with M3 - it’s not absolutely safe though, since we just don’t know which new core features will be in M3.
My wild guess is that the GPU cannot access an external bus. If all the connections between the GPU, CPU and RAM are all on chip or substrate there is no reason to connect them to external pins. Finding a way to fan out all the signals to external pins may sound trivial but it is not so I doubt it those signals are run out to external pins if they don’t have to be. My guess is Apple doesn’t publish the pinouts of their custom chips but has anyone reversed engineered their functions yet?
No not really. In simple terms: The problem here is that for off chip GPUs to work you have a way to shove things from CPU RAM to GPU RAM (and back). And it’s not only CPU RAM vs GPU RAM… it’s CPU RAM vs GPU1 RAM vs GPU 2 RAM vs …
This „addressing“ feature is something that needs stuff done in the Apple Silicon architecture (see some of the patents already mentioned here).
Since Apple is as silent as ever on this, we do not know when or even if those features will appear. We can already see that it was nothing to stuff into the already designed M2 line so we at least have to wait for M3 or maybe longer
If it solves this, this doesn’t mean that it just will be a feature to support AMD or Nvidia. It will be support for “GPU computation unit extensions”. They then might also start providing extension cards with compute unit chips proprietary to, but also perfectly optimized to the Apple silicon platform in hardware and software.
The high performance computing pro market is a rounding error in the computer industry and has been for a long time.
Not even close to being true. The workstation, small server and HEDT markets are booming. There was actually a Threadripper shortage last year. Intel launched Sapphire Rapids (Xeon W-2400 and W-3400) this year to compete with Threadripper in the desktop workstation market. AMD is going to launch the Threadripper 7900X this year along with the usual Pro 7900WX in order to compete with Intel in HEDTs. I am still wondering what is keeping Nvidia from getting in on the act with a stripped down version of their Grace chips targeting this market.
Engineers, architects, researchers etc. have always needed workstations. Do you think that these entire professions stopped existing? Moved to the cloud? Or can get their work done on MacBooks now? And now they have company. Thanks to YouTube and all that, the number of people into serious video editing and computer animation has gone through the roof. Add to those the AI/ML boom the past few years and now the LLM types is going to mean still more. Yet rounding error you say.
Apple not being able to do the Extreme until TSMC's 2nd gen 3nm process is ready until 2024, which forced them to just stick an M2 Ultra in a cheese grater and call it a day because needing to move forward with Sonoma meant that they couldn't wait any longer, is no reason to just go and make up stuff OK?
As I mentioned, it's worth targeting the HEDT enthusiast market (around $4k). The non-server market above this is a rounding error.
High-end Threadripper, Epyc and Xeons are mainly for servers. The Ultra chips are competitive with the cheaper desktop Threadrippers and i9s.
Many engineers, architects, researchers, Youtubers, artists have been able to migrate to Macbook Pros. The Max chips perform comparably to higher-end desktops and can outperform the 2019 Mac Pros.
The Ultra Mac Studio and Mac Pro are aimed at the market that buys single 4090 GPUs. The market above this ($5k+) is well below 1%. The HEDT/enthusiast market is around 5-10%.
The M2 Ultra falls short of a $4k i9-13900 + 4090 but it's at least competitively priced in the Mac Studio. Chasing after a minuscule portion of the high-end market isn't a high priority, it can easily wait for M3 for hardware RT and possibly some extra GPU chip.
Apple Silicon has been great for Apple, and the Mac faithful. It's got industry-leading computing grunt to power consumption, and it does it all quietly and in a very cool package. No more lap-singeing MacBook Pros!
It's restored pricing at the low-end that hasn't been seen in a long time. It's also broken the shackles of promises Intel made for years that led to engineering choices that couldn't handle the heat as well as they should have been able to when the chips that fell short of those lofty goals were delivered.
Honestly guys this isn't true. At all. The consensus - see Tom's Hardware and Wccftech is that the M2 Ultra's Geekbench scores are essentially the same as the Intel Core i9-13900K (Geekbench 5 has the i9 slightly higher, Geekbench 6 the Ultra slightly higher). Now the Core i9 is a chip primarily for desktop gamers. A hypothetical Mac Pro refresh would have had a Xeon W9-3495X that crushes the Core i9 and should have had an AMD Threadripper 7000 that when it launches in September crushes the Xeon W9. Claiming that an Intel Core i9 system with 256 GB RAM, 2 TB SSD and an Nvidia RTX A5000 (or AMD Radeon Pro W6800X) will set you back $10,000 just isn't being honest. Apple Silicon is only a better value than AMD systems in particular when you don't have the added expense of a discrete GPU, like the $599 Mac Mini whose integrated GPU is better than the entry level AMD, Nvidia and Intel discrete options. Otherwise x86 systems perform better and are cheaper.
When the M3 Extreme arrives next year, maybe. But an M2 Ultra workstation with CPU performance no better than a desktop PC chip and graphics performance akin to a midlevel pro dGPU? Not a chance.
I have no idea what your statement has to do with what you quoted out.
"In every case, the lack of PCI-E GPU support is the killing factor for the new Mac Pro"
You gave a music industry example. Except none of the DAWs use intense GPUs or process data on GPUs. Prior to the M series, most buying Macs leave the video alone and focus on the CPU as that was what bound the performance of DAWs. I can easily imagine folks getting the new one and filling it w/PCI-e cards for Pro Tools and such. Rack that and some interfaces and you have a very strong setup with plenty of extra expandability via Thunderbolt.
I just saw the M2 UItra Metal result on GB: 281,948, which IS better than I can get with my eGPU, which gives me 191,426 on GB 6.1. So I guess Apple has accomplished what I was hoping they would.
Three awful mistakes in a row - the wedge-less MBA, the monstrosity Studio and now the Mac Pro which alienates all its key markets. I was present for the second coming of Apple but I feel that golden era has sadly passed. I doubt the 'Vision Pro', with such limited use cases and such a high entry price will alter Apple's destiny.
As an investor, I would start buying up unopened Macs from a few years ago and putting them into storage. They will never be as good again.
The wedgeless MacBook Air is Apple's best selling laptop so that statement about it being a mistake is BS. Just because you don't like it doesn't mean the rest of the market doesn't. What is the monstrosity about the Mac Studio? How is it any different than the iMac Pro it replaces other than you don't get the 27" display attached to it which IMO is better anyways?
And nobody gives a rats ass about you being an investor. I cannot stand when people say this, or the fact that they've been an Apple fan for over 20-30yrs. It doesn't make your statement anymore true by saying shit like that.
They have an interesting challenge here. They are selling a computer with integrated high performance graphics, so if they enable customers to use external GPUs then the next complaint will be having to pay for the integrated graphics even though with external GPUs it’s not being used. A can’t-win situation. Then again, the performance gains for the integrated graphics come in part from unified memory and would be lost with external GPUs. Can’t win, again.
The interesting point to me is having so many hardware announcements for WWDC. Like they are clearing the decks for September. What comes then? They must have something else big to announce, not just an iPhone refresh.
Good report Mike. I also don’t think 2023 Mac Pro will sell nearly as well as the 2019 Pro. I hope that dip in sales prompts them to double down on improving future Mac Pros, rather than abandoning it altogether.
I’d guess that they would have liked to do more with AS Mac Pro, but for whatever reason were not able to get it done this year. Hopefully they will keep pushing forward. From taking a year longer than their 2 year transition goal, to failing to deliver the planned Jade 4C “Extreme” SoC in both M1 and M2 generations, seems like they really hit some roadblocks from their original intent. I highly doubt they would have given 2019 Mac Pro such dramatically better expansion capabilities if they knew they wouldn’t be able to maintain that level of performance after Apple Silicon transition.
One has to suspect the ongoing braindrain on the Apple Silicon design team is part of these stumbles. Let’s hope Srouji gets the big Mac back on track, and their patented multi-GPU tech comes to light soon!
Brain drain? I think whatever talent Apple had was working on that R1chip, and anything else to support the Apple Vision Pro, and anyone else left over we’re working on the iPhone chips.
Abandoning MPX isn't a negative. It's a good sign, looking ahead, not back. As I understand it (correct me if I'm wrong), MPX basically added Thunderbolt to PCIe. It was a solution Apple developed for a particular problem, but (again, as I understand it) it serves no purpose with Apple silicon and PCIe 4 and Thunderbolt 4.
Also, it's probably wrong to suggest the Pro Workflows group was not consulted. The video interview featuring Anand Shimpi a while back included a manager from that group. They were consulted, and they likely played a critical role in the decision to not build whatever it was (Shimpi made it very clear in that interview that *something* was aborted), because it didn't deliver a compelling product for exactly that Pro Workflows group.
The release of this Mac Pro now may well indicate that they have had success developing the next generation and (unlike the M1 and M2 generations) the decision has been made to build it (whatever it is). So it is past the point in the process where the earlier attempts were aborted. My own uneducated guess is that it's an Apple Silicon PCIe GPU/Unified Memory extension which doubles the GPU power of the Ultra, and not the rumored 4x "Extreme" design.
Looks like a mistake by Primate Labs. That score is more than 2x the score of the M2 Max. Unless there are a lot of Compute sub benches that are memory bottlenecked which the M2 Ultra overcomes, hard to believe that score.
Comments
https://everymac.com/systems/apple/mac_pro/specs/mac-pro-12-core-3.3-xeon-w-silver-tower-workstation-2019-specs.html
That was a 12-core Intel with Radeon 580X (5TFLOPs), $7k.
A single Radeon W6900X (22TFLOPs) cost $5600.
By comparison, the M2 Ultra is faster than the 28-core Intel chip and faster (27TFLOPs, assuming both GPUs fully used) than the higher-end Radeon GPUs.
Apple's new Mac Pro is designed exactly for the majority of their target audience with M2 Ultra, 64GB RAM, 1TB SSD for $7k.
The PCIe slots are primarily used for networking, IO and processing AV streams, not additional GPUs.
There will be no noticeable sales dip with this model or sales jump with another model that does anything differently. This is complete fiction. The high performance computing pro market is a rounding error in the computer industry and has been for a long time.
The only market worth targeting is the enthusiast market that would today buy an i9-13900K + Nvidia 4090 for around $4k. The M2 Ultra is within 30% of the CPU and probably 1/3-1/4 the GPU (lower when using hardware raytracing) for $7k. All they need to improve is GPU performance to be competitive here and M3 Ultra next year will close that gap a bit and maybe offer an opportunity to make a higher-end chip like (M3 Max -> M3 Extreme GPU <- M3 Max = 100TFLOPs) but they'd charge an extra $3k+ for it.
The RAM limit will sort itself over time. That limit doubles every 3 years or so, one day Apple will offer 384GB, then 768GB.
Apple Silicon is only a better value than AMD systems in particular when you don't have the added expense of a discrete GPU, like the $599 Mac Mini whose integrated GPU is better than the entry level AMD, Nvidia and Intel discrete options. Otherwise x86 systems perform better and are cheaper.
When the M3 Extreme arrives next year, maybe. But an M2 Ultra workstation with CPU performance no better than a desktop PC chip and graphics performance akin to a midlevel pro dGPU? Not a chance.
Engineers, architects, researchers etc. have always needed workstations. Do you think that these entire professions stopped existing? Moved to the cloud? Or can get their work done on MacBooks now? And now they have company. Thanks to YouTube and all that, the number of people into serious video editing and computer animation has gone through the roof. Add to those the AI/ML boom the past few years and now the LLM types is going to mean still more. Yet rounding error you say.
Apple not being able to do the Extreme until TSMC's 2nd gen 3nm process is ready until 2024, which forced them to just stick an M2 Ultra in a cheese grater and call it a day because needing to move forward with Sonoma meant that they couldn't wait any longer, is no reason to just go and make up stuff OK?
When thinking about this stuff it is always important to look at the easiest explanations for things being the way they are.
Since the introduction of Apple Silicon much of the things are now part of chip design that formerly were not. Even if Apple Silicon did impress in lots of ways there still were parts that were not fully scaled out. Think of the very limited display support of M1. Think of the growing but still limited RAM support. Think of the number of Thunderbolt controllers. Think of the video encoding units on Chip on M1 vs M2.
M2 did not have any on-die capability to address dedicated off chip GPU address spaces from Apple Silicon CPU cores. This isn’t something to just add by refining M2.
Still Apple was late in giving its answer to an Apple Silicon Mac Pro. If it hadn’t come - most people would have speculated the death of the Mac Pro. Now that it’s there many people still speculate that it means the death of the Mac Pro because of the missing multi GPU support. This is from a thinking that Apple just needs to say »oh let’s „allow“ multiple GPUs« and then it is finished. It completely misses the fact that there is at least one further chip generation necessary to reach that point.
So speculation about the death of the Mac Pro after the new introduction of a new variant of that product is very very premature. It’s a widely nonsensical assumption. We could see this resolved with M3 - it’s not absolutely safe though, since we just don’t know which new core features will be in M3.
Since Apple is as silent as ever on this, we do not know when or even if those features will appear. We can already see that it was nothing to stuff into the already designed M2 line so we at least have to wait for M3 or maybe longer
If it solves this, this doesn’t mean that it just will be a feature to support AMD or Nvidia. It will be support for “GPU computation unit extensions”. They then might also start providing extension cards with compute unit chips proprietary to, but also perfectly optimized to the Apple silicon platform in hardware and software.
High-end Threadripper, Epyc and Xeons are mainly for servers. The Ultra chips are competitive with the cheaper desktop Threadrippers and i9s.
Many engineers, architects, researchers, Youtubers, artists have been able to migrate to Macbook Pros. The Max chips perform comparably to higher-end desktops and can outperform the 2019 Mac Pros.
https://www.tomshardware.com/news/nvidia-maintains-lead-as-sales-of-graphics-cards-hit-all-time-low-in-2022-jpr
Notebooks continue to erode desktop relevance. Lowest overall sales in over 10 years and falling. Nvidia's growth is in the server space with AI.
According to this, Nvidia sold about 30 million desktop GPUs in 2022. Notebook GPUs would be at least as many and their GPU revenue was $12b.
$12b / 60m units = $200 ASP.
High-end GPUs like the 4090 ($1600+) sell around 1m units per year. The majority of GPUs sold are consumer gaming cards:
https://store.steampowered.com/hwsurvey/videocard/
The Ultra Mac Studio and Mac Pro are aimed at the market that buys single 4090 GPUs. The market above this ($5k+) is well below 1%. The HEDT/enthusiast market is around 5-10%.
The M2 Ultra falls short of a $4k i9-13900 + 4090 but it's at least competitively priced in the Mac Studio. Chasing after a minuscule portion of the high-end market isn't a high priority, it can easily wait for M3 for hardware RT and possibly some extra GPU chip.
You gave a music industry example. Except none of the DAWs use intense GPUs or process data on GPUs. Prior to the M series, most buying Macs leave the video alone and focus on the CPU as that was what bound the performance of DAWs. I can easily imagine folks getting the new one and filling it w/PCI-e cards for Pro Tools and such. Rack that and some interfaces and you have a very strong setup with plenty of extra expandability via Thunderbolt.
And nobody gives a rats ass about you being an investor. I cannot stand when people say this, or the fact that they've been an Apple fan for over 20-30yrs. It doesn't make your statement anymore true by saying shit like that.
https://browser.geekbench.com/metal-benchmarks?mibextid=Zxz2cZ
Either future versions get the following or you can see it die out, much like the un-upgradeable trash can Mac Pro.
allow RAM upgrade, perhaps going to Apple Store to do it.
upgradeable CPU slots, perhaps space for 3 allowing at least the next generation or two to replace existing SoC
Also, it's probably wrong to suggest the Pro Workflows group was not consulted. The video interview featuring Anand Shimpi a while back included a manager from that group. They were consulted, and they likely played a critical role in the decision to not build whatever it was (Shimpi made it very clear in that interview that *something* was aborted), because it didn't deliver a compelling product for exactly that Pro Workflows group.
The release of this Mac Pro now may well indicate that they have had success developing the next generation and (unlike the M1 and M2 generations) the decision has been made to build it (whatever it is). So it is past the point in the process where the earlier attempts were aborted. My own uneducated guess is that it's an Apple Silicon PCIe GPU/Unified Memory extension which doubles the GPU power of the Ultra, and not the rumored 4x "Extreme" design.
These folks? https://youtu.be/VA4e0NqyYMw