Yeah but what else could they say really…. I have no doubt they have workshopped chip designs with zero (or few) gpu cores so they can use dedicated gpus in the Mac Pro - either instead of or along with the integrated ones. For all we know they have their own AS-based gpus on the table with all (or mostly) gpu cores. considering the kind of work (and money) that goes into these things and what they are able to charge for them at the end we might all be way off the mark with our low expectations of what they can achieve on a AS Mac Pro. As far as RAM goes they could just make it moot by shoving so much in there that no one would ever complain.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP All Models Intel may be the only way to save the Mac Pro
- Thread starter Canubis
- Start date
- Sort by reaction score
Yeah but what else could they say really…. I have no doubt they have workshopped chip designs with zero (or few) gpu cores so they can use dedicated gpus in the Mac Pro - either instead of or along with the integrated ones. For all we know they have their own AS-based gpus on the table with all (or mostly) gpu cores. considering the kind of work (and money) that goes into these things and what they are able to charge for them at the end we might all be way off the mark with our low expectations of what they can achieve on a AS Mac Pro. As far as RAM goes they could just make it moot by shoving so much in there that no one would ever complain.
Spec prediction:
up to 4 user replaceable M2 Ultra modules with up to 192GB RAM (768GB total), 8TB SSD (32TB total), 4 TB5 ports (16 total)
4 PCIe Gen 5 x 16 slots
2 10G ethernet
BT 5.3
WiFi 6e
up to 4 user replaceable M2 Ultra modules with up to 192GB RAM (768GB total), 8TB SSD (32TB total), 4 TB5 ports (16 total)
4 PCIe Gen 5 x 16 slots
2 10G ethernet
BT 5.3
WiFi 6e
The question is how far can Apple Silicon scale up and be competitive in the computer market? Apple's SoCs have proven competitive in smartphones, smartwatches, laptops, and small desktops. But will the chip Apple installs in the new Mac Pro be competitive in the workstation market?
Mac Pro thread. I read that as Intel Mac Pro. Sorry.
After Apple introduced the 2020 27" iMac they sold over 100's of thousands of Intel Macs. Note that 'Intel Macs' is just as plural as Tim Cook's statement.
It is an odd 'stand on your head and squint' reading of cooks Statement that there was going to some broad product line of Intel Mac models (note models is plural) coming. He has sjust made a statement that the whole line up is transitioning.
At that point in time, the iMac Pro was 2017 stale (about 3 years) and 'plain' iMac 27" had no T2 chip (also a substantiive stale design that was way overdue. ) . Mac Pro 2019 had a T2 chip and , for the Mac Pro, relatively been recently refreshed (versus 3 years for 2010-2013 and 6 years 2013 - 2019 ). Apple had over a decade of not going rapid or regular Mac Pro updates as a track record.
Jumping up and hyper excited about the Ice Lake W-3300 ... there is no technical foundation to be excited about that at all given Intel's execution level in the 1H of 2020. (or 2019 ) .
My question is simple:
Why would anyone even purchase a Mac Pro, even if it has Apple Silicone?
The only people I can see who are interested in a Mac Pro are hobbyists who still live in the mid 1990's and who want to make the computer last as long as possible.
For businesses, computers are a deductible write off.
Why would anyone even purchase a Mac Pro, even if it has Apple Silicone?
The only people I can see who are interested in a Mac Pro are hobbyists who still live in the mid 1990's and who want to make the computer last as long as possible.
For businesses, computers are a deductible write off.
The W-3400 set at "Max power consumption, max overclocker" mode does. If set the power profile to 'balanced' the idle is 42W.
Pretty good chance macOS and Mac firmware isn't going to have some "Super overclocker" setting.
The W-2400 doesn't has this problem at the same extreme limits, doesn't blow out macOS's 64 thread limit, and pretty much matches up to the PCI-e lane provisioning of the current W-3200. And more affordable ( because a more 'value packed' Mac Pro at the current price levels would help. )
People keeping pointing at making the most expensive Mac Pro possible using Intel parts (and max expensive Nvidia cards ). A Mac Pro that is some kind of HP Z8 Fury 'killer' . The HP z4 is likely a bigger threat to the Mac Pro line up than the Z8 Fury is.
If look at the Puget charts it is pretty clear that the 3400 (and likely 2400 which they haven't done yet) are better than the W-3300 ( Ice Lake ) stuff. So these aren't 'horrible'. There are a number of 'issues' fixed relative to the what they tried to deliver in late 2020 ( and dribbled out in tame volumes in 2021 and very few major systems vendors have touched.). Intel has to start somewhat. It would be a substantive uplift over what the MP 2019 is running at the foundational level.
The bigger problem with these are that they are 'late' rather than 'horrible'. For Apple these are probably 1-1.5 years too late to be useful. Intel is making progress against AMD. AMD has a 7000 Threadripper 'counterpunch' queued up for later in 2023 , but aren't going to pragmatically ignore it like they did with the W-3300 (waited until Apple introduced to the M1 Ultra to drop the 'hammer' on that one. Intel didn't really matter at all in 2021. )
Its not a overlocker mode. It is "High Performance mode" which can be found in Windows Settings.
A rumoured(if it exists) 32 P core M2 can match a 56 core Xeon in CPU performance and trounce it in ST.
At this point I changed my mind about wanting an Intel Mac Pro, in applications like Blender and Vray the new Xeons gets beat out by 2 year AMD TR Pro.
These AMD/Intel systems going to do ProRes RAW faster than M2/M3 'Ultra' class count of embedded de/ecode fixed function logic? Probably not.
Heavy use of apps which use Apple Numeric acceleration libraries that effectively use the NPU , Apple AMX , and fixed function accelerators present on Apple Silicon also have heavy traction.
Heavy use of several high cost ( and sunk cost) A/V PCI-e cards that are already in a Intel Mac Pro and the drivers are available. ( there are already 50+ PCI-e cards that would work in a Mac Pro. It is not like there is nothing to put inside.)
That is mainly the folks still buying and using Mac Pro 2009-2012 models over the last 4-6 years. That really isn't the folks who bought the Mac Pro 2019's. The MP 2019 increased the entry price by 100%. Lots of the hobbyist, 'tinker on the most affordable', 'most time stretched budget possible' folks got left behind in 2019.
That never was Apple's primary target market either in 2009-2013 era or in the 2019-2021 one. There are a set of folks who piggybacked on the Mac Pro product ( especially in the used , 2nd/3rd/4th generation owner market) , but that isn't who Apple was primarily trying to sell new systems to.
You can only deduct if have substantive revenue. Some folks have businesses with either thinner margins or higher cash holding / self insurance requirements.
I guess never say never- but there is basically no way Apple would introduce a new intel based Mac Pro in the near future. They have made it pretty clear that they are planning to transition their entire Mac lineup to Apple Silicon, and there are huge disincentives to continuing to offer x86 systems- from software/OS/dev resources, hardware costs (profits), and even "about face"- they need to complete the transition. I would also argue that, while there may be some exceptions, as a Mac Pro user amongst many Mac Pro users, most of us are very eager for the Apple Silicon versions.
When the transition was first announced and some intel machines were said to be in the pipe, it left more room for speculation that Apple might keep an intel Mac Pro in the line-up- (or at least offer an interim update of one) -but over the last couple of years or so, the transition for most (even pro) software to support Apple Silicon has been pretty quick and smooth. There are still a certain number of folks tied to intel either because of bootcamp or some very specific workflows, but that's not a lot of users and a lot of those people are stuck with legacy software and can continue to use "legacy" intel Mac pros.
The much bigger issue IMO is the apparent cancellation of the "Extreme" variant of the chip. This was the only chip option that would seem to really set the Mac Pro apart from other Macs, and justify a $10k+ starting price for users that needed/wanted that level of performance. I really hope those rumors are incorrect, but it may be that the new Mac Pro is not much different in raw performance from the Mac Studio. Apple may indeed "artificially" hold back the Studio by not updating it with the M2 series, which will help differentiate the Mac Pro- but we all know that M2 is not /that/ much faster than M1.
A lot of folks have also bemoaned the lack of memory expansion. I just dont think that was ever in the cards for an M-series based machine. The on-package memory is a key component of the architecture and a lot of what makes these chips so much faster and more efficient. There is/was the possibility of adding a "Second Level" memory that could be user upgraded- basically a much larger, expandable amount of RAM that worked at a much lower speed- but the amount of hardware and software complexity this would add- the likely large performance inconsistency many apps would encounter when making memory requests, and other technical drawbacks would make this an unattractive option. Rather I expected the M2 extreme would be made available with some even higher capacity RAM options- so while not user upgradable, you could buy a system with say, 512GB of RAM built on package- for a very, very high cost- but that would cover the bases of 99.x percentage of "pro" users.
That is not to say the current rumor version of the Mac Pro isn't a bit disappointing- without the Extreme chip option, it wont be the blow out performance machine many of us were hoping for. And if the design is really almost completely unchanged, it would be a rather bulky system for the hardware it contains. But one shouldn't under-estimate the utility of having a bunch of internal PCIe bandwidth. For many Pros- those slots and bandwidth, along with the plethora of thunderbolt buses, was the most important aspect of the intel Mac Pro (and what we hated about the trashcan).
While GPU upgrades seem to get the most attention, GPU and storage are far from the only thing we add to machines like the Mac Pro- all sorts of add-in cards are important in pro workflows. And I am also not convinced Apple wont give us some sort of GPU card options- they could do this in a couple ways-
1) They could re-start support for AMD (or in theory even nVIDIA/intel) GPUs- its really a simple matter of partnering up and making drivers available- there is nothing inherent in Apple Silicon that breaks third party GPU support- Apple just needs to allow it and there need to be corresponding drivers.
2) They could make their own GPU MPX modules- basically creating large core count versions of the GPUs they use on M-series chips, with their own integrated VRAM, etc. Somewhat like the "upgradable RAM dilemma", these GPUs would not have the exact same performance characteristics as the on-die GPU has- but it could certainly be useful in workflows that are highly parallel and GPU/metal dependent - which describes a lot of pro workflows.
Even though I think its kind of pointless to riff on a theoretical new intel Mac Pro- using the w2400 series as the article postulates makes no sense at all- the maximum core count would be lower than the current intel Mac Pro options, and while the new CPU cores are a good deal faster now- thats not enough. The internals of the chassis would likely need to be modified a little for the higher TDP of the W3400 series and high core count CPU options- but this is not the trashcan, and there is plenty of room to improve on an already fantastic thermal management system in the current Mac Pro. That said, a theoretical M2 Extreme would probably fit easily into the same thermal management system that the current Mac Pro uses, without the need to modify it in any significant way.
The bottom line is, without the M2 Extreme option, the theoretical 56 core intel XEON based Mac Pro would almost certainly be significantly faster in some/most workloads than the M2 Ultra Mac Pro will be. And that sucks. But another intel Mac Pro is not coming- and M2 Ultra will at least be faster in most workloads than a 24 core modern XEON. It was always clear that building an Apple Silicon chip for the Mac Pro was going to be the biggest challenge in Apple's Mac line-up. Remember these chip designs started out in phones and iPads, before getting sophisticated and enhanced enough for use in modern PCs. The Apple Silicon focus on energy efficiency and integration were paramount design accomplishments for mobile devices- and frankly its super impressive how well Apple has been able to scale this chip design into the performance desktop space- But energy efficiency is of much lesser concern on a machine like the Mac Pro - and the M-series chips have historically lacked other important features of a workstation chip- such as massive amounts of I/O and PCIe connectivity. Add to this the fact that these will probably be relatively minuscule sales volume parts compared to the M-series chips that now populate the entire Mac (and iPad) line- and the high cost of developing and manufacturing such chips- it was always clear the Mac Pro would be the last upgrade to arrive at Apple Silicon. And it looks like Apple just hasn't gotten their designs all the way to where they should be for a high cost, high performance desktop workstation- at least if the M2 Extreme cancellation rumor is true. But it wont suck, and Apple could still bridge this gap with a future M-series "Extreme" chip or some other branch off their designs that is optimized more for desktops, especially with intel now much closer on their heels with new chips, and serious core counts.
g\
When the transition was first announced and some intel machines were said to be in the pipe, it left more room for speculation that Apple might keep an intel Mac Pro in the line-up- (or at least offer an interim update of one) -but over the last couple of years or so, the transition for most (even pro) software to support Apple Silicon has been pretty quick and smooth. There are still a certain number of folks tied to intel either because of bootcamp or some very specific workflows, but that's not a lot of users and a lot of those people are stuck with legacy software and can continue to use "legacy" intel Mac pros.
The much bigger issue IMO is the apparent cancellation of the "Extreme" variant of the chip. This was the only chip option that would seem to really set the Mac Pro apart from other Macs, and justify a $10k+ starting price for users that needed/wanted that level of performance. I really hope those rumors are incorrect, but it may be that the new Mac Pro is not much different in raw performance from the Mac Studio. Apple may indeed "artificially" hold back the Studio by not updating it with the M2 series, which will help differentiate the Mac Pro- but we all know that M2 is not /that/ much faster than M1.
A lot of folks have also bemoaned the lack of memory expansion. I just dont think that was ever in the cards for an M-series based machine. The on-package memory is a key component of the architecture and a lot of what makes these chips so much faster and more efficient. There is/was the possibility of adding a "Second Level" memory that could be user upgraded- basically a much larger, expandable amount of RAM that worked at a much lower speed- but the amount of hardware and software complexity this would add- the likely large performance inconsistency many apps would encounter when making memory requests, and other technical drawbacks would make this an unattractive option. Rather I expected the M2 extreme would be made available with some even higher capacity RAM options- so while not user upgradable, you could buy a system with say, 512GB of RAM built on package- for a very, very high cost- but that would cover the bases of 99.x percentage of "pro" users.
That is not to say the current rumor version of the Mac Pro isn't a bit disappointing- without the Extreme chip option, it wont be the blow out performance machine many of us were hoping for. And if the design is really almost completely unchanged, it would be a rather bulky system for the hardware it contains. But one shouldn't under-estimate the utility of having a bunch of internal PCIe bandwidth. For many Pros- those slots and bandwidth, along with the plethora of thunderbolt buses, was the most important aspect of the intel Mac Pro (and what we hated about the trashcan).
While GPU upgrades seem to get the most attention, GPU and storage are far from the only thing we add to machines like the Mac Pro- all sorts of add-in cards are important in pro workflows. And I am also not convinced Apple wont give us some sort of GPU card options- they could do this in a couple ways-
1) They could re-start support for AMD (or in theory even nVIDIA/intel) GPUs- its really a simple matter of partnering up and making drivers available- there is nothing inherent in Apple Silicon that breaks third party GPU support- Apple just needs to allow it and there need to be corresponding drivers.
2) They could make their own GPU MPX modules- basically creating large core count versions of the GPUs they use on M-series chips, with their own integrated VRAM, etc. Somewhat like the "upgradable RAM dilemma", these GPUs would not have the exact same performance characteristics as the on-die GPU has- but it could certainly be useful in workflows that are highly parallel and GPU/metal dependent - which describes a lot of pro workflows.
Even though I think its kind of pointless to riff on a theoretical new intel Mac Pro- using the w2400 series as the article postulates makes no sense at all- the maximum core count would be lower than the current intel Mac Pro options, and while the new CPU cores are a good deal faster now- thats not enough. The internals of the chassis would likely need to be modified a little for the higher TDP of the W3400 series and high core count CPU options- but this is not the trashcan, and there is plenty of room to improve on an already fantastic thermal management system in the current Mac Pro. That said, a theoretical M2 Extreme would probably fit easily into the same thermal management system that the current Mac Pro uses, without the need to modify it in any significant way.
The bottom line is, without the M2 Extreme option, the theoretical 56 core intel XEON based Mac Pro would almost certainly be significantly faster in some/most workloads than the M2 Ultra Mac Pro will be. And that sucks. But another intel Mac Pro is not coming- and M2 Ultra will at least be faster in most workloads than a 24 core modern XEON. It was always clear that building an Apple Silicon chip for the Mac Pro was going to be the biggest challenge in Apple's Mac line-up. Remember these chip designs started out in phones and iPads, before getting sophisticated and enhanced enough for use in modern PCs. The Apple Silicon focus on energy efficiency and integration were paramount design accomplishments for mobile devices- and frankly its super impressive how well Apple has been able to scale this chip design into the performance desktop space- But energy efficiency is of much lesser concern on a machine like the Mac Pro - and the M-series chips have historically lacked other important features of a workstation chip- such as massive amounts of I/O and PCIe connectivity. Add to this the fact that these will probably be relatively minuscule sales volume parts compared to the M-series chips that now populate the entire Mac (and iPad) line- and the high cost of developing and manufacturing such chips- it was always clear the Mac Pro would be the last upgrade to arrive at Apple Silicon. And it looks like Apple just hasn't gotten their designs all the way to where they should be for a high cost, high performance desktop workstation- at least if the M2 Extreme cancellation rumor is true. But it wont suck, and Apple could still bridge this gap with a future M-series "Extreme" chip or some other branch off their designs that is optimized more for desktops, especially with intel now much closer on their heels with new chips, and serious core counts.
g\
Last edited:
It is overclocking . just in the standard controlled way by Windows you get out the box. The fact that you can set it "don't gratitiously waste power mode" . Intel Turbo and Max Turbo 3.0 mode etc. That is all putting the silicon into a overclocking mode. The silicon isn't running at the natural mode the fab process supports. Intel just sell that as a feature ( since about a fab node (or more) behind in process being used to implement).
Yeah there is an even more 'ludicrous' mode version of overlocking.
That they can get back ~100W just idling is systemic that they are chasing the fringes of what the fab process supports. idle should be just that idling (doing almost nothing). A reasonable power management system should be kicking in turn stuff off that you are not using anyway. They are throwing power away just to hyper minimize the 'spin up' latency time out of idle. It is more a play to maximize some geek tech porn, short burst, single thread benchmark than be generally useful.
But at what price point? If the Mn "Extreme" costs 2-3x more than the W-3400 are folks really going to be 'happy' with that?
If Apple hasn't made a major adjustment to their chiplet approach that 'Extreme" isn't coming any time soon.
The TR5000 is as old as the M1 Ultra. (got announced at about the same day) It isn't two years old.
March 8, 2022
March 8, 2022
Apple unveils all-new Mac Studio and Studio Display
Apple introduced Mac Studio and Studio Display, a desktop and display designed to give users everything they need to build the studio of their dreams.
www.apple.com
AMD squatted on the TR5000 until the M1 Ultra came . Intel didn't matter as much.
P.S. TR5000 didn't ship in decent real volume until about 1 or 2 quarters later. ( primarily because it makes more sense to use those same base dies to take server market share than improve opportunities in the workstation market. TR7000 same thing. Better long term economies to do it later given limited supply. )
Last edited:
The problem is that the workstation market really isn't a 'scale up' market. It is a smaller market. The more expensive the smaller it gets. Once the number of units sold gets 'too small' it isn't going to be economically viable for Apple.
Apple only has about 10% of the general PC market. Mac sell around 24-30M a year. If the super high end workstation market is just 0.8 of the overall market and Apple takes 10% of that... do you even have a whole number left?
If the price of development the chip is too high for the volume you'll sell then generally don't get the chip made.
In the x86 high end workstation market they really are not grounding the prmarily volume source for the package on the workstation market. It is a re-tweaked server package with a different label.
Is Apple going to be able to build every SoC for everybody? No.
That said, the CPU / GPU / NPU core counts of the M 2 Max are tractable to multiply by 3 or 4 without going crazy on costs. They need a far better disagregation strategy than the relatively very primitive and chunky one they used in the M1 generation.
Is that going to cover the extreme upper ends of the workstation market? No. Do they need to cover those? Not really ( again the more extreme they go the smaller the market gets and there are no other systems to put these increasing niche SoCs into to generate more volume. )
I agree with everything you say except for this part. In no way does UMA require soldered RAM chips. That's completely orthogonal.
Not everything is about compute. The Xeon W2400 series has 64 PCIe gen. 5 lanes, the W3400 even has 112. That's a lot of expandability.
UMA is about a single memory space for CPU and GPU that allows pointer-passing instead of copying data. It doesn't require soldered RAM chips.
Correct. And if that is the case, that in the foreseable future, there wont be a Mac with card slost, than they can simply rename Mac Studio to Mac Pro (MP) or they can cancel the whole Mac Pro line altogether and just keep the Studio.
Although I admit that it would be interesting to see the expressions on the faces of the existing MP owners, that there will be no continuation.
Apple's chips are clearly designed for mobile products first. The Mac Pro in general goes against the "AS philosophy". I hope we don't have to wait too much longer for Apple to release something. At least we will be able to discuss an actual product on its merits instead of speculating wildly to no gain.
It isn't just x86 binaries. It old UEFI firmware. An aging T2 and its firmware. The subset of x86 binaries that Rosetta 2 also uses probably are not that much of big 'pain'. Apple doesn't support software APIs forever (macOS 9 code isn't running without substantive work arounds now either).
The problem more so is the huge imbalance of new Intel systems sold. The major volume got covered in year 1. The Mac Pro all by itself is problematic because the volume is so low. When the transition started there were 100M Intel Macs. That isn't going to disappear in two or three years.
The problem with a new Intel Mac Pro is that it kicks the can down the road on the support window with no way of replacement of Intel Macs back into the pool faster than old Intel Macs are spilling onto the Vintage/Obsolete list.
As the pool shrinks too small it isn't just Apple's 'pain'. Developers aren't going to want to put major money into vastly shrinking pool of users also.
More so on 'clickbait-ium'. When there is a subgroup of pissed off Mac users write an article that subgroup can commiserate around and it typically generates a substantive number of clicks. Art of writing what people want to hear so happier in their chosen social media silo.
Last edited:
And there is nothing that would prevent Apple from integrating more PCIe lanes into their desktop products. But then again, why do they need so many? Main consumer of PCIe lanes is the GPU, which is irrelevant for Apple Silicon. Other devices don’t need nearly as much PCIe bandwidth.
There have been some claims that the Mac Pro will be based on CXL. In that case I can see something like 128 PCIe 5 lanes from a separate controller (located on a logic board) and everything else (including the SoC) sitting on an MPX module. Nobody says that the PCI lanes need to come off the SoC…
Yeah just like Jobs said that they had a lot of exciting PowerPC products coming in 2005... come on
I am certain they did plan to update the Mac Pro with new Xeons, but those got delayed by a couple of years... Intel only managed to get Sapphire Rapids out literally couple of days before and they make somewhat pitiful figure next to AMD's offerings. Would have been a great CPU in 2021, not so much in 2023.
Any yet the sever systems that these very same CPU packages go into with a different label relatively rarely have buckets of GPUs in them.
If sitting on a household 500Mb/s Internet maybe. but high speed NAS/SAN network controller . 4 M.2 PCi-e v5 SSD adapter add-in card.
AFterburner ( uncompress video. ) [ Some other card could do something besides only ProRes. Apple is dragging their feet on AV1 (although beta Safari updates appears that maybe over on next iteration. ) )
The point is not support to cut the Mac Pro off from some future accelletor that is just an Apple GPU.
112 lanes of aggregate bandwidth. Nah...Apple isn't going to match that toe-to-toe. Because of the iGPU they have to sped a distortional die edge space on memory.
64 lanes aggregate. Probably not. Decent chance Apple is going to lean on PCI-e v4/5 2x faster to do fewer lane aggregation to go sideways on bandwidth.
If want to hit a high level of Perf/Watt you do. Same with lower PCI-e switching latencies.
Apple use a very large PCI-e PLEX switch on the Mac Pro main logic board. The MP 2019 already does that. If they updated that two input switch to PCI-e v4/5 (from v3) that would not be a shocker at all. But that won't be 128 lanes of aggregate bandwidth.
I'm kind of skeptical about PCI-e v5 (and CXL ) though. At least for 2023. If they can't do DPv2.1 or AV1 in a timely fashion in 2022 , highly doubtful they aren't not going to get caught flat-footed on the uptake of CXL also. (Especially if following the "no 3rd party GPUs so bandwidth doesn't matter" meme. )