Seems like today's rumors are reinforcing the idea that Apple is going to go to AR/VR with a future M chip in a tetherless headset, so it does sound like Apple will be going all in on its GPU architecture. That means that Apple has 10 years to shrink today's workstation-class dual 8K 120Hz 3D graphics down to something that can run portably. The Pro line becomes the vehicle for funding that R&D but, given the end goal of what will be essentially a wearable Pro chipset, we'll see power consumption playing a much bigger role than a desktop-only strategy.
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 7,1 Mac Pro Apple Silicon is going to eat our lunch
- Thread starter rondocap
- Start date
- Sort by reaction score
On the other hand, Apple's CPU for their autonomous car is said to run very hot ... I think it will be "horses for courses". If Apple want to boost the desktop performance of today's M Pro Max processor, then besides having more than one, the other alternative is to increases its clock rate which increases its thermal load. Cheap and simple. And a lot more power efficient than using 3rd party GPUs as well.
Does Apple enjoy a distinctive advantage of its GPU over Nvidia/AMD's "state of the art"?
I don't believe so. Instruction sets for GPUs are private and vendors could change at will from one generation to the other to suit their goals. Hence, unlike CPUs, they aren't bounded by complexity imposed by x86-64. I very much doubt that once Apple scales up GPU clocks, they'll still enjoy power efficiency bragging right except they have process node manufacturing advantage over, say, AMD's current "state of the art".
If we do a quick & dirty comparison between RX6900XT (AMD) and M1 Max 36C GPU (Apple), logically they're surprisingly similar to me:
ALUs: 5120 (AMD) vs 4096 (Apple)
Texture units: 320 (AMD) vs 256 (Apple)
ROPs: 128 (AMD) vs 128 (Apple)
Peak clock: 2250MHz (AMD) vs 1296MHz (Apple)
Memory bandwidth: 512GB/s (AMD) vs 400GB/s (Apple)
Power: 300W (AMD) vs 60W? (Apple)
Process: TSMC 7nm (AMD) vs TSMC 5nm (Apple)
Apple doesn't seem to me having a distinctive advantage over AMD/Nvidia. We could predict how much higher Apple could clock their current GPU without destroying their marketing rhetoric of power efficiency.
I also believe the integrated design (CPU+GPU) is a dis-advantage here. It limits how high Apple could push its GPU clocks. Higher freq > more heat > heat spill into CPU clusters. That might result in degraded CPU performance in concurrent CPU+GPU workloads. Apple will have to strike a balance of clocks between CPU and GPU and the cooling solution they'll deploy in the Workstation Mac Pro.
Another limiting factor perhaps will be memory bandwidth. With future 8.5GT/s LPDDR5X, Apple will get 544GB/s on par with RX6900XT. Anywhere less than that, by the wisdom of AMD, we could expect it's futile to push GPU clocks to near as high as AMD's because the many cores can't be fed with enough data to crunch.
I don't believe so. Instruction sets for GPUs are private and vendors could change at will from one generation to the other to suit their goals. Hence, unlike CPUs, they aren't bounded by complexity imposed by x86-64. I very much doubt that once Apple scales up GPU clocks, they'll still enjoy power efficiency bragging right except they have process node manufacturing advantage over, say, AMD's current "state of the art".
If we do a quick & dirty comparison between RX6900XT (AMD) and M1 Max 36C GPU (Apple), logically they're surprisingly similar to me:
ALUs: 5120 (AMD) vs 4096 (Apple)
Texture units: 320 (AMD) vs 256 (Apple)
ROPs: 128 (AMD) vs 128 (Apple)
Peak clock: 2250MHz (AMD) vs 1296MHz (Apple)
Memory bandwidth: 512GB/s (AMD) vs 400GB/s (Apple)
Power: 300W (AMD) vs 60W? (Apple)
Process: TSMC 7nm (AMD) vs TSMC 5nm (Apple)
Apple doesn't seem to me having a distinctive advantage over AMD/Nvidia. We could predict how much higher Apple could clock their current GPU without destroying their marketing rhetoric of power efficiency.
I also believe the integrated design (CPU+GPU) is a dis-advantage here. It limits how high Apple could push its GPU clocks. Higher freq > more heat > heat spill into CPU clusters. That might result in degraded CPU performance in concurrent CPU+GPU workloads. Apple will have to strike a balance of clocks between CPU and GPU and the cooling solution they'll deploy in the Workstation Mac Pro.
Another limiting factor perhaps will be memory bandwidth. With future 8.5GT/s LPDDR5X, Apple will get 544GB/s on par with RX6900XT. Anywhere less than that, by the wisdom of AMD, we could expect it's futile to push GPU clocks to near as high as AMD's because the many cores can't be fed with enough data to crunch.
Last edited:
I don't think power efficiency is a big issue for single workstations. It is for battery driven hardware. Cloud computing too I guess.
With the very fast GPUs, there are different markets. The mass market high end GPU use is for PC games, and Apple isn't in the market. Professional workstation high end GPUs are dominated by Windows users, in CAD and industrial design (correct me if I'm wrong). Apple's total market share for Macs is I think 1/10th of the PC market. Of that share, the Macs that Apple sells with the capability of installing a GPU, is just one machine, and that unit is quite costly.
Is it worth Apple spending a lot on a unique architecture that allows their CoS single memory platform, to also run 3rd party GPUs? I personally doubt the critical mass is there to do that. Such expenditure would IMO, be better spent on utilising more than one SoC ... and doing so would also make software support cheaper for all Macs, compared to having a single computer capable of separate GPUs, and the rest of Apple entire computer range not having separate GPUs.
We've heard how Apple led the reporting on Intel chip issues; and that promises of thin CPUs that never shipped from Intel caused a late redesign of an Apple notebook. That Intel quality was a cause of Apple doing their own thing. And with GPUs, Apple have not been friends with Nvidia for some time. AMD has been different, but the same thing could happen, and factually, Apple had no choice but to go along with AMD.
It seems to me that using external GPUs has many downsides for Apple. Whether that will shrink Apple's professional market for Macs, who knows ...
With the very fast GPUs, there are different markets. The mass market high end GPU use is for PC games, and Apple isn't in the market. Professional workstation high end GPUs are dominated by Windows users, in CAD and industrial design (correct me if I'm wrong). Apple's total market share for Macs is I think 1/10th of the PC market. Of that share, the Macs that Apple sells with the capability of installing a GPU, is just one machine, and that unit is quite costly.
Is it worth Apple spending a lot on a unique architecture that allows their CoS single memory platform, to also run 3rd party GPUs? I personally doubt the critical mass is there to do that. Such expenditure would IMO, be better spent on utilising more than one SoC ... and doing so would also make software support cheaper for all Macs, compared to having a single computer capable of separate GPUs, and the rest of Apple entire computer range not having separate GPUs.
We've heard how Apple led the reporting on Intel chip issues; and that promises of thin CPUs that never shipped from Intel caused a late redesign of an Apple notebook. That Intel quality was a cause of Apple doing their own thing. And with GPUs, Apple have not been friends with Nvidia for some time. AMD has been different, but the same thing could happen, and factually, Apple had no choice but to go along with AMD.
It seems to me that using external GPUs has many downsides for Apple. Whether that will shrink Apple's professional market for Macs, who knows ...
Last edited:
I agree, power is not linear. Apple is optimizing the M1 for "Max Q". I think Apple is in a good position in terms of targeting consumers, prosumers, and small studios. No need to compete with Enterprise, which Intel, Microsoft, and Nvidia dominate.
To capture enterprise market, it is more than just an efficient and visually appealing product. It needs the support, compatibility/stability, scalability, and training.
Intel's business CPUs like Xeon are always slower than consumer models, but feature more stable and secure features, it's not just about the clock speed and benchmark scores.
Apple is also focused on security, but it relies mainly on its closed architecture. For enterprise use, it needs to be secure and open at the same time, that is difficult.
My Mac Pro can't even support Nvidia GPUs or 3rd party devices natively without workarounds, why would companies use them.
To capture enterprise market, it is more than just an efficient and visually appealing product. It needs the support, compatibility/stability, scalability, and training.
Intel's business CPUs like Xeon are always slower than consumer models, but feature more stable and secure features, it's not just about the clock speed and benchmark scores.
Apple is also focused on security, but it relies mainly on its closed architecture. For enterprise use, it needs to be secure and open at the same time, that is difficult.
My Mac Pro can't even support Nvidia GPUs or 3rd party devices natively without workarounds, why would companies use them.
By the end of 2022, AMD will have RX7900XT with >15,000 cores. AMD has already demonstrated how it'll be done in AMD Instinct MI250 today. Also, Nvidia will be on par if not slightly ahead. So by 2023, Apple will need something close to make a statement in Mac Pro. Otherwise, Apple will look bad, and appear weak. They should have learned it years before by tapping into AMD's roadmap (hehe..). Something like stitching four M1 Max dies alike together will fill the bill.
The latest rumor that Apple "officially" leaked through Information no longer mentioned 4x M1 Max dies. It still mentioned 2x M1 Max dies but 4x dies will be based on the successor of M1 Max.
So next year Smaller Mac Pro will come with two M1 Max stitched together. And IMO, it'll come in a SoC daughterboard. The box will support a single daughterboard, together with a couple of PCIe slots, socketed proprietary SSD blades and without DIMMs.
I imagine the full-tower replacement will come in 2023 or later. Support more than one SoC daughterboard. For the next few years, I think Apple's gameplay is simple. Increase number of cores per die. Increase number of SoCs per daughterboard. Increase number of daughterboards per box. If Apple is serious about Workstations/Servers, perhaps they will produce a top10 candidate for the Top500 supercomputer list.
The latest rumor that Apple "officially" leaked through Information no longer mentioned 4x M1 Max dies. It still mentioned 2x M1 Max dies but 4x dies will be based on the successor of M1 Max.
So next year Smaller Mac Pro will come with two M1 Max stitched together. And IMO, it'll come in a SoC daughterboard. The box will support a single daughterboard, together with a couple of PCIe slots, socketed proprietary SSD blades and without DIMMs.
I imagine the full-tower replacement will come in 2023 or later. Support more than one SoC daughterboard. For the next few years, I think Apple's gameplay is simple. Increase number of cores per die. Increase number of SoCs per daughterboard. Increase number of daughterboards per box. If Apple is serious about Workstations/Servers, perhaps they will produce a top10 candidate for the Top500 supercomputer list.
"on par if not slightly ahead" will probably translate in real world terms to AMD's best being roughly equivalent to the second or third tier you can get from Nvidia, if history is any guide.
Really, what they should have done is treat the Mac like a standalone meta-platform instead of begging for table scraps at the iOS technology dinner, and built everything about it to be the best host for whatever the industry in general is doing. Be it CUDA, DirectX, OpenGL, Vulkan, whatever. The mac was always at its best when it was an Apple-flavoured standard PC, and always at its worst the more Apple-y it became.
Grab your popcorn for the wheels to fall off Apple's ambitions as the Chinese & Taiwanese chip industries collapse under a one-two-three punch of covid, tradewar, and drought. The American government will effectively buy Intel new foundries on American soil to keep them in business and supplying the entire American PC (and defence) Industry, as a basic national security issue. They're not going to extend that to Apple for processors only appearing in Apple's products.
So a Beast Canyon NUC, but without the RAM upgrade capability, and storage you won't be able to replace the moment a single manufacturer, Apple, decides not to sell it any more.
If you're an independent contractor, using your own hardware, like most of the people who "work for" the companies in the credits on every movie you see, you're going to buy that instead of a PC with the same performance and same software, but which you can upgrade yourself on each job, because...
I remember UMAX made a nice mac clone, that had twin CPU daughter card slots, you could run it with one, or two daughtercards. Daystar made similar machines with their GenesisMP line, that could have 2 or 4 cpus with twin daughter cards.
That has changed since last year with AMD introducing RX6900XT which is a few percentage points less performant than Geforce RTX3090 overall. Ray tracing cores though still noticeably inferior to RTX3090 and hopefully that'll be addressed in next gen cards. Intel Mac Pro users should anticipate Radeon Pro W7900X in 2023. lol
Nvidia's Ampere architecture has a few die sizes. The biggest one is one tier above RTX3090 and primarily for data centre and high performance compute products. AMD was very much lagging behind in this category until the introduction of Instinct Mi250X/MI250 very recently.
With PCIe slots, users can still do m.2 NVMe through expansion cards. A baseline Smaller Mac Pro with a rejected M1 Max bin perhaps will be quite affordable. A larger user base hopefully will attract 3rd-party sellers to do controller-less SSD blades. So that's another possibility.
You may say the SoC daughterboard a NUC. I tend to think it's a MPX-like daughterboard. Maybe depends on where you first started pondering this idea. I wish the board has a sturdy structure like Apple official MPX builds. When no extra daughterboards are installed (like in the future full-tower Mac Pro), the slots should be usable for standard PCIe cards. Preferably backward compatible with MPX modules. Since the SoC daughterboard will require way more PCIe lanes, inter-processor communication and other connectivities, the proprietary extension of the socket will be longer and/or fatter.
The slot's panel of the SoC daughterboard may carry up to six USB-C ports. The rest and majority of the area will be used for ventilation. Well conditioned silky smooth airflow runs from the front to the back to cool the board...The USB-C ports will function as display, thunderbolt and USB4. Other peripheral connectivities off the SoC daughterboard may have to fly by stealthy jumper cables to the top or front of the chassis. That's assume this is necessary or better than routing through the MPX-like socket. But if Apple wants to lock down the number of ports users can use, they may choose routing through the socket anyway.
Because it's a Mac Pro, Apple could choose to bundle 16/32/64/128GB of soldered RAM even for a reject bin of M1 Max in the baseline models. Some of less common combinations will be BTO configs. Since SoC comes in a daughterboard, it's fairly easy for Apple to manufacture and manage the inventory. It also creates ample opportunities for users to do hardware swap/trade in the 2nd hand market. Perhaps let's stop here for now.
Sure, but just as 2019 mac pro owners are going to discover a few years down the road, when Apple decides to stop making their storage blades, what do they do if a blade dies? Does the 2019 even boot without any blades installed?
I posted a concept a while ago, for a "MacStation" which is basically what you're suggesting - a chassis with 3 MPX-like slots, plus the current mac pro's slimline-style power supply. Processing is from a self-contained unit similar to a standard doublewidth MPX module, which if you look at the current motherboards and what's available in the Beast Canyon NUC compute unit, should easily be able to contain processor, storage and socketed ram. That module could go in a single slot chassis for the smallest desktop that only requires integrated graphics, or 3, 6, 9 whatever slot chassis for more processing, or more graphics.
But that relies on Apple producing an M-Series mac pro that works with off-the-shelf graphics cards, because there's no market for a "mac pro" / professional workstation that doesn't.
Which applications need Mac and huge graphics (or rather compute) cards? Video encoding has dedicated coprocessors. Apple are behind the curve in 3D modelling and rendering (how they missed that opportunity is a mystery for me). The competitor is WinPC due to software not whether there is a dedicated GPU support or not. The SoC challenge the old fashioned CPU/GPU architecture based on "slow" buses and copying information and therefore dedicated third part GPU is not obvious in a future Mac Pro. There will a long wait (WWDC 2022?) for big MP sneak peek and lots of opportunities to discuss architectures
I believe Beast Canyon NUC is Intel's direct response to SFF popularity in DIY markets. People achieved the same with desktop off-the-shelf parts a couple of years earlier and cheaper. Intel put the processor in a PCIe card form factor since NUC 9 (?). It serves little purpose IMO. Intel perhaps expects people to buy a next gen processor card to upgrade their NUC setup...
This also raises the possibility that even if Apple goes with the idea of SoCs on a daughterboard but they may still design a brand new slot connector which abandons PCIe mechanical compatibility. For Smaller Mac Pro supporting a single SoC daughterboard, there is little reason for the same slot space to accept a standard PCIe card. It's only in the full-tower Mac Pro supporting multiple SoC daughterboards. PCIe slot compatibility provides a purpose as the same space can be used for PCIe cards when not populated with SoC daughterboards. Either way, each has its own merits.
The motherboard will be more sophisticated than ever and hence expensive IMO. Between two SoC daughterboards, high speed inter-processor communication will go through the motherboard. Because a bridge-like connector such as Infinity Fibre Link for current GPUs is rather a creation for PC world. Apple will be happy to make it look cleaner and simpler.
Going back to next year's Smaller Mac Pro, supporting a single SoC daughterboard also reduces the cost of its motherboard and hence lowers entry price. In case, people haven't read the whole thread, one main reason for a SoC daughterboard is to make more efficient use of internal space, and a more compact box.
The MPX-like SoC daughterboard is already Apple's discrete GPU
There are more possibilities about the daughterboard we can discuss on another day.Anything that needs to generate a realtime 3D environment as its workspace - CAD, 3D, VR etc. Graphics cards, not Compute cards - ie cards that are dedicated to putting as many pixels onto screen as fast as possible.
The lack of the ability to remain on the cutting edge of graphics performance, where several years of graphics cards can be upgraded before a CPU / system upgrade is necessary, on the Mac, and the inability to pair a powerful GPU with a relatively inexpensive computer overall, again with said ability to remain current, is why the Blender foundation had to be paid off by Apple to keep working on a Mac version, and it's why Valve gave up on SteamVR for macOS.
They missed the opportunity because their preferred model for computing is people having 3.5 years of secondrate experience, to get ~6months of current (actually slightly less secondrate) experience, on a 4 year replacement cycle.
3D users would like to have zero secondrate experience, by being able to buy a new GPU each year for a fraction of the cost of a new computer.
Spending 25% the price of a computer every year, to remain at the cutting edge the whole time, is better than spending 100% every 4 years, and being 1 year behind, then 2 years behind, then 3 years behind the leading edge, and spending most of that time with a machine that is uncompetitive with everyone on current GPUs, which means where does the money come from for that next 100% purchase?
Software which is dependent on a vibrant, and healthy GPU upgrade market. Apple does not provide this.
The SoC challenge the old fashioned CPU/GPU architecture based on "slow" buses and copying information and therefore dedicated third part GPU is not obvious in a future Mac Pro. There will a long wait (WWDC 2022?) for big MP sneak peek and lots of opportunities to discuss architectures
The "old fashioned" architecture will outperform the SOC architecture at the same pricerange, whereby performance includes "annual cost of keeping a workstation at the leading edge of performance".
Apple tried a non-upgradable "pro" workstation. They were laughed out of studios and workflows worldwide, and had to mea-culpa by making literally the very thing they had claimed everyone would abandon to follow their model.
Well it's Apple's problem, and the problem of any company who isn't American, and who fabs chips in China and Taiwan.
it seem like you have a limited view of what a work station can be. The 2013 MP morphed into a iMac Pro that was well received. Both would have shone with ASi inside especially if they were updated every 1 or 2 year. Both would be able to support a Jade 4C in terms of power draw and Jade 4C will likely has far better for performance for video editing work than a fully loaded 2019 MP. I would not be surprised if a Jade 4C machine cost 5000$ just as much as a 6900X does. There are not many who bother to change a 5000$ GPU every year to have the latest and greatest. People change every three year when the GPU is worth naught according to the economy department. Alternative they lease the machines.
People said those who wanted a slotbox to replace the 5,1 had a limited view of what workstations could be, and that the trashcan was going to revolutionise the industry, and everyone would be making workstations like it from then on.
It was a one-off, that went nowhere, and left no legacy.
People said the iMac Pro should be more than enough for anyone who wanted a Mac Pro. The only people who praised it, were iMac customers, and Apple-appeasing / Apple-dependent media shills. Not even the developers who were partners for its launch followed through on their promises of software for it.
It was a one-off, that went nowhere, and left no legacy.
So, what do you think is going to happen the third time, when Apple tries to "redefine the workstation" and make yet another appliance, with the same full-replacement-price-for-GPU-upgrade pricing model?
There seems to be this weird idea amongst a proportion of Apple fans, that the slotbox is some sort of compromise, or lazy failure to do something better, when if fact it's a superbly adapted evolved answer to the market. You can't make a better workstation than a slotbox, because there is no "innovation" that is worth more than the flexibility (and psychological value of that flexibility, even if unused) a slotbox provides. Apple understood this prior to 2013, and acknowledged it in 2019.
Video editing is a niche, a niche. It is not the be all and end all of Mac Pro use, as observed by the fact that whenever Apple makes a machine that's all hardcore foussed on video editing, to the exclusion of being good at other niches, it fails in the market.
Again, the gaslighting that the trashcan would have been a success if only Apple had updated it each year, ignores that no one wants to replace a workstation afresh every year. It failed because users couldn't post-purchase upgrade it, not because Apple couldn't new-model upgrade it.
Mocking it as a non-upgradable appliance was literally HP's major Z-Series marketing campaign.
The 6900XT is more like $2300, not $5k. That ~$2000-3000 (at current overpriced levels) is where the bulk of professionally used graphics cards sit. Most professionally used workstations are going to be running 3080/3080ti setups.
Apple is not going to sell, or lease you a machine with performance equivalent to that GPU's range, staying at that relative level of performance, for 2k-3k / year.
Apple is not going to sell you a machine that eats into the current "big iMac" sales. The "mini" Mac Pro is going to start above the iMac, for the computer alone.
Last edited:
Workstation is relative these days. Many PCs are capable of workstation tasks. Cloud computing is also another option diminishing the need for large power consuming expensive workstations. No need to upgrade on the cloud, just elasticity based on the need, instead of 32 cores, 64 cores, how about hundreds to thousands of cores.
The current Mac Pro kind of shows how much Apple is supporting it. There is not much MPX modules and upgrades. Very expensive GPUs that are not industry leading and incompatible with standard GPUs .
Upgradeability is also irrelevant these days. Technology changes so rapidly, that even upgrading the Mac Pro to the max is no point. It supports only Cascade Lake, Ram is only 2933MHz, and only PCIe 3.0. The best way is just buy the most you can afford and need in the beginning and just upgrade to a whole new computer when time comes. After a few years of workstation use, no point of upgrading an old computer since the existing hardware would need to be refreshed along with it. ie. worn out power supply, CPU, motherboard, etc. which would be difficult to get from Apple.
The current Mac Pro kind of shows how much Apple is supporting it. There is not much MPX modules and upgrades. Very expensive GPUs that are not industry leading and incompatible with standard GPUs .
Upgradeability is also irrelevant these days. Technology changes so rapidly, that even upgrading the Mac Pro to the max is no point. It supports only Cascade Lake, Ram is only 2933MHz, and only PCIe 3.0. The best way is just buy the most you can afford and need in the beginning and just upgrade to a whole new computer when time comes. After a few years of workstation use, no point of upgrading an old computer since the existing hardware would need to be refreshed along with it. ie. worn out power supply, CPU, motherboard, etc. which would be difficult to get from Apple.
Yes, the new Apple silicon introduces the concept of appliance computing; you buy what you need & when you upgrade, you upgrade the entire appliance...
But...! For those whose workflow can use a collection of appliances to offload large computes (rendering, code compiling, video encoding, etc.), when you get a new appliance, you drop the old one into the compute farm...?
But...! For those whose workflow can use a collection of appliances to offload large computes (rendering, code compiling, video encoding, etc.), when you get a new appliance, you drop the old one into the compute farm...?
I see little reason to debate this old topic if Apple splits Mac Pro into two designs: Smaller Mac Pro and Full-tower Mac Pro. When that happens (users will know fairly quickly), its intention is clear and cloud - Apple is going to take care of both crowds (the trashcan and the realman).
So far in this thread we've (or mostly my wishful thinking) defined Smaller Mac Pro: one SoC daughterboard, a couple PCIe slots, socketed SSDs. Half the size of Intel Mac Pro. Personally I think the size will be very close but not bigger than a compact microATX box.
What about the future Full-tower Mac Pro?
Believe it or not. You don't need a crystal ball into Apple's future but look at the current and past of the PC world. I had a poke on Nvidia and saw this exciting NVswitch thing:
Each connection in the diagram is one NVlink (AMD's close equivalent known as Infinity Fibre Link). Each NVlink is 600GByte/s, bidirectional combined. One NVswitch has 12 NVlink "ports". So theoretically any two GPUs can talk to each other at any moment through NVswitch and do it super fast.
I expect if Apple doesn't treat Mac Pro as hobbies, they'll design something similar. Think of each of those GPUs in the diagram an Apple SoC daughterboard. Of course for Full-tower Mac Pro, I believe Apple at most will go with four SoC daughterboards. Who and how many will need a more beasty machine? Basically then the next Full-tower Mac Pro will be up to four "Radeon Pro RX7900 Extreme equivalent" GPUs that also happen to come with a bunch of fast CPU cores for free.
Another use-case of Full-tower Mac Pro that I touched on previously: laying down the Full-tower Mac Pro horizontally is a 5U (?) rack mountable unit. Just like Intel Mac Pro, Apple will design a rack casing for this. Can you imagine what will happen next if Apple is serious about data centre business?
So far in this thread we've (or mostly my wishful thinking) defined Smaller Mac Pro: one SoC daughterboard, a couple PCIe slots, socketed SSDs. Half the size of Intel Mac Pro. Personally I think the size will be very close but not bigger than a compact microATX box.
What about the future Full-tower Mac Pro?
Believe it or not. You don't need a crystal ball into Apple's future but look at the current and past of the PC world. I had a poke on Nvidia and saw this exciting NVswitch thing:
Each connection in the diagram is one NVlink (AMD's close equivalent known as Infinity Fibre Link). Each NVlink is 600GByte/s, bidirectional combined. One NVswitch has 12 NVlink "ports". So theoretically any two GPUs can talk to each other at any moment through NVswitch and do it super fast.
I expect if Apple doesn't treat Mac Pro as hobbies, they'll design something similar. Think of each of those GPUs in the diagram an Apple SoC daughterboard. Of course for Full-tower Mac Pro, I believe Apple at most will go with four SoC daughterboards. Who and how many will need a more beasty machine? Basically then the next Full-tower Mac Pro will be up to four "Radeon Pro RX7900 Extreme equivalent" GPUs that also happen to come with a bunch of fast CPU cores for free.
Another use-case of Full-tower Mac Pro that I touched on previously: laying down the Full-tower Mac Pro horizontally is a 5U (?) rack mountable unit. Just like Intel Mac Pro, Apple will design a rack casing for this. Can you imagine what will happen next if Apple is serious about data centre business?
It's an interesting idea, but I think the full-size Mac tower is just the current Intel Mac Pro with updates as necessary to extend it. Even a new motherboard is probably possible. As some folks on this thread have made abundantly clear, nothing short of an Intel slotbox that can be upgraded annually with a new GPU card, ideally from Nvidia, is going to meet the need for the committed tower buyer (Not that I think you're going to see Nvidia). That computer already exists today and can be bought at any local Apple Store. Apple Silicon is going to be used where it excels and that doesn't come from trying to force it to act like a PC.
The NVswitch is fascinating tech, although I doubt these will be in PCs anytime soon, let alone going to Mac. This is something that is used in places like AI research laboratories.
The current Mac Pro can do Infinity Fabric quad GPUs, but Mac OS and hardware is too constricted for many industries and does not have the enterprise customer support that the larger and more open OEMs can provide. (Ie. where can one find a new motherboard and power supply, if it fails. Is not any standard form factor like ATX. Unless that new right-to-repair program now has a kit, please let me know) I don't think Apple is trying to get into that market.
Apple is mainly a consumer tech and entertainment business. The main purpose of Macs these days are for development in IOS and creator contents that goes back into Apple ecosystem like App Store, Apple TV, Arcade, etc.
Apple has a unique thing going, I agree, don't need to follow the PC market. That was their whole philosophy, not to be like PC, but an alternative way of computing.
The current Mac Pro can do Infinity Fabric quad GPUs, but Mac OS and hardware is too constricted for many industries and does not have the enterprise customer support that the larger and more open OEMs can provide. (Ie. where can one find a new motherboard and power supply, if it fails. Is not any standard form factor like ATX. Unless that new right-to-repair program now has a kit, please let me know) I don't think Apple is trying to get into that market.
Apple is mainly a consumer tech and entertainment business. The main purpose of Macs these days are for development in IOS and creator contents that goes back into Apple ecosystem like App Store, Apple TV, Arcade, etc.
Apple has a unique thing going, I agree, don't need to follow the PC market. That was their whole philosophy, not to be like PC, but an alternative way of computing.
Just because some end users do not need PCIe expansion slots does not make them any less relevant; your "the trashcan and the realman" phrasing sounds like the "trashcan" users are irrelevant...
"Half the size" does not automatically mean "half as tall", or "half as wide"; "half as" can cover all dimensions, so half the overall volume might be more realistic (which would place it close to a 20L microATX box)...
"Bidirectional combined", so 300GB/s each way...? M1 Max memory bandwidth is 400GB/s, and the theoretical M1 Max Duo & Quad may have bandwidth of 800GB/s & 1.6TB/s; so the 300GB/s bandwidth of the NVLink is slower than any of those...?
The upgrade path for the 2019 Mac Pro would be the W3300/Ice Lake CPUs, which would require a new mobo because they use a different socket than the current 2019 Mac Pro...
The Mac really took off when DTP (Desk Top Publishing) was fresh, Macs were the backbone of many grassroots publications; without DTP & the Macintosh, I doubt Adobe would be where they are today...?
Today, some of the "niche" markets the Mac could command are video, audio, & DCC (3D et al)...
Apple has decent video & audio suites (Final Cut & Logic), there are plenty of quality third-party a/v suites (Resolve, Pro Tools, Media Composer, etc.)...
There are also a good number of 3D/DCC suites available on macOS, hopefully these will be improved to meet the performance available in the new Apple silicon Macs...
And Apple could do to add a 3D/DCC suite into their software product lineup...?
Hijacking what Professional is? So a professional = upgradable computer? The most popular professional computer is the MBP which by the way is also non upgradable. After that the iMac come and then nothing nothing and then the MP.
I doubts video editing is a niche market. Providing the afterburner card for a small number of people seem strange. High end Macs outside of video editing field - do you have numbers of the size of that market?
3080 cards are gaming cards as far as I know. Are you discussing the corporate pro market or what?
The PC industry do a lot of mocking these days. Mocking is signs of weakness. Apple mocked PC earlier and Apple was for sure an underdog at that time.
Also MP can also be seen as an appliance; configure, buy, use and dispose. That is probably the majority of users but who knows?
Yes, the definition (or as good as a definition as you can find) of a "professional" workstation is a slotbox. Professional products are by their nature, utilitarian, ergonomic, user & field serviceable, downtime minimising, flexibility enhancing.
Look at vehicles - a professional might use a range rover, but a Professional utility truck, even one as styled as a Ford F150 has a removable rear tray, and can have any other body module bolted on instead - it might even be sold as a cab with a naked chassis at the back. Or a van - a professional might use a Chrysler Voyager, but a Professional van is an empty shell, that can be configured by the user, and reconfigured by the user whoever the circumstances may require.
That independent reconfigurability is what makes them Professional Products, as opposed to Consumer products used to make money.
That's why every professional workstation maker, be it Apple, HP, Lenovo, Dell, Puget, Boxx etc. makes slotboxes.
A laptop is about being portable, it's not the same thing as a Desktop. Desktop workstations used in a professional environment (ie an environment configured for professional work standards) don't have "small" as a requirement. If someone needs a desktop workstation as small as the trashcan, the bigger problem is they're not operating in a professional manner, by failing to have a sufficient workspace.
If any professional needs upgradability, and the product is not upgradable, it is by virtue of that not a professional product.
Well, Apple made a video-specific Mac Pro. It was a failure even in the video industry, and no other industries flocked to it in their place, so I guess that answers your doubts.
Afterburner is exactly the sort of thing Apple should be doing - making generic workstations, and then providing tailored modules to make them specialised.
That way, the video editor can have their machine, the audio specialist can fill the machine with audio cards, and have their machine, the scientist or engineer can have their specialist data acquisition / equipment control card etc, and have their machine.
You're going to argue consumer technology like a Macbook Pro is "professional", while arguing that a consumer GPU isn't? Until the 2019, Apple didn't have any products with actual professional GPUs in them (Quadro-fitted Mac Pros excepted), they just slapped a "pro" sticker on gaming cards.
Guess what cards are in the workstations that are used by multinational corporate professional studios to create games (a bigger industry than the film industry)?
Go on, guess.
Well HP (and frankly the rest of the workstation industry, and most of the professional content creation industry) was able to mock Apple for 6 years on that one joke of a system, all while eating Apple's professional market, so not really all that weak.
It's a piece of capital plant, not an appliance. Just because something can be depreciated over 4 years, doesn't mean a business is eager to re-buy it every 4 years. The more disposable a product is, the worse its lease terms will be, because the leasing company can't sell it afterwards.
Last edited: