MP 7,1 Would the Mac Pro 2019 (7.1) capable to upgrade from Intel CPU to the new Apple Silicon CPU??

[Kentrent]

Hello my wonderful Mac Pro 2019 owners and community!!!

Do you think Apple would make the Mac Pro 2019 (7.1) capable to upgrade from Intel CPU to the new Apple Silicon CPU??

Since a lot of Mac Pro 2019 owners spend a lot of money on this machine, do you think Apple would design something to allow all of current Mac Pro 2019 owners swap out the Intel CPU for the new Apple Silicon CPU? If this happens, used Mac Pro 2019 would gone up in prices and in hot demand. What do you think?? Let’s discuss.

 

[Boil]

1 - Silicone is for breast implants, silicon is for computer chips...

2 - No...

3 - There is no three... ;^p

 

[Algr]

Apple hasn't done CPU upgrades in a long time. The development cost and likely low sales would prohibit it. You'd be better off just putting a Mac Mini on top of your pro, for when you need M1.

 

[MisterAndrew]

I don't think the current 7,1 will be able to have an Apple Silicon CPU inserted into it. Don't expect 7,1 prices to remain high after the Apple Silicon version is released. If the new version is not expandable then the 7,1 will be in stronger demand from a certain group of users, but will still not be desirable overall since Intel support will end sooner than later.

If you need one right now I suggest to buy a lower cost base model before deciding to spend a lot of money on upgrades until we see what the new version is like.

 

[deconstruct60]

Hello my wonderful Mac Pro 2019 owners and community!!!

Do you think Apple would make the Mac Pro 2019 (7.1) capable to upgrade from Intel CPU to the new Apple Silicon CPU??

It isn't going to get a new generation Intel chip let alone something from a different vendor. The CPU socket changed for the next gen Intel product. There is no next gen Intel CPU that will fit or is compatible with the PCH chipset present in the MP 2019.

Apple really isn't in the computer parts selling business. Apple didn't sell a MP 2019 container and will in the future sell major parts to swap and and keep the container running. They don't sell retail motherboards or "build it yourself" kits.

The main logic board is highly likely not going to be the same size. There may be some design cues shared between systems. However, it is unlikely they are going to use the exact same board dimensions. the exact same power supply, or the exact same case for the next Mac Pro. Highly likely the Thunderbolt ports of the 2019 model won't be future compatible either.

Since a lot of Mac Pro 2019 owners spend a lot of money on this machine, do you think Apple would design something to allow all of current Mac Pro 2019 owners swap out the Intel CPU for the new Apple Silicon CPU?

No. Because Apple largely presumes that folks who spent money on a Mac Pro 2019 can generate enough value (revenue ) from using the Mac Pro 2019 for it to pay for itself. It isn't like Apple "owes you something" for buying the Mac Pro. You hand over money and Apple hands over a Mac Pro. That is the swap. If you buy used then you didn't even hand Apple any more directly. So they owe you what extra? Paying $700 or $7,000 or $14,000 for a Mac doesn't buy different status or technical support classification with Apple.

If this happens, used Mac Pro 2019 would gone up in prices and in hot demand. What do you think?? Let’s discuss.

https://www.apple.com/shop/trade-in

If check Apple's trade in values for a Mac Pro .... that really isn't a future prices hype train area for them.

 

[DrEGPU]

Agree with all of the above, but it would pretty neat if they could build an M1/M2 chip into a PCIe card or MPX thing, kind of like an FPGA card like the afterburner. I’m not an engineer so I don’t know if that’s even possible, much less it’s practicality

 

[TobiasT]

Put a cheap price and sell the MP7.1 before it is too late.

 

[Algr]

1 - Silicone is for breast implants, silicon is for computer chips...

But what if Apple made silicone silicon for a well rounded experience in personal computing ?

 

[Boil]

1 - Silicone is for breast implants, silicon is for computer chips...

But what if Apple made silicone silicon for a well rounded experience in personal computing ?

Sweet, the Apple iBot...!

 

[macguru9999]

I had a performa 6300 .... I put in a 6400 motherboard and a POWERPC upgrade card plugged in the PDS (processor direct slot). Those were the days ... sigh

 

[Colstan]

If you've got a time machine, what you are asking for has existed, in a fashion.

Behold, the OrangePC Model 620:

It was a Windows compatibility card featuring a blinding fast WinChip 200Mhz C6 with 16MB of SDRAM on a PCI card, compatible with any Macintosh that had a free PCI slot. You can read Macworld's review from 1999 of a similar model if you want. They managed to upgrade it to a 300Mhz AMD K6-2 using the ZIF socket, and even the RAM could be upgraded.

What are the chances of getting this sort of thing with Apple Silicon for the 2019 Mac Pro? Somewhere between zero and none.

I understand that the x86 era is going to be hard to let go of. Some Mac users will have difficult choices ahead of them, and not everyone is going to be pleased with the end result, which happens with all major product transitions. This is just my opinion, but I think the Apple Silicon Mac Pro is going to be a substantial departure from what Mac Pro users are used to. I think upgradeability is going to be significantly limited, forget about upgrading old Intel systems.

I think we've seen the last of the MPX modules for the 2019 Mac Pro. The W6600X is the final third-party GPU upgrade that Apple will ever release. Maybe they will have their own discrete GPU solution for Apple Silicon Macs, but it won't come from AMD or Nvidia (and certainly not Intel Arc). Nor will we see the likes of Intel or AMD CPUs in a future Mac. Apple is in the process of eliminating everything Intel from their products, as they continue to integrate their own solutions, locking out third-party suppliers in the process.

I think that if there are ever any future upgrades for the 2019 Mac Pro then they will come third-party suppliers, in whatever limited form they may take, assuming anyone bothers, but they will not be from Apple...and I say this as someone who just purchased a 2019 Mac Pro two weeks ago.

 

[macguru9999]

With such low sales worldwide , I doubt there will be many 3rd party accessories for the 2019 mac pro, apart from what has already been released. Still, I hope to get one after the 8,1 is released. It will still be a useful machine.

 

[Algr]

Behold, the OrangePC Model 620:

It was a Windows compatibility card...

I remember those things. They sounded great until you realized that they were barely any cheaper then a complete Windows PC.

 

[mattspace]

If you've got a time machine, what you are asking for has existed, in a fashion.

Behold, the OrangePC Model 620:

Apple used to sell PowerPC Macs with an Intel PC PCI card pre-installed, as a factory-standard SKU / Named Model, e.g PowerMac 6100/66 DOS Compatible.

 

[theluggage]

Agree with all of the above, but it would pretty neat if they could build an M1/M2 chip into a PCIe card or MPX thing, kind of like an FPGA card like the afterburner. I’m not an engineer so I don’t know if that’s even possible, much less it’s practicality

I think that is one road that a mythical new Mac Pro could take - a box of MPX-like slots that you could fill up with M? Max/Ultra modules - with the MPX bus providing power, PCIe and video inter-connection. Whether they'd actually be compatible with the 2019 MP is another matter (but it would make some of the investment in developing MPX pay off).

The hurdle with the Apple Silicon Mac Pro is that while it would be technically feasible for Apple to develop an Apple Silicon "Xeon Killer" with the same sort of insane RAM capacity and PCIe bandwidth as the Xeon-W, (a) that would be a major departure from the design philosophy of Apple Silicon to date - use on-die GPUs + on-package unified RAM & reap the advantages of tight integration and short, fast interconnections, mainly thunderbolt for I/O - and (b) it would be hugely expensive for Apple to develop a new Apple Silicon die dedicated to the Mac Pro, that would only sell in relatively tiny numbers. It would be far more economical to re use the M? Max design on which all the other higher-end Macs are based. Since even the rumoured M? Extreme (4xMax) doesn't completely solve the issue of the more challenging RAM and GPU requirements the solution could be to go for clusters of M? Ultras instead.

 

[Boil]

It would be far more economical to re use the M? Max design on which all the other higher-end Macs are based. Since even the rumoured M? Extreme (4xMax) doesn't completely solve the issue of the more challenging RAM and GPU requirements the solution could be to go for clusters of M? Ultras instead.

You do know that the Mn Ultra is two Mn Max dies connected with UltraFusion...?

 

[theluggage]

You do know that the Mn Ultra is two Mn Max dies connected with UltraFusion...?

Yes, so what? "Ultrafusion" makes the two Maxs work like one big chip - it's not the same thing as clustering.

There's a rumour of an "extreme" chip with 4x Maxs - and also quite a lot of debate as to how it could work physically. I doubt it could go beyond 4 - and the cost of manufacturing is going to go up exponentially with each Max you add because each version is going to have a smaller market. At some point it will make more sense, and give better "scalability" to keep the chips separate and cluster them together using PCIe or Thunderbolt links and some NUMA-like technology to pool RAM. My suspicion is that the Ultra is about as large as it's worth going in a single package.

The 2019 Mac Pro supports up to 1.5TB of RAM and presumably some users need that - seeing as Apple use the expensive m-suffix version of the Xeon W in order to support that.

A M1 Ultra has up to 128GB RAM. The rumoured Extreme would have 256GB - maybe more if the RAM technology has improved in a "M2 Extreme", but remember that the M1 Pro/Max/Ultra already got the LPDDR5 upgrade that went into the regular M2. They're not gonna use UltraFusion to hook 16 Max dies together to get 1TB RAM. However, 3-4 Ultras on a MPX card, 2-3 cards in a Mac Pro? Too rich for my blood, but still a lot more feasible than one megachip for a tiny audience...

 

[hwojtek]

Do you think Apple would make the Mac Pro 2019 (7.1) capable to upgrade from Intel CPU to the new Apple Silicon CPU??

What? No.

 

[Boil]

Yes, so what? "Ultrafusion" makes the two Maxs work like one big chip - it's not the same thing as clustering.

So what..? So just throw the whole UMA thing out the door...?

Because clustering is not the same as UltraFusion; Ultrafusion is designed to make all attached SoCs act as and be seen as one "mono SoC"...

There's a rumour of an "extreme" chip with 4x Maxs - and also quite a lot of debate as to how it could work physically. I doubt it could go beyond 4 - and the cost of manufacturing is going to go up exponentially with each Max you add because each version is going to have a smaller market. At some point it will make more sense, and give better "scalability" to keep the chips separate and cluster them together using PCIe or Thunderbolt links and some NUMA-like technology to pool RAM. My suspicion is that the Ultra is about as large as it's worth going in a single package.

Again, throwing the entire UMA thing out the door...

Mn Extreme would be four Mn Max SoCs stitched together with UltraFusion to appear as a single SoC to macOS...

The N3/N3E (N3P/N3S/N3X) processes should allow for more transistors in the same area, meaning more cores in the die...

It has been mentioned that this might allow Apple to get the core count found in an Ultra into a single Mn Max size die; this would basically allow four much more core-loaded Mn Max SoCs to be combined into one Extreme SoC; this could give us the equivalent of eight M1 Max SoCs worth of compute in a Mn Extreme SoC...

The 2019 Mac Pro supports up to 1.5TB of RAM and presumably some users need that - seeing as Apple use the expensive m-suffix version of the Xeon W in order to support that.

A M1 Ultra has up to 128GB RAM. The rumoured Extreme would have 256GB - maybe more if the RAM technology has improved in a "M2 Extreme", but remember that the M1 Pro/Max/Ultra already got the LPDDR5 upgrade that went into the regular M2.

If you extrapolate from the M2 SoC, with a maximum RAM of 24GB, then a fully loaded M2 Extreme SoC should be able to go up to 384GB...

If Apple decides t use LPDDR5X SDRAM for the Mac Pro, then they could get up to 1TB RAM into a system...

They're not gonna use UltraFusion to hook 16 Max dies together to get 1TB RAM. However, 3-4 Ultras on a MPX card, 2-3 cards in a Mac Pro? Too rich for my blood, but still a lot more feasible than one megachip for a tiny audience...

They cannot use UltraFusion to hook eight SoCs together, because then they would be exceeding the 64-core limit (for CPU cores) that macOS can handle...

And with multiple SoC-loaded MPX cards, you once again both ignore the core limit and throw UMA out the door...

Apple might put out a M2 Ultra / M2 Extreme Mac Pro within the next six months, but I feel it would be a placeholder...

The real ASi Mac Pro would be a M3 Ultra / M3 Extreme system, with a much denser core count, higher RAM limits, and hardware ray-tracing...

Apple may also put out a GPGPU add-in card for dedicated rendering while allowing the iGPU to stay free for maximum display/viewport output (resolution/frame rate/etc.)...

 

[deconstruct60]

I think that is one road that a mythical new Mac Pro could take - a box of MPX-like slots that you could fill up with M? Max/Ultra modules - with the MPX bus providing power, PCIe and video inter-connection. Whether they'd actually be compatible with the 2019 MP is another matter (but it would make some of the investment in developing MPX pay off).

The major problem is that the Max just provisions x4 PCI-e v4 lanes ( and the x8 for the Ultra). Mainly it is just enough to hook up ethernet, wifi and some seondary I/O on fewer lanes (rather than a focus on high end bandwidth). M2 with a similar die edge allocations to Memory bus , UtraFusion , Thunderbolt , PCI-e , and etc. would not bode for much of a Mac Pro . It doesn't relatively buy much of a bandwidth to any other PCI-e Card. Could do some hackery with internally consumed TB controller output and another internal TB peripheral controller to get some slot allocation ( Rube Goldberg internal TB PCIe enclosure to get some PCI-e v3 slots. ) .


It would certain not get you two "long distance" ,low NUMA coupled Max/Ultra . You couldn't mix and match the core types (stuck to a single generation). And largely would be just decoupling the cards. Effectively it would be more of a blade server enclosure where you'd be putting multiple independent Macs in a single box with some shared power supply and fan/cooling infrastructure.

Ultrafusion using 10K connections to hook together just two Maxes. Maybe some of those are unused. However, even at half that (5k), how could the run that many connections out the bottom of an add-in board? UltraFusion primarily works because the pads being used are extremely small and extremely low powered. A card that plugs into a logic board in a fashion like the standard PCI-e cards do is not either of those.
[ NOTE: the infinity Fabric links come out the other side of the MPX GPU modules that have them and are no where near that wide or fast bandwidth wise. ]

The hurdle with the Apple Silicon Mac Pro is that while it would be technically feasible for Apple to develop an Apple Silicon "Xeon Killer" with the same sort of insane RAM capacity and PCIe bandwidth as the Xeon-W, (a) that would be a major departure from the design philosophy of Apple Silicon to date - use on-die GPUs + on-package unified RAM & reap the advantages of tight integration and short, fast interconnections, mainly thunderbolt for I/O - and (b) it would be hugely expensive for Apple to develop a new Apple Silicon die dedicated to the Mac Pro, that would only sell in relatively tiny numbers. It would be far more economical to re use the M? Max design on which all the other higher-end Macs are based. Since even the rumored M? Extreme (4xMax) doesn't completely solve the issue of the more challenging RAM and GPU requirements the solution could be to go for clusters of M? Ultras instead.

There is an in between path where the Ultra isn't strictly built from just two laptop Maxes. If the studio (and iMac Pro) Maxes could sell in sufficient numbers Apple could aggregate enough units to fund a "desktop" base chiplets that has some more flexible and desktop oriented I/O. Not completely backtracking out of the iGPU , unified memory stance into ultra-modularity, but built better to scale up.

Putting the UltraFusion connector on MBP 14/16 SoCs is somewhat of waste. It is a decent chunk of silicon that will never get used. Ye,s they are getting the MBP buyers to pay for it. But asking folks to pay for it and they can't use it is somewhat dubious in the long run. Apple could make more laptop Maxes with fewer wafers if it didn't have that pragmatically laptop useless die area allocation on them. ( current make more Pros without it. ). Likewise 4 laptop Max chips with 3 redundant SSD controllers and secure enclaves, at least 8 superfluous TB controllers, and four dead embedded display controllers is a collective waste of wafer space also if make 50K of those kinds of packages.



Apple has choked of PCI-e provisioning in the Max (and lower ) because there are not dual 10GbE Ethernet systems (or even any Ethernet on half of the line up). Less than 5 ports. etc. The dGPUs disappeared in the rest of the line up and the embedded x16 allocation just got tossed.

The Mac Pro can't completely do it by itself but loop Studio , iMac Pro , Mac Pro (i.e., get some models that go all the way down to the $2K to generate demand volume. ) that it can possibly pay for some limited deviations that brings in some more sensible desktop oriented I/O infrastructure. It won't get them a Threadripper or Xeon W-6xxx 'killer' SoC, but something better than 4 Mac laptop design driven chips.


Long term I doubt the Ultra and Extreme will be build from 'laptop' Maxes if Apple is anywhere near serious in keeping up in the desktop space across the zone where the upper iMac range and entry-mid Mac Pro range has been historically. They probably will 'give up' on the far fringe top end Mac Pro. In the Intel era that was all 'freebie' stuff that fell out of just buying the Intel CPU/Chipset products in that space. Apple's primary concern there was slapping their mark-up on top and collecting the extra money. ( the >1 TB CPU choice for the Mac Pro 2019 is a extra tax that Apple gets to collect at least as much as any user technical requirement thing. Intel had an extreme revenue tax on those CPUs and Apple just piggybacked off of that. )

 

[Boil]

Great point on all of the redundant systems four Mn Max SoCs would bring...

Having two Mn Max SoCs, each paired with a GPU-specific SoC, that might help cut down on redundant systems and allow more GPU horsepower...

But having a series of SoCs designed from the beginning for high performance desktop/workstation usage, that seems the smart way to go...

That way the desktop lineup might look like...

Mac mini = M2 & M2 Pro

Mac Studio = M2 Max & M2 Ultra

Mac Pro = M2 Extreme - choice of CPU heavy or GPU heavy core count options

M3 family on N3/N3E (N3P/N3S/N3X) though, that is where we should really see a powerful Mac Pro; denser dies for more cores, hardware ray-tracing, PCIe slots, more RAM, etc. ...?

M3 Extreme on N3X process, only in the Mac Pro...!

 

[Kentrent]

If you've got a time machine, what you are asking for has existed, in a fashion.

Behold, the OrangePC Model 620:

View attachment 2079474

It was a Windows compatibility card featuring a blinding fast WinChip 200Mhz C6 with 16MB of SDRAM on a PCI card, compatible with any Macintosh that had a free PCI slot. You can read Macworld's review from 1999 of a similar model if you want. They managed to upgrade it to a 300Mhz AMD K6-2 using the ZIF socket, and even the RAM could be upgraded.

What are the chances of getting this sort of thing with Apple Silicon for the 2019 Mac Pro? Somewhere between zero and none.

I understand that the x86 era is going to be hard to let go of. Some Mac users will have difficult choices ahead of them, and not everyone is going to be pleased with the end result, which happens with all major product transitions. This is just my opinion, but I think the Apple Silicon Mac Pro is going to be a substantial departure from what Mac Pro users are used to. I think upgradeability is going to be significantly limited, forget about upgrading old Intel systems.

I think we've seen the last of the MPX modules for the 2019 Mac Pro. The W6600X is the final third-party GPU upgrade that Apple will ever release. Maybe they will have their own discrete GPU solution for Apple Silicon Macs, but it won't come from AMD or Nvidia (and certainly not Intel Arc). Nor will we see the likes of Intel or AMD CPUs in a future Mac. Apple is in the process of eliminating everything Intel from their products, as they continue to integrate their own solutions, locking out third-party suppliers in the process.

I think that if there are ever any future upgrades for the 2019 Mac Pro then they will come third-party suppliers, in whatever limited form they may take, assuming anyone bothers, but they will not be from Apple...and I say this as someone who just purchased a 2019 Mac Pro two weeks ago.

Share the specifications of your Mac Pro with us… and how much did you got it for? so after all, you still love the 2019 Mac Pro? I love it too… I recently bought one with 28 cores, 384 GB Ram, 8 TB SSD, 2x Vega II Duo, and afterburner card. This will be my desktop for many more years to come. I paid mine for $10k.

 

[Colstan]

Share the specifications of your Mac Pro with us… and how much did you got it for? so after all, you still love the 2019 Mac Pro? I love it too… I recently bought one with 28 cores, 384 GB Ram, 8 TB SSD, 2x Vega II Duo, and afterburner card. This will be my desktop for many more years to come. I paid mine for $10k.

Yes, I absolutely adore my 2019 Mac Pro. I would say that it's the best computer that I have ever owned, and I've owned many over the years. I got mine about three weeks ago. I already mentioned it in another thread, but I'm perfectly happy to repeat it, because I switched the graphics card for better performance.

Specs:

16-core 3.2Ghz Xeon W-3245
96GB RAM
2TB SSD
580X GPU

None of this is too remarkable, compared to other Mac Pro models, until you find out how much I paid for it: $2,000 in U.S. currency. How did I manage this? Ebay search error. The computer company that sold it on Ebay didn't bother to check to make sure it came up when searching for "2019 Mac Pro", hence I was the only bidder. The same company has been trying to sell similar models for $7,000, so to say I got a good deal is an understatement. These go for almost $10,000 on Apple's website brand new.

I've paired it with a 21.5-inch LG UltraFine for the full "Retina" experience, which I got for half MSRP last year. Of course, the only weak spot in this system is the GPU. I originally had a W6600X inside, but sent it back to Apple, because I wanted the highest performance GPU I could find. So, I replaced it with a 6900XT.

It took me forever to find a 6900 that will fit inside the Mac Pro case, but I managed to find one open box, and saved a few dollars in the process. The "Gigabyte Gaming OC 16GB" is one of the few that will squeeze inside the housing. I figured that a beastly machine like this Mac Pro needed an equally mammoth graphics card to go inside it.

All told, after rendering unto Caesar and shipping costs, I paid about $2,860 total for the Mac Pro and GPU. A year or two ago the GPU alone would have taken up most of that. Obviously, I got a stupid good deal.

 

[CooperBox]

But what if Apple made silicone silicon for a well rounded experience in personal computing ?

And even more authentic if manufactured in Bristol.........😜

 

[theluggage]

The major problem is that the Max just provisions x4 PCI-e v4 lanes ( and the x8 for the Ultra).

...plus 4 Thunderbolt 4 ports on the Max, at least 6 on the Ultra (presumably 8, but the Studio tops out as 6) - so there's plenty of TB4 ports to pay with, each of which uses the equivalent of a 4 PCIe lanes. At best, it may be possible to configure some of the TB4 ports as 4x PCIe slots - at worst you could use TB-to-PCIe bridges to get the PCIe lanes back. Or, just use TB4 as the processor interconnect.

Effectively it would be more of a blade server enclosure where you'd be putting multiple independent Macs in a single box with some shared power supply and fan/cooling infrastructure.

Well, yes - except the communication between nodes would be via PCIe or Thunderbolt rather than Ethernet - so faster than your typical clustering system, not as fast as a "proper" NUMA system with dedicated links.

Putting the UltraFusion connector on MBP 14/16 SoCs is somewhat of waste. It is a decent chunk of silicon that will never get used. Ye,s they are getting the MBP buyers to pay for it. But asking folks to pay for it and they can't use it is somewhat dubious in the long run.

It's pretty common in the semiconductor industry for it to be more economical to ship high-end chips with components disabled than to have a separate design & manufacturing process for a unique "low-end" chip. You'd need to have all of Apple & TSMC's confidential information to work out the economics of leaving the ultrafusion connector on the Max - but I guess Apple have done the math.

Ultrafusion using 10K connections to hook together just two Maxes. Maybe some of those are unused. However, even at half that (5k), how could the run that many connections out the bottom of an add-in board?

Ultrafusion is for stitching together dies within a single package to make big, single processors with all-local storage, CPU and GPU cores. Even a "proper" NUMA system would use something simpler that could be used to connect between separate packages, something with plug-in cards would probably use PCIe or Thunderbolt.

Long term I doubt the Ultra and Extreme will be build from 'laptop' Maxes if Apple is anywhere near serious in keeping up in the desktop space across the zone where the upper iMac range and entry-mid Mac Pro range has been historically.

...except they've pretty much achieved that with the higher-end MacBook Pro and Mac Studio while only needing two die designs (the M1 Pro and Ultra being variations on the Max). Laptops are Apple's big seller, so a way of scaling laptop dies to higher-end desktop machines is the holy grail for them. The Studio Ultra already has the Mac Pro beat on CPU power - the sticking points are the relatively small number of users who need (a) extreme RAM capacity and (b) heavy duty AMD GPUs. Making a Xeon-W class Apple Silicon chip with PCIe for external GPUs and external RAM could be a very expensive hobby, which backpedals on the key design features of the M1/M2. If - as you suggest - Apple are prepared to throw those users under the bus, then all they need is a rackmount Studio Ultra & matching Thunderbot-PCIe enclosure so people can mount their specialist audio/video cards.

Super-powerful, big box 'o' slots machines like the Mac Pro really are the area where ARM's Unique Selling Point of high performance-per-watt is the least significant: it's vital in laptops - long battery life and good thermals, good for small-form-factor/all-in-one desktops, and good in high-density computing/server farms (lower electricity bills - which is why Amazon, Ampere etc. are already there) but high-end personal big-box 'o' slots workstations really have the least to gain: the power consumption is constrained by those big AMD Space-Heater (TM) GPUs, the size is constrained by the size of those cards and they're really not the best machines for high-density computing.