The Mac Pro still has auxiliary power connectors, so you could theoretically install a GPU in it, just not the highest end ones. No drivers so you can't really do anything with them. I do suppose Nvidia could maybe do CUDA drivers for a card, but would Apple even allow it?
@paul could they be used for virtualization? PCIe passthrough to Linux and windows guests?
@paul The two SATA ports are also a little confusingly still there - couldn’t slap some NVMe ones? Guess you could put them on an expansion card…
@ckelley87 that's just copy/paste from the 2019, probably uses the same module which I think is fine.
@paul WTH, How many watts do GPUs take now?? I haven't bought a separate GPU for nearly 20 years…
@pixel @paul My new 4090 came with a power plant.
@steveriggins Holy crap. some f the 4090s need 1200 watts. TWELVE HUNDRED. https://www.tomshardware.com/news/1200w-power-requirement-rtx-4090
Some RTX 4090 Graphics Cards Recommend a 1200W Power Supply

That might be overkill, even for an RTX 4090

Tom's Hardware
@pixel My falcon came with a 1000w PS for the 4090 and 13900ks
@steveriggins At some point (probably already) people will have to install a new circuit too.
@pixel no joke my ceiling leds flicker once when I power on lol
@pixel well, there goes my fantasy of a new MacPro loaded with extra A100s.

@Dhmspector @pixel My understanding is that the new Mac Pro doesn’t support discrete GPUs.

I’d love to be wrong about this, though.

@jeff been the case for a long while… but I can dream.
@pixel @paul Same, last time I assembled a PC was in 2004, and then a few years later I switched to Macs - I'm now looking for a Windows computer for my sister, and apparently the GPU is now half the cost of the whole box 😐

@paul You could theoretically have said the same about thunderbolt eGPU’s on M1 machines. Marcan had posted back on that old social network a couple years ago that the actual problem was related to the base addresses available on the PCIe lanes for hardware and at least in Linux's case too much stuff was hard-coded to rely on how it works elsewhere. (Warning: This is my understanding of what he said)

I had asked a few months ago if that changed for M2 and he had not looked into it yet.

@paul they might let AMD, but I doubt they’ve gotten over their issues with NVIDIA
@sladewatkins @paul Craig did say during John Gruber’s event that Nvidia does a great job on AI and it isn’t something Apple is focused on. Sort of opens the door for Nvidia in non-Metal roles on the Mac…
@paul IIRC it's entirely a limitation of the apple silicone preventing the OS using additional GPUs. I think virtualization pass through as others mentioned might be the only use.

@paul

They really should make nice with NVidia just to get cuda running on Macs.

@paul Didn’t Nvidia do that a few years ago?IIRC they made a driver for Macs so their GPUs could be used in eGPU enclosures, only for Apple to refuse to sign it and welp
@paul The number of people this machine makes sense for is not zero, but I bet they could all go to dinner and not have to push more than two tables together.
@paul I almost wonder if driverkit allows that without apple’s intervention.
@paul what's even the point of that if you can't put a normal GPU in? i would love to switch back to macOS as my workstation, but the M series processors just suck when it comes to 3D rendering
@paul I still think apple is working on their own gpu card, there were rumors about it. it just wasn't ready. the whole mac pro wasn't ready cause it’s not bringing much to table you can't just get with a studio and good thunderbolt 4 station for less. I really am just gonna wait for m3 for both 3nm and to see what they do with more time. if by then they still haven't given me a reason to care about pro, i'll get m3 studio.
@paul is Windows/PC still the way to go if you want raw GPU compute? I haven't kept up on the GPU side of things for the M2 Ultra benchmarks.