I think the only thing the CPU does in this card is to store data from the high speed connections the card has, read it and pass it to the VRAM (and the other way around when gathering results from the GPU).
I think that particular CPU is more than enough since it doesn't really need to crunch numbers or similar heavy computations.
Yes. We have done pieces on the BlueField-2 DPUs running Ubuntu and doing things like running ZFS and iSCSI off of the DPU's Arm cores as well. This is the BlueField-3 base, so a faster Arm core complex and more memory bandwidth.
I think for most people, that's called a "GPU", and the programs they execute are called "Shaders", which are what calculates the lighting, shadows, 3d matrix manipulations, of video games.
Probably only be able to fit 2 similar cards in a Mac Pro with the needed cooling. But that could seriously add up total specs with how integrated their chips are.
Space-wise, the Mac Pro can fit 4 double-slot cards (which fits all 3 models). Hardware-wise, the most immediate problem would be powering another 230-1400w of power. The other big issue would be bandwidth - the AX800 is a PCIe 5.0 card, and the Mac Pro chassis only supports PCIe 3.0 busses last I checked.
Frankly, I wouldn't bet on miraculous results adding Nvidia cards to Apple silicon anyways. Apple's compute hardware focuses very hard on saturating on-chip resources with bandwith, while providing comparatively small IO bandwidth. It's not a bad strategy for mobile hardware, but it can struggle to keep up with bespoke distributed systems.
19 comments
[ 2.8 ms ] story [ 61.2 ms ] threadI think that particular CPU is more than enough since it doesn't really need to crunch numbers or similar heavy computations.
https://turingpi.com/product/turing-pi-2/
I would love to have some real computers inside my computer, preferably able to things my box cant or just do it faster.
Frankly, I wouldn't bet on miraculous results adding Nvidia cards to Apple silicon anyways. Apple's compute hardware focuses very hard on saturating on-chip resources with bandwith, while providing comparatively small IO bandwidth. It's not a bad strategy for mobile hardware, but it can struggle to keep up with bespoke distributed systems.