I don’t get the negative comments here. This is the c/LocalLLaMa community, and this post is about adding 16 GB of VRAM for fairly cheap. This feels perfectly useful and appropriate.
For reference, the GPU used is the Tesla V100 SXM2 16GB. The author adds that:
The V100 SXM2 is not the only option. The P40 gives you 24GB for similar money, though it is slower and has no Tensor Cores. The V100 32GB variant costs more but still undercuts any consumer GPU with that much VRAM.
AI haters ventured outside of the Fuck_AI community and the outside world confused, bewildered and angered them.
Think of cavemen emerging from stasis, seeing airplanes up in the sky, then throwing rocks up at the terrifying sky demons, and the falling stones hitting people on the ground.
The digital equivalent is happening here.
Slightly misleading title imo. Like yeah you put it “inside” a gaming PC but its not in any way usable for that. Its still just a datacenter “GPU”
I’m not the author, I shared the title as is. You’re not wrong though
Are you the author ? Anyway, thanks for sharing, that was a very interesting read
not the author, just thought it’s interesting
Im expecting there might be masses of gpus being sold for cheap once the ai bubble pops.
I hope someone turns these adapter boards into a consumer grade product, once that happens. Certainly has potential. The question is: can you game on these?
You can do CUDA on these, which is not nothing. With that you can run DOOM and pipe it to a video out. If there is a glut of cheap new versions of these after a bubble pop, perhaps the effort will be made to bodge these into pretending they’re a 5090 or something for the drivers, probably on Linux. It’s certainly possible, but significant work. If they’re cheap enough, and plentiful enough, life will find a way.
What they are good at, right now, is running local LLMs, scientific computing, etc., and it is done reasonably commonly by hobbyists. Likely also Photoshop and similar if you want the pain of running them on windows.
A while back I was able to use an nvidia m40 as a GPU while using embedded intel as the primary video output. This worked back then because laptops with nvidia GPUs rely on this hybrid GPu setup built into windows to conserve power and optimize for apps that actually need acceleration (windows surface book had a detachable nvidia GPu built into the keyboard). I did have to tweak some things in the driver to pick the default GPu. I don’t remember exactly how I did it back then and I’m not sure if it’s relevant any longer. I’m sure it’s better now that USB-C docks provide acceleration as well without needing to be connecting directly to the output device.
I’m sure anything and all of these datacenter systems will eventually find creative places in the aftermarket.
I put a brick in my gaming PC. Applause, please.
