Hi, I’m going to create an inference/training workstation. I heard somewhere that Tesla P100 will be better than Tesla P40 for training, but the situation is the opposite for output. Does it make sense to create a workstation with two variants of video cards, or will only P40 or P100 be enough? And why? (I’m going to build a workstation with 4 graphics cards in a year).
And one more question: P40 supports CUDA 6.1 and P100 supports 6.0, but now torch uses CUDA 11.8/12.1… Will this cause a problem? I’ve seen people run LLM on P40, but because of the CUDA situation i don’t understand how it works at all(
Forgot to mention, what are the processor requirements? I won’t be able to provide a big budget for the build, so I’m looking at inexpensive processors like EPYC 7302P (it seems to be able to support 4 video cards).