Hello,
As far as I know A100 takes MIG (Multi-instance GPU) and RTX 3090 does not support MIG.
So, I am wondering Amere (i.e., RTX 3090) supports Multi-process Service (MPS).
Thanks,
Hello,
As far as I know A100 takes MIG (Multi-instance GPU) and RTX 3090 does not support MIG.
So, I am wondering Amere (i.e., RTX 3090) supports Multi-process Service (MPS).
Thanks,
Looking at the docs, “MPS requires a GPU with compute capability version 3.5 or higher.”, so you should qualify:
I am wondering whether it supports hardware-based MPS.
I know hardware-based MPS works for Volta and Turing, but not sure ampere, especially RTX 3090.
I admit I am speaking from a position of ignorance, but my take from Robert’s comment here Hyper-Q technology is that all hardware since 3.5 has hardware features for MPS use.
Edit: I see your point re the extra MPS hardware features for Volta and Turing, but looking at the Turing arch doc https://developer.nvidia.com/blog/nvidia-turing-architecture-in-depth/, this only applies to the Tesla versions, so perhaps this is just a Tesla (Quadro?) extra?