I have been trying to run concurrent dl4j process on nvidia 3060 GPU on ubuntu 20.04 with cuda 11.2 and cudnn 8 using navidia-cuda-mps-control. however, each time i start nvidia-cuda-mps-control and attempt to run 2 dl4j processes, both processes are getting terminated. Does dl4j support nvidia mps?
@ajmakoni Make sure you’re using snapshots for cuda 11.2 support. https://deeplearning4j.konduit.ai/config/config-snapshots
We’ll be publishing a new release within a few weeks that will cover cuda 11.2 otherwise.
As far as I’m aware we haven’t tried it yet. But for the 30 series gpus you’ll need to use snapshots, as @agibsonccc has said.
If it still doesn’t work with that, can you please post the actual output you are seeing?