当前位置：网站首页>[problem handled] -nvidia SMI command cannot obtain the GPU process number of its own container and the external GPU process number

[problem handled] -nvidia SMI command cannot obtain the GPU process number of its own container and the external GPU process number

2022-07-01 01:29:00 【Ye Laila spicy】

Introduction ： In the deployment docker When the service is running, it is found that nvidia-smi The command cannot get the of its own container GPU Process and external GPU Process number . Verify and find some solutions .

1 Use docker Parameters pid=host This will be with the container systemctl Command conflict

docker run --rm   --privileged=true --runtime=nvidia  -e NVIDIA_DRIVER_CAPABILITIES=all    --pid=host   --gpus all nvidia/cuda:10.0-base nvidia-smi

If it is all in one The container of Need to use systemctl The command cannot be used , At present, we have encountered such a situation . Algorithm binding is behind it gpu Processed .

2 Use py3smi