Sign inSign up

bialekamir/nvidia-k8s-device-plugin-multi-gpu

By bialekamir

•Updated almost 2 years ago

enabling applications on k8s to access specific GPUs through proper configuration.

Image
Integration & delivery
0

211

bialekamir/nvidia-k8s-device-plugin-multi-gpu repository overview

⁠nvidia-k8s-device-plugin-multi-gpu

⁠About

base on NVIDIA Device Plugin, this image supports multi-GPU nodes. The final result allows you to have multiple GPUs on a single node, enabling applications to access specific GPUs through proper configuration.

⁠Instructions

  • Deployment in k8s should be done with Nvidia GPU Operator helm chart, but adjust the image ref in values.yaml:
devicePlugin:
  repository: bialekamir
  image: bialekamir/nvidia-k8s-device-plugin-multi-gpu
  version: 0.17.0

Make sure the chart version match the docker version. example: For image v0.17.0 , the chart version should be GPU Operator v24.9.0

  • To deploy as docker:
  1. Create a config.yaml file with the following:
 version: v1
 flags:
   migStrategy: none
 sharing:
   timeSlicing:
     resources:
       - name: "nvidia.com/rtx4090"
         replicas: 5
       - name: "nvidia.com/rtx3090"
         replicas: 2
 resources:
   gpus:
     - pattern: "GPU-.*4090.*"  # Matches RTX 4090 GPUs
       name: "nvidia.com/rtx4090"
     - pattern: "GPU-.*3090.*"  # Matches RTX 3090 GPUs
       name: "nvidia.com/rtx3090"


  1. Run the image with the config as volume:
docker run --rm   --name nvidia-device-plugin-working   --privileged   --security-opt=no-new-privileges   -v /var/lib/kubelet/device-plugins:/var/lib/kubelet/device-plugins   -v  $(pwd)/config.yaml:/config/config.yaml  bialekamir/nvidia-k8s-device-plugin-multi-gpu:0.17.0   --config-file /config/config.yaml

Tag summary

Content type

Image

Digest

sha256:a63bc3431…

Size

176.2 MB

Last updated

almost 2 years ago

docker pull bialekamir/nvidia-k8s-device-plugin-multi-gpu:0.17.0