gpu_docker_tensorflow_mat_multi

GPU Docker based TensorFlow Matrix Multiplication

Using a single GPU on a multi-GPU system

Requirements:

• Nvidia drivers in the directory /var/lib/nvidia on host node

Usage:

$ docker build -t gpu_docker_tf_mat_multi .
$ docker run -it  --privileged -v /var/lib/nvidia:/usr/local/nvidia/lib64 gpu_docker_tf_mat_multi

• --privileged makes the docker container can access the GPUs on host node
• -v /var/lib/nvidia:/usr/local/nvidia/lib64 map the Nvidia dynamic libraries on host to docker container

Node Affinity

• Scheduling GPUs
• Assigning Pods to Nodes
• affinity exmaples

reference

Using GPUs