You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'd add another item to your list: Simplify GPU plugin deployment options.
I would vote to only have two examples:
NFD + GPU-plugin with shared-dev-num=1 and monitoring
Basic use case, should work for most
NFD + GPU-plugin with shared-dev-num>1, resource management, monitoring and extended resources
GAS use case, for those who need it
Then I'd add notes about configuration options, using shared-dev-num without GAS etc. into a different file (advanced-deployment.md or similar). As you say, the current README is pretty long.
IMHO "using shared-dev-num without GAS" should be documented as being for "dedicated cluster/nodes with a single GPU workload, where share count equals to how many instances of that workload fit into a single GPU (with required QoS)".
AFAIK other uses for it are non-production ones, so they do not need to be mentioned.
GPU plugin documentation is missing following things:
integpu-job.yaml
referred inREADME.md
Testing and Demos section: https://github.com/intel/intel-device-plugins-for-kubernetes/blob/main/demo/intelgpu-job.yamlThe text was updated successfully, but these errors were encountered: