How to do it, can aliyun.com/gpu-mem and nvidia.com/gpu are compatible ?

[goversion]

I hope that the two resources both aliyun.com/gpu-mem and nvidia.com/gpu can coexist in k8s system.

Currently, pods using aliyun.com/gpu-mem resources and pods using nvidia.com/gpu resources are actually applying for the same physical device GPU and do not perceive each other, that is, actually Its are not compatible.

I hope that the two will coexist. What is the idea of the community? Thanks !

[cheyang]

I hope that the two resources both aliyun.com/gpu-mem and nvidia.com/gpu can coexist in k8s system.

Currently, pods using aliyun.com/gpu-mem resources and pods using nvidia.com/gpu resources are actually applying for the same physical device GPU and do not perceive each other, that is, actually Its are not compatible.

I hope that the two will coexist. What is the idea of the community? Thanks !

I suggest to use device plugin and gpushare device plugin in the different node. They can exist in the same cluster, but not the same node.

[Svendegroote91]

@cheyang does that mean that nodes that are labeled with gpushare=true will not be schedulable with the nvidia.com/gpu resource requirement? So you have to decide which nodes (and thus GPUs) you open up to GPU sharing and which not, is that correct?