Time-Slicing GPUs

Is it possible Time-Slicing GPUs on Charmed Kubeflow? I have 2 servers with NVIDIA GeForce RTX 4090. How can I configure Time-Slicing GPUs? kubeflow 1.7/stable

Hi Petr. I’m not aware of this use case specifically. I’ve raised the question with the engineering team on our Mattermost channel. Join us there to continue the chat: LINK TO CHAT.

Hi Petr. We tested our solution on FGX hardware some time ago. As for time-sharing configuration question. It should work but we have never tested it before. Feel free to play around. First thing you need to is install microk8s on your machine with gpu addon https://microk8s.io/docs/nvidia-dgx. Next you may have to set some cluster wide configuration (as mentioned here https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/gpu-sharing.html#applying-one-cluster-wide-configuration). As said before, I have never tested it. Let me know how it goes.

Hi @bhandalc @michalhucko. Feel free to contact me if GPU Fraction is important for you My email - > kirson@run.ai

Hi Peter.

You can try this. Nvidia MPS

It works.

Thanks for the information. I’ll try to figure it out, unfortunately my skills are not very high. I’ll start experimenting after the 11th, so as not to drop everything for the holidays.