Hi Oliver, Thank you so much for your reply! Please make a blog :) I am very interested to see. On Tue, Jan 7, 2025 at 10:41 AM Oliver Weinmann <oliver.weinmann@me.com> wrote:
Hi Sathish,
I have this running with Magnum since a couple of months now. What I can say is that it is working. Magnum itself is not in any way aware of the GPUs.
You basically just need to set up GPU passthrough on at least one compute node, have a flavor with the GPU and add an additional node group to your k8s clusters with this flavor. It works very well when using CAPI (Ubuntu) since you can simply deploy NVIDIA GPU Operator HELM chart. This automatically detects your NVIDIA GPUs, installs drivers etc. and simply works. I also managed to get it working with FCOS but this involves some more manual work, since the NVIDIA GPU Operator HELM chart is not 100% compatible with FCOS. You need to manually install the nvidia GPU driver in the running worker node. At least that was the state back in summer 2024. Maybe it is compatible now.
I planned to write a blog post from my personal notes, but it is not finished yet.
Cheers,
Oliver
Am 07.01.2025 um 16:07 schrieb Satish Patel:
Folks,
We are running magnum with openstack to spin up k8s clusters and everything works as expected. Now I want GPU support so we can do some AI/ML stuff with k8s. Does magnum support GPU and how does it work?
In the past I used GPU in passthrough mode to expose hardware to VMs but not sure how it works for k8s and what are the challenges?