This message was deleted.

this didn't fully work. It failed on the driver install so there must be more to it. I tried the default and latest drivers and neither enabled the time-sharing via the driver.

I will be tackling this next with an HP DL380a node w/ 8 NVIDIA L4 GPUs in my lab. Thanks for blazing the path. 🙏🐉

@microscopic-diamond-41007 I never fully got the labels to work. There is a file nvidia_config that is supposed to get written to for the driver install. Based on these settings.

<http://cloud.google.com/gke-gpu-sharing-strategy=time-sharing|cloud.google.com/gke-gpu-sharing-strategy=time-sharing>
<http://cloud.google.com/gke-max-shared-clients-per-gpu=20|cloud.google.com/gke-max-shared-clients-per-gpu=20>

It looks like the labels are not the only thing needed because those entries are missing. When deployed via rancher. So what I did was I deployed my primary nodes that don't have a GPU via rancher. Then I deployed the GPU nodes from GCP. Those GPU nodes will get added and in fact you can control them from Rancher after that.