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
Not sure sure if i'm doing something wrong or theres an easier workaround for this.
I'm using the nvidia gpu operator in a very vanilla config.
This results in the generation of several runtimeClass resources.
In this configuration, where one component of the nvidia gpu operator is handling the driver instead of me manually installing the driver on the host, it seems that pods which want to have access to the GPU require runtimeClassName: nvidia to be in the pod spec.
I'm not sure if I'm missing it or not but I dont see a way to specify runtimeClassName in the CRD for a model. Thanks for all the hard work! I'd really like to use your project!
The text was updated successfully, but these errors were encountered:
Hi, thanks for the issue. And thanks for the in-detailed statement.
Indeed, after a brief read through documentations, I found runtimeClass requires extra parameters with no workaround (retricted by immutable of Pods, too, so no directly modifications are allowed.)
Not sure sure if i'm doing something wrong or theres an easier workaround for this.
I'm using the nvidia gpu operator in a very vanilla config.
This results in the generation of several
runtimeClass
resources.In this configuration, where one component of the nvidia gpu operator is handling the driver instead of me manually installing the driver on the host, it seems that pods which want to have access to the GPU require
runtimeClassName: nvidia
to be in the pod spec.I'm not sure if I'm missing it or not but I dont see a way to specify
runtimeClassName
in the CRD for a model. Thanks for all the hard work! I'd really like to use your project!The text was updated successfully, but these errors were encountered: