GitHub user sueun-dev added a comment to the discussion: Gpu on Kubernetes Cluster
Check whether those GPU offerings are customized. In the [4.23.0.0 picker](https://github.com/apache/cloudstack/blob/4.23.0.0/ui/src/views/compute/CreateKubernetesCluster.vue#L726-L745), `iscustomized` must be `false`, and `cpunumber`/`memory` must meet the selected Kubernetes version's minimums. The [backend](https://github.com/apache/cloudstack/blob/4.23.0.0/plugins/integrations/kubernetes-service/src/main/java/com/cloud/kubernetes/cluster/KubernetesClusterManagerImpl.java#L648-L661) also rejects custom service offerings. The blank CPU/RAM columns in your screenshot make this worth checking. I ran the unchanged selection method from 4.22.0.0 and 4.23.0.0 in isolated Node with synthetic API responses. It excluded customized GPU offerings and retained fixed-size GPU offerings meeting the minimums. This tests the picker logic, not GPU passthrough or a Kubernetes deployment. If yours have `iscustomized: true`, create a separate GPU offering with fixed CPU/RAM satisfying the CKS/version minimums, then recheck the picker. Which CloudStack release are you running, and what do `iscustomized`, `cpunumber`, and `memory` show for those two offerings in [`listServiceOfferings`](https://cloudstack.apache.org/api/apidocs-4.22/apis/listServiceOfferings.html)? GitHub link: https://github.com/apache/cloudstack/discussions/14264#discussioncomment-18659986 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
