Skip to content

nvidia-k8s-device-plugin: update to v0.19.3 - #1022

Draft
Sparksssj wants to merge 1 commit into
bottlerocket-os:developfrom
Sparksssj:freshness-nvidia-device-plugin
Draft

nvidia-k8s-device-plugin: update to v0.19.3#1022
Sparksssj wants to merge 1 commit into
bottlerocket-os:developfrom
Sparksssj:freshness-nvidia-device-plugin

Conversation

@Sparksssj

@Sparksssj Sparksssj commented Aug 20, 2026

Copy link
Copy Markdown
Contributor

Description of changes:

  • Update nvidia-k8s-device-plugin to v0.19.3.

Testing done:

Built the aws-k8s-1.36-nvidia AMI against the updated kit and validated on a g4dn.xlarge GPU node.

1. NVIDIA CUDA smoke

Variant Arch Instance Passed Failed
aws-k8s-1.36-nvidia x86_64 g4dn.xlarge 429 0

GPU detected, nvidia.com/gpu scheduling works, all CUDA sample workloads (vectorAdd, immaTensorCoreGemm, reductionMultiBlockCG, …) completed.

2. OCI hooks validation

Per the device-plugin / container-toolkit testing runbook, verified the OCI hooks in the container spec for both device-list strategies:

  • CDI (device-list-strategy=cdi-cri): no prestart hook; only createContainer hooks (create-symlinks, update-ldcache, disable-device-node-modification).
  • Legacy (device-list-strategy=volume-mounts): prestart hook present (nvidia-container-runtime-hook prestart).

No new/unexpected hooks were introduced by v0.19.3 (the disable-device-node-modification hook comes from nvidia-container-toolkit v1.19, which is unchanged in this PR).

Terms of contribution:

By submitting this pull request, I agree that this contribution is dual-licensed under the terms of both the Apache License, version 2.0, and the MIT license.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant