[BUG] GPU-PV broken for Windows 11 24H2
- Dominant language
- PowerShell
- Stars
- 66
- Forks
- 50
- PR merge metrics
- No merged PRs in 30d
Description
**Describe the bug**
We're using the GPU-PV functionality of AKS-Edge and this has been working great for us so far. However after updating one of our machines to Windows 11 24H2 (coming from 23H2) the nvidia-device-plugin no longer seems to be working. It fails to start with the error:
Error: failed to create containerd task: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: error during container init: error running hook #0: error running hook: exit status 1, stdout: , stderr: Auto-detected mode as 'legacy' nvidia-container-cli: initialization error: nvml error: unable to load the nvml library: unknown
**To Reproduce**
Steps to reproduce the behavior:
1. Install Windows 11 24H2
2. Follow the GPU acceleration guide from: https://learn.microsoft.com/en-us/azure/aks/aksarc/aks-edge-gpu
3. observe the error "unable to load the nvml library"
**Environment (please complete the following information):**
- AKS Edge Essentials Version: 1.9.262.0
- Kubernetes version 1.29.6
- Windows Host OS
- Edition: Professional
- Version: 24H2 build 26100.3476
- NVIDIA RTX A5000
- NVIDIA driver 572.83
Contributor guide
Research direction
Start with the GPU acceleration guide and the nvidia-device-plugin setup described in the issue, then compare behavior between Windows 11 23H2 and 24H2 with the listed driver and AKS Edge versions. Done means the plugin starts successfully on Windows 11 24H2 without the NVML library initialization error.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- kubernetes
- Domain
- infrastructure, operating-systems
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100