Hi All. I am looking for some guidance regarding u...
# harvesterp
plain-eye-78517
10/07/2026, 6:09 AMHi All. I am looking for some guidance regarding using vGPUs on RKE2 cluster deployed in Harvester (1.8.2v). We were able to map the GPUs (nvidia L4) as vGPU to the VM. "lspci" command from the vm shows it is available. But when I try to deploy the GPU operator using helm chart (helm.ngc.nvidia.com/nvidia) , the chart is unable to deploy pod with below error related to unable to find the image. The k8s VMs OS is OEL 10.2.
"ErrImagePull: rpc error: code = NotFound desc = failed to pull and unpack image "nvcr.io/nvidia/driver:595.91.07-ol10": nvcr.io/nvidia/driver:595.91.07-ol10: not found"
I have tried to manually deploy drivers at OS level but that were giving failures too.
any idea on how to proceed further with troubleshooting or resolution?
c
clean-cpu-90380
10/07/2026, 8:33 AMAre you sure NVIDIA releases that? I can only find these tags.
Copy code
595.91.07-rhcos4.18
595.91.07-rhel10
595.91.07-rhel10.0
595.91.07-rhel10.1
595.91.07-rhel10.2
595.91.07-rhel8
595.91.07-rhel8.10
595.91.07-rhel9
595.91.07-rhel9.4
595.91.07-rhel9.6
595.91.07-rhel9.7
595.91.07-rhel9.8
595.91.07-rocky10
595.91.07-rocky8
595.91.07-rocky9
595.91.07-ubuntu22.04
595.91.07-ubuntu24.04
595.91.07-ubuntu26.04b
bland-article-62755
10/07/2026, 10:09 PMNot RKE2, but K3s... and we're doing things a little different... but let me upll up the chart/helm we're using....
bland-article-62755
10/07/2026, 10:10 PMgpu-operator:26.3.2bland-article-62755
10/07/2026, 10:11 PMConcessions: We found that this only works for us on Ubuntu 24.04 LTS
bland-article-62755
10/07/2026, 10:12 PMWe tried other Versions but landed on that for our cloud image. (26.04 is broke and so was the alma version we tried back in May. Your millage may vary)
bland-article-62755
10/07/2026, 10:14 PMIt also needed this in teh cloud-config:
Copy code
#cloud-config
package_update: true
package_upgrade: true
packages:
- qemu-guest-agent
- linux-headers-$(uname -r)
- nvidia-container-toolkit
runcmd:
- - systemctl
- enable
- --now
- qemu-guest-agent.service
- systemctl enable - --now - nvidia-persistencedp
plain-eye-78517
10/12/2026, 11:21 AMThanks for the replices. Moved forward a bit but still unable to use the GPUs. What i did was, used RHEL 10 as base OS and latest RKE2 kubernetes. Passed the previous error but encountered a different issue during GPO operator installation.
a
ancient-pizza-13099
10/15/2026, 7:55 AMwhat's the latest errors now?
or could you create a Harvester GH issue to list all required details for a better analysis. Too few information to dig more. thanks.
p
plain-eye-78517
10/26/2026, 11:06 AMWe progressed with gpu drivers deployed in guest os. Apparently nvidia restricts gpu driver images if downloaded from opensource for helm installs, even if you have licenses. You need to enable private repository and use the nvidia driver images.