Summary
gpu-operator/install-gpu-operator-vgpu.rst presents building the vGPU guest driver container from the licensing-portal ZIP as the only way to get a driver image, and never says who that workflow is for. NVIDIA AI Enterprise customers have a prebuilt vGPU guest driver image on NGC (nvcr.io/nvaie/vgpu-guest-driver), and the page should say that so they do not build an image they could have pulled.
This request has come up more than once recently, including from a customer who read the page as mandatory.
Current text
https://github.com/NVIDIA/cloud-native-docs/blob/db636fa/gpu-operator/install-gpu-operator-vgpu.rst#L67-L68
The prerequisites make Git, Docker, and a private registry unconditional:
* You have access to a private registry such as NVIDIA NGC Private Registry and can push container images to the registry.
* Git and Docker are required to build the vGPU driver image from source repository and push to the private registry.
https://github.com/NVIDIA/cloud-native-docs/blob/db636fa/gpu-operator/install-gpu-operator-vgpu.rst#L97-L101
"Build the Driver Container" opens with no scoping statement, and nothing on the page mentions NVIDIA AI Enterprise, NGC entitlement, or the prebuilt image.
Elsewhere the docs already show the prebuilt image being used, so the two pages disagree on whether building is required:
Suggested change
- At the top of "Build the Driver Container" (or in the prerequisites), state that building the driver container is for customers who have a vGPU entitlement but not NVIDIA AI Enterprise, and that NVIDIA AI Enterprise customers can instead pull the prebuilt
vgpu-guest-driver image from the nvaie NGC org and skip the build and download steps.
- Cross-reference :ref:
nvaie-ocp (and the equivalent guidance for non-OpenShift clusters) for how to set driver.repository / driver.image / driver.version and the NGC pull secret when using the prebuilt image.
- Make the Git/Docker/private-registry prerequisites conditional on the build path.
Summary
gpu-operator/install-gpu-operator-vgpu.rstpresents building the vGPU guest driver container from the licensing-portal ZIP as the only way to get a driver image, and never says who that workflow is for. NVIDIA AI Enterprise customers have a prebuilt vGPU guest driver image on NGC (nvcr.io/nvaie/vgpu-guest-driver), and the page should say that so they do not build an image they could have pulled.This request has come up more than once recently, including from a customer who read the page as mandatory.
Current text
https://github.com/NVIDIA/cloud-native-docs/blob/db636fa/gpu-operator/install-gpu-operator-vgpu.rst#L67-L68
The prerequisites make Git, Docker, and a private registry unconditional:
https://github.com/NVIDIA/cloud-native-docs/blob/db636fa/gpu-operator/install-gpu-operator-vgpu.rst#L97-L101
"Build the Driver Container" opens with no scoping statement, and nothing on the page mentions NVIDIA AI Enterprise, NGC entitlement, or the prebuilt image.
Elsewhere the docs already show the prebuilt image being used, so the two pages disagree on whether building is required:
vgpu-guest-driverimages fromnvcr.io/nvaie)nvcr.io/nvaie/vgpu-guest-driver:...in sample output)Suggested change
vgpu-guest-driverimage from thenvaieNGC org and skip the build and download steps.nvaie-ocp(and the equivalent guidance for non-OpenShift clusters) for how to setdriver.repository/driver.image/driver.versionand the NGC pull secret when using the prebuilt image.