Inconsistent RHEL9 package naming breaks precompiled build for driver version 580.65.06

[radikaled]

Unfortunately, it looks like an inconsistency with the package naming for:

• nvidia-fabric-manager
• libnvidia-nscq

Breaks precompiled builds for OpenShift (RHEL9) when specifying driver version 580.65.06.

For example:

580.65.06
https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/nvidia-fabricmanager-580.65.06-1.x86_64.rpm
https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/libnvidia-nscq-580.65.06-1.x86_64.rpm

• nvidia-fabric-manager is now formatted as nvidia-fabricmanager
• libnvidia-nscq does not have the driver branch (i.e., 580) in the version string. Ideally, it would be: libnvidia-nscq-580-580.65.06-1.x86_64.rpm

In comparison to:

575.57.08
https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/nvidia-fabric-manager-575.57.08-1.x86_64.rpm
https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/libnvidia-nscq-575-575.57.08-1.x86_64.rpm

The Dockerfile currently is unable to account for this discrepancy:

gpu-driver-container/rhel9/precompiled/Dockerfile

Line 163 in 407b170

dnf install -y nvidia-fabric-manager-${DRIVER_VERSION} libnvidia-nscq-${DRIVER_BRANCH}-${DRIVER_VERSION} ; \

I hesitate to open a pull request in case this is an oversight with the package naming. In the meantime, it is easy enough to work around the issue by modifying the string formatting in the Dockerfile.

Cheers,

[julio-matarranz]

Dear,
It would be nice that someone backport those commits to precompiled rhel9 :

• 78fa4ac
• b8f5eaa