Hi all,
I have OpenShift 4.8.35 with NVIDIA GPU Operator 1.9.1 installed on VMware vSphere and I would like to configure the vGPU setup.
The ESX had A100 with 40GB installed and the drivers installed.
We have configured a new VM with profile NVIDIA GRID vGPU grid_a100-7-40c and joined that VM into OpenShift cluster.
I have installed the NFD and NVIDIA GPU Operator.
When the Operator tries to install the driver, the module can not be loaded because if fails due to the Device seems not to be supported by the driver in version 470.
But I think there are further steps needed for the vGPU setup.
According to NVIDIA vGPU I have to build my own driver container image.
I have cloned the mentioned repo, setup the env vars, but when I try to build the docker image, I get the error:
# pwd
/root/driver/rhel8
# docker build --build-arg DRIVER_TYPE=vgpu --build-arg DRIVER_VERSION=$VGPU_DRIVER_VERSION -t ${PRIVATE_REGISTRY}/driver:${VERSION}-${OS_TAG} .
Sending build context to Docker daemon 336.2 MB
Step 1/51 : ARG CUDA_VERSION
Please provide a source image with `from` prior to commit
Have I to set up a CUDA VERSION as build arg?
But in which verison and why that is not mentioned in to detailed workflow?
Thanks guys