I would like to perform EUD testing Extended Utility Diagnostics (EUD) — NVIDIA DCGM Documentation latest documentation
However, I couldn’t find the package to install nvidia-diagnostic DCGM Diagnostics — NVIDIA DCGM Documentation latest documentation
Requesting to share link to download nvidia-diagnostic for ubuntu 22.04, current environment has nvidia-driver-535
Sharing our actual workflow for obtaining and deploying EUD packages in an air-gapped environment
We ran into the exact same issue during GB300 platform validation — the nvidia-diagnostic packages aren’t reachable through standard channels in an air-gapped setup, and /usr/share/nvidia/diagnostic/ doesn’t get created automatically either.
Here’s the three-step workflow that worked for us:
1. Extract the nested package
The EUD package itself is wrapped inside a local-tegra-repo .deb structure, so you need to extract the inner repo first, then extract the actual binaries from there:
mkdir -p ./eud_temp && dpkg-deb -x nvidia-diagnostic-local-tegra-repo-ubuntu2404-580.126.12-mode1_1.0-1_arm64.deb ./eud_temp
cd ./eud_temp/var/nvidia-diagnostic-local-tegra-repo-ubuntu2404-580.126.12-mode1/
mkdir -p ./binary && dpkg-deb -x nvidia-diagnostic-580_580.126.12-1_arm64.deb ./binary
2. Manually create the target directory and inject the files
This directory doesn’t exist by default, so you have to create it yourself:
sudo mkdir -p /usr/share/nvidia/diagnostic
cd ./binary/usr/share/nvidia/diagnostic && sudo cp -r ./* /usr/share/nvidia/diagnostic/
3. Confirm MIG mode is disabled before running the deep diagnostic
The EUD deep stress test (dcgmi diag -r eud) needs full, unpartitioned GPU access:
sudo nvidia-smi -mig 0
sudo dcgmi diag -r eud -p "eud.passthrough_args='gpu_diag'"
This completely bypassed the installation issue in our environment, and the EUD deep test itself runs for over 10 hours — it’s what we use as the formal basis for hardware RMA determination.
Hope this helps anyone running into the same situation.