I have a k8s cluster, an ubuntu host as the master node, NX as the working node, when I use the cluster to call NX’s GPU resources, I successfully deployed nvidia-device-plugin on the master node, but it cannot detect NX’s GPU.
Most GPUs that you know of are discrete GPUs (dGPU) which are on the PCI bus. Tools to detect and query are for PCI. Jetsons have an integrated GPU (iGPU) which is wired directly to the memory controller, and so this is likely why your tools do not see the Jetson.
The tool nvidia-smi is what mediates the PCI dGPU detection. You can find your release of L4T with “head -n 1 /etc/nv_tegra_release”, and if you have L4T R36.x, then you have a “minimal” version of this. However, it isn’t until you reach Orin that L4T R36.x is usable. I don’t know if there is any alternative way to fix that, but I’m guessing you are out of luck if not using Orin (and if you do use Orin I don’t think it is a guarantee).
Maybe someone knows of a way to adapt a program to run with the iGPU. I’ve not heard of a tutorial for this, but it would be a great idea if there were some sort of tutorial on how to modify a program expecting PCI GPUs to work with the iGPUs.
Since you have (on the TX2, but maybe or maybe not on the NX) L4T R32.7.5, then any documents or software would come from R32.7.5 here (adjust for differences on the NX if it isn’t R32.7.5):
https://developer.nvidia.com/linux-tegra
Someone else (e.g., maybe @dusty_nv ) might have some advice on how to add detection of the Jetson GPU (I’ve never tried to adapt this).
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.