I am using the AGX Orin 64 GB version, and have been trying to reflash it to Ubuntu 22.04 using the following guide:
I used an external USB stick to reflash and have basically completely followed these instructions. Unfortunately, I get a host of GPU errors when running some of the test commands at the end.
When I run nvidia-smi, I get:
Failed to initialize NVML: Driver/library version mismatch
NVML library version: 540.03
My understanding is nvidia-smi may not be supported on Jetson devices, so that’s not as worrying.
However, when I run the CUDA Device Query I get:
CUDA Device Query (Runtime API) version (CUDART static linking)
cudaGetDeviceCount returned 801
→ operation not supported
Result = FAIL
and when I run the sample algorithm selector, I get:
&&&& RUNNING TensorRT.sample_algorithm_selector [TensorRT v8602] # ./sample_algorithm_selector
[06/17/2024-21:50:44] [I] Building and running a GPU inference engine for MNIST.
[06/17/2024-21:50:44] [I] Writing Algorithm Cache for MNIST.
[06/17/2024-21:50:44] [W] [TRT] Unable to determine GPU memory usage
[06/17/2024-21:50:44] [W] [TRT] Unable to determine GPU memory usage
[06/17/2024-21:50:44] [I] [TRT] [MemUsageChange] Init CUDA: CPU +0, GPU +0, now: CPU 20, GPU 0 (MiB)
[06/17/2024-21:50:44] [W] [TRT] CUDA initialization failure with error: 801. Please check your CUDA installation: 1. Introduction — Installation Guide for Linux 12.5 documentation
&&&& FAILED TensorRT.sample_algorithm_selector [TensorRT v8602] # ./sample_algorithm_selector
I’m not sure what has gone wrong in the CUDA set up here, I followed the exact process and commands suggested in the doc, and have rebooted my system several times.
Wondering what I can do to fix this, or if I should just reflash following a different set of instructions, like the one here: Jetson Linux 35.3.1 | NVIDIA Developer