CUDA memory usage by process in Orin/DriveOS

Autonomous Vehicles DRIVE AGX Orin DRIVE AGX Orin General
1

I’m trying to obtain CUDA memory usage per process on Orin. The issues I face are:

  1. There is no nvidia-smi.
  2. To work around (1), I try to write a Python program using py3nvml.py3nvml, but it fails due to lack of libnvidia-ml.so.

When I tried to find an arm version of libnvidia-ml.so, all I could find in various distributions (ex. CUDA for SBSA) was a stub. At the same time, x86 distributions do contain the actual implementaion.

Could someone please help me find the arm version of that file, or suggest a different way of getting per-process GPU memory utilization?

Thank you.

Software Version
DRIVE OS 6.0.8.1

Target Operating System
Linux

Hardware Platform
DRIVE AGX Orin Developer Kit (not sure its number)

Host Machine Version
native Ubuntu Linux 20.04 Host installed with DRIVE OS Docker Containers

3

It is removed from DRIVE OS 6.0.8.1 package.
Could you check PyTorch + CUDA11.4 on 6.0.8.1 - #6 by SivaRamaKrishnaNV if it helps?

5

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.