We are building a workstation that uses 4 RTX 2080ti. When running some tests using the software GPU Burn ( GitHub - knorth55/gpu-burn: Forked version to run on CUDA9 and 10 or original version GitHub - wilicc/gpu-burn: Multi-GPU CUDA stress test ) to push the GPU to 100% in order to do some temperature and sound tests, we noticed some repeated beeps coming from the GPU.
I tried to investigate by checking the system logs but could not find any error and gpu_burn does not report any error. All the work is done successfully (GPU running at 100% without any crash).
The system used is Ubuntu 18.04.2, up-to-date with NVidia drivers 418 and CUDA 10.0 installed from Nvidia’ repository.
The kernel is 5.0.0-17-generic provided by the linux-generic-hwe-18.04-edge package.
The graphic chipsets are all RTX 2080 ti Turing : tu102, using PALIT GeForce RTX™ 2080 Ti GamingPro OC.
The sound is a high-pitched beep, but in a low volume that repeat every second like a wave. All cards seems to be beeping at the same time, so it is hard to define which one is beeping exactly. However, it seems that all 4 cards are beeping continuously.
The sound disappears as soon as we stop the software and the GPU is no longer running at 100% (goes back to 0% with low work and normal state).
The issue happens both when running the program from gnome-shell’ terminal and from a TTY without any X server started.
I was not able to run
startx -- -logverbose 6 as requested (no idea how to run gnome this way).
Also, if NVidia as any software/benchmarking tool it recommends for pushing the GPU to its maximum, I’d prefer to use it rather than a soft on Github. However, I was not able to find anything like this on the website (whether it is in a docker image of as a independent software).
nvidia-bug-report.log.gz (2.7 MB)