0

I have 22.04 LTS, and an nvidia 4070 card. I installed the driver via Cuda toolkit v12.3, and blacklisted nouveau (which doesn't even fully work with 4070 still). When I suspend the system, and try to resume, I get the following errors:

[41.212199] NVRM nbsiReadRegistryDword: osReadRegistryDword called in Sleep path can cause excess delays!
[41.212201] NVRM nvAssertFailedNoLog: Assertion failed: 0 @ nbsi_osrg.c:107

which repeat a bunch of times. And then the system doesn't resume and I have to force reboot the computer. How can I fix this?

sodiumnitrate
  • 291

1 Answers1

0

I'm running into similar issues with different nvidia cards, drivers and CUDA Toolkit versions. Next approach I'll try is this:

sudo rmmod nvidia_uvm
sudo modprobe nvidia_uvm

Have you had any luck with the answers to these similar questions?

CUDA not working after returning laptop from sleep

How do I fix CUDA breaking after suspend?

Weid cuda driver state after forgetting to clean up resources before suspending the computer

smcnally
  • 396
  • Removing and re-probing nvidia_uvm works to address CUDA errors after resume. Details here https://askubuntu.com/a/1506973/941887 – smcnally Mar 08 '24 at 21:35