TL:DR I’m not sure if this is an Ubuntu problem or an Nvidia problem. My configuration lasts exactly once and doesn’t stay after a reboot.
I see release dates are jiving with the testing I’m doing so I’d hope to contribute something along the way. Part of my difficulty is chasing these behaviors with “black box” testing. I’ve never debugged drivers, so this part is new. The last several weeks have seen several updates to this range of drivers and I’ve tried them all with the same results.
Ubuntu 22.04 LTS
Nvidia 3070 card (8GB)
510-545 drivers all do the following behaviors.
v545 finally worked but this underlying issue has been a problem for weeks. Whenever I get a stable installation, it lasts only once. Why?
Odd details:
When it crashes, I’m seeing what looks like artifacts on the bootup log screen. It looks to be trying to put up a rest API bubble or two before the graphics are done loading. When this happens, bootup gets stuck in a loop trying to start endless sessions. Maybe something was left in there during testing or it’s on my end? I don’t know.
When I’m not getting the above bubble artifacts, I get a black screen or green bars and dots and have to start over.
I’ve tried both ubuntu installs and nvidia drivers and the cuda toolkit. I know how to clean up the libs and lacklist nouveau. This last round of testing was the first time 545 and 12.3 CUDA has worked for me. I had to mix the ubuntu autoinstall with the Nvidia CUDA install to make that work.
If I could get this to just stick, I’d be plenty happy and can move on studying AI.