[396.24] Duel GPU - black screen crash and gpu fans go 100%

I have a 980pro and 1080ti that I use for rendering. On Windows 10 there are no issues, however I get an issue on Linux after an hour of two of use.

Almost randomly after an hour (under little activity other than browsing), the screen will go black (linux crashes) and the GPU fans are running at full speed. I can reset the computer without issue, but the problem happens again after an hour or so. The issue does not happen if I only have the 1080ti active, and remove the 980pro from the motherboard.

Fedora 28 - Linux 4.16.8
Nvidia: 396.24
850watt platinum psu
Ryzen 2700x
(non sli) 980pro and 1080ti with the later primary with screen out
x470 Gaming 7

nvidia-bug-report.log.gz (126 KB)

I also couldn’t find anything in the error logs that looked suspect to me. I thought it might be power, but I’ve plugged the psu directly into the wall-socket, and the psu is a brand new “Corsair HX850 850W 80 Plus Platinum” which should easily support duel video cards. I also don’t think it’s power because Windows 10 does not show any issue with rendering on both GPUs.

I had just taken out the 2nd gpu when I ran the nvidia-bug-report. I think that’s okay, but I’ll rerun the report with the 2nd gpu if needed.

Put the 980 back in and see if using the persistence daemon fixes the issue. If the issue hits again, try to ssh in from another system and then run nvidia-bug-report to catch the error.

I’ve looked at the example nvidia-persistenced-init to set up persistence daemon in my systemd startup. So far with vidia-persistenced, I have not experienced a crash when both gpus installed. I’ll update here if I do experience a crash in the next 24hrs, which I usually experience a crash or two by then.

There is a nvidia-persistenced-init installer gunzip bundle that comes with the nvidia drivers. The script detects the init type of system and does the persistence installation, but why is this not installed by default with the drivers?

Depends on your use case if you want/need the persistence daemon.