pc restarts without any message

Graphics / Linux Linux
1

hello
when i train neural net with gpu, sometimes the pc restarts without any message. i am running ubuntu 16.04 on pytorch.

please, help

Regards
Milton

2

First of all, check the kernel logs for anything just before the time of reboot. Then check the power consumption while the GPU is heavily utilized using nvidia-smi command. You may be nearing the power load that your PSU can supply.
Your PSU should have a factor of safety of around 150 Watts. For example, if your GPU can consume 200Watts, the box may say “Recommended minimum 450 Watts PSU”. That means you really want a 450+150 = 600Watts PSU.

3

hello
Thanks for reply. i have 2 1080ti whose power is (320w+250w). my ps is 850w. when i use single gpu, it still shows same restart automatically sometimes.

how can i view the kernel log?

regards
milton

4

If you are on systemd, journalctl.
For example, if last reboot was within the last 5 hours, type (as root):
journalctl --since “5 hours ago”

If you are not using systemd, check /var/log/kernel.log

You can always consult your distribution on reading log files so it neither option works, just ask them for instructions on log locations :)