BOINC occasionally crashes my computer when using GPU
14 Oct 2020 20:06:37 UTC
Topic 223716
(moderation:
The GPU I have is NVIDIA GeForce GTX 1070 (4095MB) with the latest driver.
After running for anywhere from 30 minutes to a few hours using the GPU, my computer freezes and I have to force shut down.
The reason I have centred on the GPU is that I suspended GPU usage in BOINC and it no longer freezes.
I ran an app to log temp/fan readouts, and nothing seems to be running partiularly hot (or anywhere near to the limits of safe ranges, anyway).
I've never had any issues with my GPU, and I normally OC it and run games that flog it to bits.
For some reason, BOINC seems to not play nice with it. I've tried removing the OC settings and even "under" clocking it, but the same issue keeps happening.
SplendidDiscovery wrote: The
)
Download a prgram like MSI Afterburner and check out your gpu's temperature, it's probably waaay too high which is shutting down the computer. If this is a laptop, your computers are hidden so I can't tell, that is the most likely problem. The Projects have gotten very good at writing their apps to use as much of the gpu as it can and that can stress a laptop gpu's more than it was designed to do.
A failing 7 yr old power
)
A failing 7 yr old power supply caused one of my boxes to do that.
I use Afterburner to OC
)
I use Afterburner to OC everything, and Open Hardware Monitor to log readings to a file (as it logs in a convenient CSV format).
Someone else had suggested that the temperature may spike in under a second, and Open Hardware Monitor only monitors per-second, so I've now also turned on Afterburner's logging to every 1/10th of a second.
As far as GPU temps go, when in use by BOINC, it sits in the mid-60's, which is a bit lower than where it tends to sit when gaming (which tends to max out at about 75).
This is a desktop PC, by the way, not a laptop.
I think I've made my computer visible so you should be able to see it now.
This is an interesting
)
This is an interesting thought. I hadn't considered that.
What weirds me out though is that this machine can handle playing high resolution, high frame-rate gaming and it's never conked out on me before. It seems to be something specific to BOINC.
Pushing pixels to a screen is
)
Pushing pixels to a screen is vastly different than crunching floating point math for a gpu.
Keith Myers wrote: Pushing
)
Fair point, but was more going on the GPU temperature, which is what, to me, sounds like a likely culprit of what could be causing the computer to crash.
Also I feel as though a GPU being used to "push pixels to a screen" is a slight oversimplification of what it's doing when gaming =P
And I'm still not sure why it's specifically BOINC that's causing this issue, and how to fix it short of just cutting the GPU out entirely (I can't find any settings to use a certain % of GPU, so it seems all-or-nothing).
SplendidDiscovery wrote: And
)
Yes that's true on or off. Also yes your pc is now visible, thanks!! As for the temps 75c is not overly warm for a gpu in a desktop. Do you leave a cpu core free just for the gpu to use or do you crunch cpu tasks with all of the cpu cores?
mikey wrote: Do you leave a
)
I have been playing around with the CPU% and it's currently set to 85% (which I guess leaves 1.8 physical/0.9 logical cores free. Math is annoying with 6 cores).
I did have that on 75% (3 physical/1.5 logical cores free) when I was trying to diagnose the problem earlier and found it was still freezing.
I might try it on a super low number like 25% to see if it makes any difference at all.
SplendidDiscovery
)
That sure is enough cpu cores free, the problem then could be cpu temps if you are down to 85% now and it's still crashing.
I was going to say running out of memory too but you have 32gb of memory so that isn't the problem either.
Wow I just noticed I
)
Wow I just noticed I embarrasingly and clearly mixed up the words "logical" and "physical" in my brain somewhere there.
I've taken the CPU usage preference right down to 25%. I'll leave it for some hours and see what happens. If that fixes the problem, I guess I can slowly increase it to work out the threshold.