Page 1 of 1

Any danger in a (possibly slightly) broken GPU?

Posted: Wed Apr 01, 2020 7:45 pm
by Joker
I have an older nVidia card that I used to run SLI with another like it. If I recall correctly, this card at some point developed some intermittent artifacting problems. It still appeared to render everything 3D correctly, but on the screen there would be times when the colors would get kinda screwed up.

I think I swapped the slots of the cards and used the other as my primary while the semi-broken one was still part of the SLI setup, and as far as I remember, that worked fine. The question is--if I try to use this broken card for F@H, and if it successfully works, is there a potential that it's returning erroneous data?

Is there ever a way to know if F@H simulations are bad? Or do they check themselves and simply fail?

-Joker

Re: Any danger in a (possibly slightly) broken GPU?

Posted: Wed Apr 01, 2020 9:19 pm
by Jesse_V
Hey Joker,

I don't know if I would recommend it. That kind of sounds like a memory problem, is my guess. You're welcome to try F@h and I'd just advise you to keep a close eye on the log to see if workunits successfully complete. The workunits check themselves and the server also does some final checks to ensure that the workunits looks sensible. You will see workunits fail halfway through or get rejected by the server. If you see that start to happen more than once, then remove the card.

F@h won't work very well with SLI so I wouldn't recommend that. That was tested in the research labs a number of years ago and it's actually more productive if each card takes its own workunit.