Page 1 of 1

Possibility of wrong F@H calculations from undervolting

Posted: Sun Mar 22, 2020 9:29 pm
by OdyWillMakeIt
Hi all,
I have a 5700 XT underclocked to 1821 MHz at 920 mV, and it gets about 850.000 to 900.000 PPD. I tested this voltage on OCCT, and it shows no errors at all, but if I lower it to 910 mV, the benchmark will start to get errors. I'm a little worried that the errors shown in the OCCT benchmark are equivalent to wrong results at F@H calculations.
Do I need to be worried at all about sending wrong results due to insufficient voltage, or any other reason?
The F@H application has never given me such a warning, except when the Radeon driver crashes as well and it's obvious that voltage is too low.

Re: Possibility of wrong F@H calculations from undervolting

Posted: Sun Mar 22, 2020 10:51 pm
by Joe_H
The GPU folding core does sanity check on the folding data using your CPU every checkpoint, so that should catch errors like what you are concerned with. So if you start seeing the progress stop with an error and then restart at the last checkpoint, then you should increase the voltage a tad. Some projects will stress the GPU more than others, so it might be best to leave a bit of a cushion.

Re: Possibility of wrong F@H calculations from undervolting

Posted: Mon Mar 23, 2020 3:25 am
by MrFrizzy
As Joe said, unless you are seeing errors in the logs, the system is folding without a problem.

I too have a 5700 XT and have been messing around with over and under clocking. I'm impressed that yours is able to maintain 1800MHz on 920mV. My current testing is at 1800MHz and I can't go lower than 1025mV. Shoot, I can't even go that low on 1700MHz. Sounds like you may have hit the silicon lottery with that one! Glad to see another Navi card folding!

Re: Possibility of wrong F@H calculations from undervolting

Posted: Fri Mar 27, 2020 5:57 pm
by OdyWillMakeIt
Thank you!
I'd like to fold 24/7 now that I'm home all day, and I've been looking for a voltage low enough to not reduce my card's lifespan. Now I've set it to 1450 MHz (1418 actual) at 750 mV, which is as low as the voltage will go, and 25 mV above idle voltage (725). I figured now it's the same as idling 24/7. It's getting 650.000 PPD on project 11743.
Maybe that's too much underclocking, and 1821 (1780 actual) at 920 mV is just right.
It always hovers around 40-50 MHz lower than what I set.