Is this expected ? Most Jobs don't finish

[FoldMeBackwards]

Hi,

I returned after a longer absence...

Is this a log you expect?
https://prnt.sc/5mC6ybqUL4d2

Occasionally, I get this, inside the WU:
https://prnt.sc/V0-JzElwT7F1

this is all GPU btw

[arisu]

No, that log is not how it's expected. WUs should almost never fail (and when they do, the failure should almost always be right at 0.0% completion). Make sure your system is stable, and make sure that you pause the WU before you shut down or reboot your computer. If you don't pause WUs, then when you try to shut down, Windows will kill the cores that are running the WU and the client will mistake that for a core crashing on its own.

[FoldMeBackwards]

Thanks for the swift response.

Might this be because my CPU is busy doing other things while the GPU does FAH ?

[arisu]

FAH is designed to handle any amount of system load. If your CPU is busy, that just means that folding will be a little slower. But it won't cause it to crash.

Open the text logs and post it, since that will give more detailed information.

[FoldMeBackwards]

Thanks, I'll dig deeper, but everything else works fine on this GPU (Gaming)

[arisu]

Are you overclocking or undervolting your GPU at all? Folding pushes the GPU a lot harder than gaming and it's also a lot more sensitive to instability. A tiny glitch in a computation for a game might cause a single pixel to be out of place which you won't even notice. One error every trillion operations probably won't corrupt a game but will definitely corrupt a simulation. So your GPU could be unstable without you even knowing it.

But that's a lot more failures than is normal. You should pause GPU folding until it's resolved (CPU folding looks fine).

[FoldMeBackwards]

This is a good Point. I was doing undervotling and overclocking.
I'll remove it and give it another test. Thanks

[arisu]

The failed logged credits thing is normal on the other hand. It just means that others failed the WU as well, which happens sometimes. Usually it takes only one try to finish a WU but if it fails, the server will send it out to someone else to try again and it continues until it succeeds (or meets a threshold of consecutive failures that would indicate that the problem is with the WU and not with the folders. In that case the WU is pulled from circulation).

That's also the reason why it's not good to let so many WUs get dumped. The server only sends it out to one person at a time so any time spent folding on a WU that gets dumped is a delay for the project.

Be careful with overclocking a GPU btw. It's not like overclocking a CPU where the first sign of trouble will be sudden and undeniable. Overclocking doesn't really give you any benefit for folding unless your GPU is liquid-cooled, because the limit will be thermal dissipation and not boost clock speed. Light undervolting can help if done carefully but you still have to monitor things. But if it's not stable for folding, you probably shouldn't game with it either because errors *are* happening even in your game, you just haven't noticed them (yet). Think about FAH as a stress test program, and the crashes indicating that it failed the stress test and that the tweaks were too aggressive.

[FoldMeBackwards]

I was doing undervolting to reduce the FAN speed, as my GPU can be loud when the FAN goes off, and overclocking to still have the same "power". The setting themselves should be stable because on GPU mining they work fine without complaints. BUT, I get your point that folding is a different story and might be less tolerant in "error handling".

[Joe_H]

With modern cards just leave the voltages to the firmware when dealing with keeping a GPU stable for folding. What works for most is setting power limits instead. Mining uses different functions, and may not have as much error detection as folding. The GPU folding core both checks for calculations that reach a NaN and will try to restart at the last checkpoint, and do sanity checks on your CPU every time a checkpoint is done. Sometimes the NaN is where the calculations will end, but it can also be from an arithmetic unit in your GPU malfunctioning from insufficient voltage. In a game that might show as a pixel with the wrong shade, you might never notice.

[FoldMeBackwards]

Hi,
I can confirm, once the modification to the GPU were removed, all jobs finished successful

https://prnt.sc/wFeilmJKM4-W

@Joe_H
I was using MSI Afterburner and did not modify the voltage settings but used the "Power limit" which indirectly modifies voltage. I did this to reduce the power costs when doing mining. I could reduce it by around 50-90 W, which is a lot if you have your PC running 24/7.
Removed now and all is good, apart from my power bill

[arisu]

Hi,
I can confirm, once the modification to the GPU were removed, all jobs finished successful

https://prnt.sc/wFeilmJKM4-W

@Joe_H
I was using MSI Afterburner and did not modify the voltage settings but used the "Power limit" which indirectly modifies voltage. I did this to reduce the power costs when doing mining. I could reduce it by around 50-90 W, which is a lot if you have your PC running 24/7.
Removed now and all is good, apart from my power bill

Usually power limit will only decrease the clock rate but if you're forcing it to overclock and then telling it to limit power, I guess it has no choice but to limit voltage. For folding and when not overclocking, setting the power limit slightly helps reduce wattage a lot without decreasing PPD very much. You can get like a 30% wattage reduction with only a 5% decrease in PPD or around that. Though decreasing power even further eventually causes a sharp drop-off in PPD.

[FoldMeBackwards]

I played a bit with then power limit (%) around, and it seems that at 87% the FAN does not go off and the TPF appears to be equal.
Not sure about Watt, but that's ok for now.

Thanks for your support.