Folding used to work, now GPU_MEMTEST_ERRORs

[rsquared]

Built a full-time F@H system, focused on Alzheimer's, in honor of my stepfather, who died of Alzheimer's. Its two GTX 580s have been folding fine for months. But for the past month or so, I haven't been able to get any GPU folding out of it; the WUs promptly fail with a GPU_MEMTEST_ERROR.

Both cards are at stock clocks, so overclocking isn't messing it up. Both are full-cover-block water-cooled, and there's no temperature problems. I haven't changed anything, just left it to folding, but then one day noticed the air coming out of the case wasn't as warm as usual, and logged in to find the failures. I've removed and reinstalled the video drivers (v331.82) and that didn't help. I deleted the cores and WUs and GPUs.txt to force the client to redownload everything and that didn't help. I removed and reinstalled F@H itself and that didn't help.

Is anyone else experiencing this? A bit of Googling turns up an instance where the assignment server was handing out the wrong kinds of GPU WUs (link), and I wonder if this is happening again?

 

[MikeTimbers]

Would need to see the start of the log but one immediate suggestion would be to revert to older drivers like the v326 or v314. v331 is known to not play nicely with folding unless you have a 780ti or Titan.

 

[rsquared]

I installed v314.22 and it has been folding without errors since. So I guess the problem is solved, but it doesn't really explain the cause. I had installed v331.82 after I saw the failures, in the hope that the Latest Version would fix it; I didn't make note of what version was installed before. Oh well... fold on.

And FWIW, v331.82 is on my main PC, with a GTX 680, and GPU folding has been working fine on it.

 

[musky]

Certain card/driver combinations just don't work with folding. Your 580s are getting pretty long in the tooth, and newer drivers aren't necessarily going to be optimized for them. The 680 is newer, which is probably why the newer drivers do work with it.

 

[bigted]

And FWIW, v331.82 is on my main PC, with a GTX 680, and GPU folding has been working fine on it.

What ppd is your 680 at? I bet it would be higher ppd with the older drivers. From about 326 and newer the drivers kill ppd on any card but the 700 series.

 

[rsquared]

So it seems I was mistaken—I checked this morning and found both GTX 580s had failed with GPU_MEMTEST_ERRORs again. It lasted long enough to fool me yesterday, though; maybe they each folded one WU successfully and failed on the next? I had to leave for work and didn't try to dig into the log.

 

[Grandpa_01]

Try deleteing the current fah core and see if that helps any.

 

[rsquared]

Deleting the cores didn't help; it re-downloaded the cores and failed again.

I was getting about 47K PPD on my main PC (4770K + GTX 680). Installed the v314.22 driver to see if it performed better. Now it, too, has failed with a GPU_MEMTEST_ERROR. Going back to v331.82.

Edit: v314.22 was not correctly installed. I had some strange half-assed installation. Ran Driver Cleaner, no wait, Driver Sweeper, nope, Display Driver Uninstaller (may have changed name again by the time you read this) and am trying again with a proper v314.22 installation. Right now FAHControl is showing 94K PPD, so yeah, it seems to have made a big difference. I will try the full Driver Whatever and reinstall routine on my dedicated folding PC.