511.23 a bad driver?

[XanderF]

As above - just released today, and I updated to it (from 496.49). Have only had 4 projects since switching, but...well, my PPD on all of them is drastically lower than the earlier driver (about half).

Curiously, although the GPU is showing the expected 100%-ish load, it's not getting hot - temperature also significantly lower than previously (indeed the fans are barely spinning).

Wondering if anyone has any different experience with this driver?

[Neil-B]

I'm seeing if anything a very small (sub 1%) upward shift in ppd from this latest driver update so far (om sample of 4 wus) ... rtx3070, win 11 (latest build)

[gunnarre]

There are no Covid Moonshot WUs at the moment. Moonshot WUs have a higher PPD than other projects, especially on modest Nvidia GPUs, so I suspect that the lower PPD you're seeing might be because you're folding projects with less PPD on your card. Once more Moonshot sprints go live, you might see your PPD reach the same level.

Does your log say "Using CUDA" when it folds? If you got the driver from Windows rather than straight from Nvidia, you might have gotten a driver without CUDA support, so check that it's actually using CUDA in your log.

One other thing you might check is if your power limit was changed when you upgraded the driver and re-started.

[toTOW]

511.23 drivers seem a little faster (about 1 to 3 second less per frame) on all my GPUs than those I had before (from November).

[Veablo]

After updating to 511.23 GPU load is 100% (ish) but temperature is around 40. ETA on current WU over 4 hours.
After a reboot, temperature at 100% is 68 and ETA on current WU are down to 2 hours 25 min.
When the current WU is completed, GPU gets cold at 100% again. New reboot fixes it...

Update:
It happens during folding also.
Went from TOTAL ESTIMATED PPD 3,579,108 to TOTAL ESTIMATED PPD 1,845,936 suddenly.

RTX-3060TI LHR

[XanderF]

Went from TOTAL ESTIMATED PPD 3,579,108 to TOTAL ESTIMATED PPD 1,845,936 suddenly.

Is that after the reboot, or before? Because that's the sort of thing I was seeing - MASSIVE drop. I'm on a regular 3060, though, so my numbers were more like 2 million PPD before and then down to under 750k. And same with the utilization at 100% but strangely low temperature.

There are no Covid Moonshot WUs at the moment. Moonshot WUs have a higher PPD than other projects, especially on modest Nvidia GPUs, so I suspect that the lower PPD you're seeing might be because you're folding projects with less PPD on your card.

Did those run out, like, yesterday? Because looking at my PPD production over time - it's been fairly consistent at 2million until it dropped like a rock yesterday (a lot of that due to the bad WUs on project 13460, but even beyond those, performance has been much lower on newer drivers).

[Veablo]

The drop from 3,579,108 to 1,845,936 happened while folding, no reboot.

[XanderF]

Went from TOTAL ESTIMATED PPD 3,579,108 to TOTAL ESTIMATED PPD 1,845,936 suddenly.

Is that after the reboot, or before? Because that's the sort of thing I was seeing - MASSIVE drop. I'm on a regular 3060, though, so my numbers were more like 2 million PPD before and then down to under 750k. And same with the utilization at 100% but strangely low temperature.

There are no Covid Moonshot WUs at the moment. Moonshot WUs have a higher PPD than other projects, especially on modest Nvidia GPUs, so I suspect that the lower PPD you're seeing might be because you're folding projects with less PPD on your card.

Did those run out, like, yesterday? Because looking at my PPD production over time - it's been fairly consistent at 2million until it dropped like a rock yesterday (a lot of that due to the bad WUs on project 13460, but even beyond those, performance has been much lower on newer drivers).

Does your log say "Using CUDA" when it folds? If you got the driver from Windows rather than straight from Nvidia, you might have gotten a driver without CUDA support, so check that it's actually using CUDA in your log.

Log showed using CUDA, yes. I do notice that driver 511.23 introduces CUDA 11.6 - 511.09 (which seems to work okay for me) is CUDA 11.5 and the 496.49 driver I had been running up until yesterday and worked the best was CUDA 11.4. Wonder if there is an issue with 11.6...

[Veablo]

Doing a clean reinstall seems to do the trick. Been stable for a good 30 min.

Update: Didnt last long. Back to high GPU usage and low temp. And TOTAL ESTIMATED PPD 1,695,519

[toTOW]

Is it related to an other use of the GPU (web browser, video playback, ...) ?

Is there some events logged in Windows logs at the same time the slowdowns start ?

[Veablo]

I downgraded my driver to a Game Ready non DCH driver and all is good now.
Version 472.12, its a bit old but works.

[gunnarre]

Did those run out, like, yesterday?

Yes, the researcher aborted the projects ahead of time due to the error in making the work units. Other information here seems to suggest that this wasn't the issue, though.

[toTOW]

Veablo, XanderF > Can we see some logs showing the part where PPD is normal and a part where PPD is reduced ?

[Neil-B]

Could you also watch the clock speeds for the gpus and report if these are changing at all when this happens?

[tulanebarandgrill]

I don't know what other people do but I use the studio drivers for Win 10, for a number of reasons. I have had a couple of runs with low PPD but they ended up aborting (I posted this to a different thread but the tail end of the log is below. I'd check for this in case it's the same thing but I noticed same symptoms, clock rate really high, temp low, not much point action until it crashed. The next WU is working well.

15:38:24:WU00:FS00:0x22:Completed 175000 out of 2500000 steps (7%)
15:49:03:WU00:FS00:0x22:Watchdog triggered, requesting soft shutdown down
15:59:03:WU00:FS00:0x22:Watchdog shutdown failed, hard shutdown triggered
15:59:04:WARNING:WU00:FS00:FahCore returned an unknown error code which probably indicates that it crashed
15:59:04:WARNING:WU00:FS00:FahCore returned: WU_STALLED (127 = 0x7f)
15:59:04:WU00:FS00:Starting