GeForce 320.49 Slow?

[N0OA]

Just an interesting observation. I just updated the drivers on most of my GTX cards to the newer NVIDIA 320.49. The upgrade process generally went very smoothly. But I did notice that 1) All of my current FAH projects on these cards were dumped and new ones downloaded and started; 2) On two of my systems, FAH control lost the configuration for at least one GPU slot. I had to reboot the system and re-add the slot to get the GPU available for folding again; and 3) My folding performance seems to be down about 10% over the previous version of the driver. I am seriously considering rolling back to the previous version. I'm going to watch this for a couple more days to see if the PPD loss persists...

-N0OA

[Kurtis200200]

Hey all,

Have been running Folding@Home now on my desktop for about 3 weeks.

With a evga GTX 660 and unaltered client type, was averaging ~21,000 points per day.

Earlier this morning, added the advanced client type tag to 7.3.6 (to take advantage of core_17) and was up in the estimated 40,000 points per day range on Project 8900 (~6.5min TPF)

Then noticed that GeForce had released a 320.49 WHQL driver on July 1st, which I today downloaded and installed.

Now the same Project 8900 work unit estimates it will yield ~18,000 PPD, with a TPF of 13.3min.

I would, if possible, like to have as-up-to-date manufacturer drivers as possible, and would like to take advantage of core_17 as well since I understand it's late-stage testing and about twice as efficient for Nvidia GPUs.

Any thoughts as to the best course of action, or any insights into why PPD has plummeted after this new GeForce driver?

[Kurtis200200]

Also,

(1) Previously I had been running the WHQL 320.18 driver.
(2) Windows 7 64-bit OS
(3) GPU-only for F@H (quad-core Q6600 CPU, three cores running BOINC's World Community Grid, one core free to feed the GPU)

[bollix47]

Hi Kurtis200200 and welcome to the folding@home support forum.

Whenever I install nvidia drivers I always use the custom method of the install and remove the checks in all boxes except the drivers(can't remove that one) and I put a check in the Clean install option. If you used the express method you might want to retry the install using the custom method to see if your PPD returns to the higher level.

p.s. another folder is reporting a PPD drop with these drivers: viewtopic.php?f=66&t=24520

[Kurtis200200]

After the passage of some time (no change in settings or drivers or configuration, still running advanced methods on the same work unit as I was in the original post), my estimated PPD has returned to ~43,000, with a TPF of 7min (which I'm not sure is any slower than it was before, as I'm not sure that the 320.18 TPF was actually 6.5min).

As such, I will keep the "clean install" method in mind in case the actual performance deviates substantially fr the estimate, but it seems to have fixed itself otherwise (fingers crossed).

Also thanks for the helpful information and link to another 320.49 user; knowing more makes the troubleshooting process go much more smoothly.

[bollix47]

Sounds good.

The estimates in v7 can be a little off when a work unit starts up ... always best to wait for at least 3 or 4 completed frames.

[N0OA]

Sounds good.

The estimates in v7 can be a little off when a work unit starts up ... always best to wait for at least 3 or 4 completed frames.

I agree, the early estimates are often way way way off I prefer to wait for about 12 hours worth of frames before I reach for a conclusion. My driver update experience is still a loss of about 10% of my PPD on the same hardware and same WU. Though, there has been about this much variation from WU to WU within the same project - so it just might be coincidence. I did not do a "custom" install. So, I might give that a try as well to see if that help with the performance observation.

-N0OA

[Kurtis200200]

I presume that, if 320.49 turns out to be slower than 320.18, Folding@Home will likely not have enough spare resources to be able to patch the PPD loss on the 320.49 driver build for an extended period of time?

[7im]

FAH has nothing to do with drivers. That's completely on the manufacturer!

[folding_hoomer]

Stanford is not able to act in programming nvidia-drivers - so if there is a loss in PPD use an older driver (like 314.xx).
This one is extensivly tested while using new Core17 and WU 7662, 7663 and 8900 . . .

[codyf86]

Hello all,

Just wanted to share my observation. The 320.xx series Nvidia drivers seem to be problematic for a good percentage of the people using them...for gaming and folding. I upgraded to the 326.01 drivers for Windows 8.1 Preview (They install and work fine on Windows 7), and they have been leaps and bounds better as compared to the 320.xx series.

I would recommend you try them out, but do a custom/clean install.

http://www.nvidia.com/object/notebook-w ... river.html (x86/32-bit)

http://www.nvidia.com/object/win8-previ ... river.html (x64/64-bit)

[Kurtis200200]

Have you compiled performance statistics on the performance of the 326.01 driver on Windows 7 for F@H?

[Kurtis200200]

Just clean installed 326.01 on windows 7, no ppd change from 320.49 on project 8074

[codyf86]

Nope not yet, only for gaming so far and it has been much improved. Also the way my card acts, while under load folding...monitoring with Precision-X has been much more to my liking.

I didn't say it would increase your PPD, was just a suggestion. If you game at all the 326.01 are definitely better. If you look at the Nvidia forums you'll see the problems with the 320 series drivers.

[Napoleon]

Didn't notice any slowdown moving from 320.18 to 320.49, nor speedup. Then again, ION and GT430 are so old that NV probably hasn't bothered to develop anything new for them in ages, just maintaining compatibility.