Folding Forum

PPD on that driver double down from 68-69k to 30k on GTX670, back to 314.07

This was first seen in the 331.40 beta drivers, where all Keplers except those based on GK110 (780, Titan) have a massive PPD hit. Others have reported the same behavior with this new set.

Another validation for op's report. 8900 (61, 0, 99) with 331.58, TPF 12:25. Reverted to 327.23, TPF 6:58 On GTX 660.Proof driver funky? Or co-incidence? Time will tell.

Any idea how to get nVidia´s attention on this topic?
I mentioned it in the "GeForce Drivers" feedback threads in the nVidia support forum but I didn´t see any entry in the known bugs sections of the Driver documentation.

Heiko

Approach it from a GPGPU angle, saying their OpenCL performance took a big hit in the latest driver, and quote your numbers. They understand that language better than saying Folding@home runs slow.

I wrote a new post in the nVidia support Forum:

https://forums.geforce.com/default/topi ... 7/#3983617

But with support from other folders it´s more likely to get nVidia´s attention.

Heiko

Yeah this is quite a predicament I need the new drivers for battlefield 4. Maybe install Ubuntu on another partition and boot to that when running folding@home. Not ideal solution but what else can you do?

Booting Linux part of the time and Windows part of the time is (almost?) always a bad proposition. Any possible gain you might think you're getting from booting something else will be lost due to the reduction in the QRB because WUs downloaded on the "other" OS will be sitting idle. Choose one or the other and stick with it.

Installing a VM that allows both OSs to run simultaneously may be possible, but it's a complicated choice and it does carry some extra overhead.

"New" drivers will most likely be replaced with newer drivers (soon?) that can run both FAH and your game (though probably not concurrently).

331.65 WHQL have the same issue. For anybody interested.

Please publish your findings in https://forums.geforce.com/default/topi ... 31-x-beta/.
Probably more postings from different people could push the topic.

Heiko

Posted, hope it helps

Calcii wrote:PPD on that driver double down from 68-69k to 30k on GTX670, back to 314.07

The 331.65 driver has some very dismal FAH performance as well (link):

art_l_j_PlanetAMD64 wrote:After allowing enough time for the PPD readings to stabilize (~1 hour on each system), I found that all 24 of my Windows 7/NVIDIA GPU systems had experienced a 50% to 60% drop in GPU folding PPD values, using the v331.65 driver, when compared to the v327.23 driver.

Going back to the v327.23 driver restored my folding performance to its previous value (== over 4 Million PPD).

Drivers 32x.xx work
I myself run 320.49 on my 3 gtx650 tis and my gtx680

art_l_j_PlanetAMD64 wrote:Going back to the v327.23 driver restored my folding performance to its previous value (== over 4 Million PPD).

Same here with my GTX 680. I get about the same PPD with 327.23 as with 314.22 (my previous favorite driver). Anything newer=massive PPD loss.

The earlier 32x.xx drivers were very unstable for me with my Kepler and Fermi cards, which is why I stayed with 314.22 for so long.

I'm getting low numbers as well on the gtx680 with 327.23. I've reinstalled the driver but only managed to increase the ppd by the 100's, havent seen any big jumps since the downgrade of driver.