better Nvidia drivers for folding?

Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
#1
Is anyone here GPU2 folding with Nvidia? If so, you might want to change your drivers to 185.20. Double your PPD almost? is the pertainant thread. I tried it on a x64 Vista machine and yes my PPD increased dramaticly (+75%) with a core_14 WU.

Drivers are avail. at: Why is my PPD so high? (Double your PPD with new drivers!). Just know that the drivers were leaked and thereby unofficial; So far, there is no clear reason anyone has discovered as to why they are doing what they do; Or even if they are producing valid folding results.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,692
Location
Québec, Québec
#2
Thanks, I'll try them and report on my results, although it will be very clear to anyone following this link :

http://folding.extremeoverclocking.com/user_summary.php?s=&u=58673

I'm folding on the same old GeForce 8800GT since April last year (maybe I started folding on the GeForce in August or September ; I don't remember).

P.S. No points for me during the last 24 hours because my client froze and I didn't check my system until this afternoon.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,692
Location
Québec, Québec
#3
I can already see the difference. Each frame of the same unit (Project 5904, Run 4, Clone 39, Gen 4 - using FAH Core 14) took almost 6 minutes with the 180.?? driver I used and they now take less than 4 minutes. I expect my output to increase by at least 50%, if not more.
 
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
#5
Stanford just released v1.25 of core_14 which is designed eliminate to speedup with 185.20 drivers. It was good while it lasted. Unfortunately, it was only a very short run.
 

Stereodude

Not really a
Joined
Jan 22, 2002
Messages
10,095
Location
Michigan
#8
Stanford just released v1.25 of core_14 which is designed eliminate to speedup with 185.20 drivers. It was good while it lasted. Unfortunately, it was only a very short run.
Why do they care if you can process the WU very quickly due to driver optimization? The WU still gets processed right? What am I missing here? :confused:
 
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
#10
ihaque said:
There was a bug in the core that interacted with some changed behavior in the 185.20 driver. The good news is that it doesn't appear that this particular error "contaminated" any runs; the effect here was that some results were simply not calculated (and therefore not returned), not that returned results were corrupt. Core execution was faster precisely because not all the calculations were done.

I've released an updated version (1.25) of the core that appears to fix the bug; your clients should automatically update when they get a new 59xx WU assigned, or you can manually force an update by stopping the client, deleting FahCore_14.exe, and restarting. Please reply to this thread if you still see the PPD "doubling" with the 1.25 core, as the core results are definitely missing some returned data under those conditions.
Basicly, they were not trying to fix the increased PPD, but rather they are striving to fix a bug that showed up with 185.20 drivers and the unfortunately the increased PPD was the casuality. Mastter of fact, the added thread syncronation needed has slightly decreased PPD for all drivers.
 
Top