Page 1 of 1

GTX 780 no longer reaches P0 state after running FAH

Posted: Sat Mar 29, 2014 2:59 pm
by biliskner
I use a GTX 780 Lightning by MSI overclocked +300/+390 Mhz core/memory. (OC 1280/3394, factory OC 980, default clocks 862/3004).

After running core17, my clock drops to 575mhz (memory remains at 3394), and even after exiting and restarting I am unable to run any games at the original core clock.
Opening NVIDIA Inspector shows that my P-state is P2 i.e. default power setting, not under load
My p-state drops to P8 normally when FAH or games are not running, but even under maximum load only goes to P2.
Attempting to force p state to p0 with nvidia inspector does not work, the p-state is shown as P0 however the clocks are still locked at 575mhz. Forcing default P-state does not resolve the issue either.

The only solution is to completely reinstall GeForce graphics drivers, and that's not really viable on a daily basis.

Also tested forcing P-states with fresh driver install under NVIDIA inspector, able to lock card in P2 state and after that revert back to normal default.
It's only after running the FAH GPU core that my card is well.. screwed basically.

TL:DR, FAH GPU core screws with my card's performance states, forcing a reinstall of the drivers every time it runs

Code: Select all

13:24:54:******************************** Build ********************************
13:24:54:      Version: 7.4.4
13:24:54:         Date: Mar 4 2014
13:24:54:         Time: 20:26:54
13:24:54:      SVN Rev: 4130
13:24:54:       Branch: fah/trunk/client
13:24:54:     Compiler: Intel(R) C++ MSVC 1500 mode 1200
13:24:54:      Options: /TP /nologo /EHa /Qdiag-disable:4297,4103,1786,279 /Ox -arch:SSE
13:24:54:               /QaxSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 /Qopenmp /Qrestrict /MT /Qmkl
13:24:54:     Platform: win32 XP
13:24:54:         Bits: 32
13:24:54:         Mode: Release
13:24:54:******************************* System ********************************
13:24:54:          CPU: Intel(R) Core(TM) i5-2500 CPU @ 3.30GHz
13:24:54:       CPU ID: GenuineIntel Family 6 Model 42 Stepping 7
13:24:54:         CPUs: 4
13:24:54:       Memory: 15.98GiB
13:24:54:  Free Memory: 14.36GiB
13:24:54:      Threads: WINDOWS_THREADS
13:24:54:   OS Version: 6.1
13:24:54:  Has Battery: false
13:24:54:   On Battery: false
13:24:54:   UTC Offset: 8
13:24:54:          PID: 3612
13:24:54:          CWD: C:/Users/Brandon/AppData/Roaming/FAHClient
13:24:54:           OS: Windows 7 Home Premium
13:24:54:      OS Arch: AMD64
13:24:54:         GPUs: 1
13:24:54:        GPU 0: NVIDIA:3 GK110 [GeForce GTX 780]
13:24:54:         CUDA: 3.5
13:24:54:  CUDA Driver: 6000
13:24:54:Win32 Service: false
13:24:54:***********************************************************************

Re: GTX 780 no longer reaches P0 state after running FAH

Posted: Sat Mar 29, 2014 3:11 pm
by Joe_H
Folding is known to push the limits on overclocked GPU's, even ones that are supplied that way from the maker. The first debugging step is to try running with clocks set at reference values for the chipset.

Re: GTX 780 no longer reaches P0 state after running FAH

Posted: Sat Mar 29, 2014 3:30 pm
by PantherX
Welcome to the F@H Forum biliskner,

Generally, with an unstable overclock (manually done or by the factory), the graphics driver can reset which will lock the GPU frequencies in 2D mode. Rebooting the system would allow the GPU to perform normally. However, as Joe_H stated, the solution is to first try with Nvidia frequencies and see if your issue is resolved. If it is, try factory overclock frequencies. Do note that there are reports that some GPUs when used at the factory overclocked frequencies were unstable for F@H (since gaming is usually less stressful when compared to folding). Thus, they had to use Nvidia frequencies.

Re: GTX 780 no longer reaches P0 state after running FAH

Posted: Sun Mar 30, 2014 8:56 am
by biliskner
I don't get it :( I can't replicate the behaviour now even though it happened over restarts and 3 driver reinstalls over the last couple of days. I also realised that since I set my card to LN2 bios there was no inherent factory overclock.

It's currently folding stable at my normal overclock.
First attempted to start folding with card set to nvidia reference clocks with low power and temperature targets to force gpu boost off, no problem there.
Gradually raised core and memory clocks to my normal overclock.. no problem
Just in case it was due to folding starting at overclock speeds (and a resultant surge or whatever) I also shut down the GPU core, loaded my overclock profile in afterburner with overvolts and all. Restarted core, no problem.

Now i's confused. :shock:
I let you guys know if it happens again, might need to test overnight