P0 state performance tweak

It seems that a lot of GPU problems revolve around specific versions of drivers. Though NVidia has their own support structure, you can often learn from information reported by others who fold.

Moderators: Site Moderators, FAHC Science Team

Post Reply
foldy
Posts: 2061
Joined: Sat Dec 01, 2012 3:43 pm
Hardware configuration: Folding@Home Client 7.6.13 (1 GPU slots)
Windows 7 64bit
Intel Core i5 2500k@4Ghz
Nvidia gtx 1080ti driver 441

P0 state performance tweak

Post by foldy »

On Windows 7 with gtx 1080ti I could overclock MEM to +500Mhz but FAH was not using the setting. So I started nvidia profile inspector and created a profile for fahcore_21.exe and set "CUDA force P2 state" to OFF. After stop and start of FAH is using the overclock setting now and I get improved TPF. So this setting although called CUDA also sets OpenCL state.
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: P0 state performance tweak

Post by JimboPalmer »

Yes, my understanding is that NVIDIA built their OpenCL support on top of their CUDA code. So anything that speeds up CUDA will speed up OpenCL and F@H.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: P0 state performance tweak

Post by bruce »

The reports that I've seen on this forum seem to suggest that overclocking MEM is rarely worthwhile. It does increase performance slightly, but more can be gained by overclocking the shaders. Increasing either setting does increase power draw (and heat) but unless you run into something limiting stability, go for shader frequency rather than Mem frequency. (Those reports were probably assuming P2 state so you'll have to evaluate how they interact in P0 state.)
MeeLee
Posts: 1375
Joined: Tue Feb 19, 2019 10:16 pm

Re: P0 state performance tweak

Post by MeeLee »

@Foldy: What do you mean by "FAH was not using the setting"?
I thought the P0 state was mainly affecting the GPU speed, not really the MEM speed when manually overclocked?
At least on the RTX cards in Linux, this is the case. Perhaps not for GTX cards under Windows...

Is your environment controlled (constant, like indoors), or are there temperature fluctuations in your card?
If there are large fluctuations (eg: 50F night, 75F day), I wouldn't force P0, for reasons below:
On Linux, I just run P0 state speeds (or higher) on a P2 state, by running a good overclock with a good power-cap on the card.
Only when temperatures go below 50C on my cards (when ambient temperatures go below ~60F), is there a chance my cards jump to P0 state, and a good GPU overclock on P2 state would be a too aggressive overclock on a P0 state.
So as long as the temps never go below 50C, you can run P2 state, pretty much close to a P0 state.


4 questions:
- Is there a possibility that one can break the card more easily, by a too high overclock in the forced P0 state?

- Are P0 speeds affected by the boost frequency? (eg: a good P0 state overclock might break the WU at boost, when the card is colder than 50C (in between WUs, or from a cold start)?)

- I also would want to know if forcing a P0 state, really benefits over a good overclock in P2 state?
I always believed that the reason those cards are clocking down in *GPU on RTX cards* speed, is when temperature goes up; and some sort of safety for the card to keep giving accurate results.
If forcing P0 state allows them to run max boost speeds even at higher temperatures, there might also be more errors?

- I'd also be interested in seeing if forcing P0 state would affect system power consumption?
Post Reply