Overclocking and PhysX

XJnine · #1 4th January 2009, 08:58 AM

Just checking to see if anyone else has done something like this to maybe save me some time.

I currently have a GTX280 as my main GPU and an 8800 GTX as my PhysX GPU. I know it's not really necessary but I was looking into overclocking the 8800GTX for the best PhysX performance and I was wondering where I should focus my efforts. Core, shader, memory? Is one of those things going to be more important to the PhysX performance than the other? I'm guessing the shader clock but I don't know how much the core frequency or memory bandwidth play a part in the overall PhysX performance. Ideally I'd like to only overclock the parts of the card that are going to make a decent difference to keep temps down and prolong the life of the card. (I know, if I wanted to do that I wouldn't be overclocking it at all).

I'm planning on testing my changes with FluidMark. Does anyone have a better suggestion as to a benchmark to use? Any recommended settings I should consider?

Thanks!

Olin Coles · #2 4th January 2009, 11:11 AM

FluidMark is a good tool for testing PhysX, and there a few free PhysX-intensive games listed on their site which would also work well for testing.

The core clock is going to be tied into the shader clock; so both with move dynamicly. I wouldn't forget about the vRAM, as that yields good performance gains too. Raising the core clock will also raise temps, but yield more improvement than a memory overclock for PhysX.

XJnine · #3 4th January 2009, 11:55 AM

Interesting side note. I've found that you can't overclock a secondary non-display graphics card through RivaTuner or ATI Tool. RivaTuner can't detect the clocks and therefor will not let me adjust them and ATI Tool will let me move the slider around but when I apply the new settings it just goes back to the default.

I can do it through the Nvidia control panel though. Go figure... The Nvidia control panel also lets me change the core and shader clock seperately although I don't know if they are still synched behind the scenes and one of the control panel settings is ignored.

Thx for the info Olin!

Olin Coles · #4 4th January 2009, 01:42 PM

If you had the time to put into a side-experiment, you could figure out the maximum software-based overclocks, and then apply the settings permanently. See the article: Overclocking the NVIDIA GeForce Video Card for step-by-step details.

XJnine · #5 4th January 2009, 09:51 PM

Well, I learned a couple of things.

First off, FluidMark is not a very good benchmark to use to test the benefits of overclocking for PhysX. The scores generated by the benchmark vary by by a few hundred points with each run with the video card settings the same. When making changes you can't tell if it's just the program giving a higher result thiat run or if it's because of the overclock.

I've also learned that OC'ing the 8800GTX for PhysX doesn't seem to be doing much good as the scores I was getting at my maximum overclock weren't measurably any better than with the stock settings.

I'll have to test with some other games and benchmarks but it looks like I might be leaving things as they are...

Olin Coles · #6 4th January 2009, 10:00 PM

The scores are going to be off (and inconsistant) because you're pairing up cards. If it were only one card, they would be much more consistant.

XJnine · #7 5th January 2009, 06:22 AM

I thought it would be more consistent since I'm using one for the display and one for the physics. Each card has its own task and I would think that would give results that were more repeatable, hmm. I'll try it with just my GTX 280 and see how those scores look after multiple runs. Also, any good benchmark program would give consistent scores across each run as long as the hardware and its parameters remain constant throughout each benchmark run. With the same settings the scores were varying by more than 5%.

Guess it's back to the drawing board. I'll let you know how things turn out.

J