[azagahl]

What good GPU stress testers exist?

I know rthdribl, 3dMark06, and Aquamark3 all stress the GPU.

But these also consume a lots of CPU time. I want something that stresses the GPU only, while consuming almost no CPU time. Then I can run it at the same time as a CPU / RAM stress tester (e.g. Orthos). Any ideas?

[Jeremy]

I don't think that's possible. You just run it overnight and keep a log of your temperatures with SpeedFan or Everest.

[azagahl]

>I don't think that's possible.

And why would it not be possible? There's no technical limitation at all. A DirectX app can issue a complex ID3DDevice9::Present() call and then the thread can yield until the GPU finishes. Very little CPU time would be consumed. I've used Rthdribl and Orthos together along with several other stress test programs at the same time; unfortunately Rthdribl consumes lots of CPU time which is basically dumbing down the Orthos torture test and making it easier on the CPU.

>You just run it overnight and keep a log of your temperatures with SpeedFan or Everest.

Yes, we all know how to use a GPU stress tester. The question is, what good ones are available?

This post has been edited by azagahl: 09 February 2007 - 07:01 PM

[Jeremy]

azagahl, on Feb 9 2007, 08:53 PM, said:

And why would it not be possible? There's no technical limitation at all. A DirectX app can issue a complex ID3DDevice9::Present() call and then the thread can yield until the GPU finishes. Very little CPU time would be consumed. I've used Rthdribl and Orthos together along with several other stress test programs at the same time; unfortunately Rthdribl consumes lots of CPU time which is basically dumbing down the Orthos torture test and making it easier on the CPU.

Well, you have a stronger pickiness towards those aspects it appears, which is fine. I don't know enough about it so I can't answer that.

azagahl, on Feb 9 2007, 08:53 PM, said:

Yes, we all know how to use a GPU stress tester.

It's not necessary to get snappy here. Some people aren't as knowledgeable.

The point is, I've run the first one you mentioned and it showed me:
1. How hot my video card got at maximum load after 7 straight hours (overnight)
2. How loud my system gets.
3. The maximums FPS my card can do when I had the settings as high as the card could handle.

I don't know what else you'd want to know after a GPU stress besides those. Cheers.

This post has been edited by Jeremy: 09 February 2007 - 07:59 PM

[noguru]

You could try Video Card Stability Test:

http://freestone-group.com/

Still uses CPU but not comparable to 3DMark or Aquamark

[azagahl]

>1. How hot my video card got at maximum load after 7 straight hours (overnight)

If you stress all components (GPU, CPU, RAM, floppies, hard disks, cd roms) simultaneously you can reach the highest temperatures. The reason I'm interested in this, is that by stressing the whole PC, one can find out what the highest stable level of overclocking is. Unfortunately, most GPU tests dumb down the CPU with instructions that do not test stability and generate little heat, meaning that you are basically testing the GPU and CPU independently instead of testing them together.

I'll try the Video Card Stability Test. Thanks, noguru.

BTW, IMHO the ideal GPU test would also count and log artifacts instead of just trying to crash the system. This goal probably conflicts with the goal of low CPU usage though.

This post has been edited by azagahl: 12 February 2007 - 03:23 PM