GPU stress test

[billford]

From the Tools List topic:

- GPU Caps Viewer:
GPU Caps Viewer contains a stress tester for GPUs. If your GPU can run the fur rendering test for an extended period of time without issues, it should be stable.

Is that still a valid comment, and how long is "an extended period of time"?

I ran it on a 780 TI for a bit over 30 minutes with nothing obvious showing as amiss, is that a long enough test for that sort of card?

[P5-133XL]

I would use FAHBench as a GPU stress test for folding. The main reason is that the program actually folds a test WU using FAHCore_17.

As To GPU Caps Viewer, I can't answer because I've never used that program.

[PantherX]

Generally speaking, for overclocking stress, an extended period of time is usually overnight to 24 hours.

As P5133XL states, FAHBench is specific in nature and a good starting point. Just remember that future projects can be more stressful than FAHBench. Having said that, using other softwares (http://www.techpowerup.com/downloads/Benchmarking/) to stress test a GPU is a good idea. Use the lowest common overclock setting and do remember to stress in the hottest possible environment that you encounter, i.e. if your ambient temperature is between 28 to 32 degrees Celsius, testing in an environment of 20 to 25 degrees Celsius would mean that if the ambient temperature rises, your overclock might become unstable (if it was on the edge of stability).

[billford]

OK, FAHBench downloaded, I'll give it a try tomorrow.

Ambient temperature isn't too much of a problem (I think), the version of OCGuru that came with the card allows the setting of a "Target" core temperature (badly named, it's a limit not a target); if it goes over the set value the clock and/or core voltage are reduced to bring it back down.

I've got it set to 65ºC at the moment which is a few degrees higher than the value I've managed to get it down to when crunching a P13000/1 at default settings. It seems stable, but it hasn't yet had any of the WUs that threw it when it was running at 75-80ºC.

[PantherX]

BTW, you can capture a live WU and test it out on other systems using FAHBench (viewtopic.php?f=38&t=25229).

[billford]

Thanks, I was just about to ask if it was possible to get hold of a "real" WU in the event that it didn't like one particular project

[7im]

Thanks, I was just about to ask if it was possible to get hold of a "real" WU in the event that it didn't like one particular project

The default test WU included with FB is very simple. If it fails that test you know you have problems.

[billford]

OK, FAHBench downloaded, I'll give it a try tomorrow.

I reviewed the history of this card overnight… when I first installed it it was fine on some projects, marginally unstable (ie it got most of the way through a WU) on others. Since then I've managed to reduce the core temperature by 15-20ºC (and learned quite a lot about using GPUs!). So I think I'll let it fold normally for a while to see whether or not I've still got a problem.

In the meantime, could someone explain a couple of terms to me (or point me to a relevant page) please?

Specifically- what the terms implicit and explicit solvent mean in this context?

[P5-133XL]

My non-technical amateur understanding:

Explicit means the water/solvent atoms are explicitly included in the WU. These WU's are much bigger needing far more calculations to complete.
Implicit means the water/solvent is assumed to be there but do not actually exist in the WU, rather the effects of its existence are simulated.

[billford]

My non-technical amateur understanding:

Just the right level for me to comprehend it, which I do

Thank you.