Stress Testing

[substratus]

Ok, I need to find out something about Prime95...

I am assuming that most of you guys are using the beta release of version 23, right?

So there's this box that you can run 3 different types of torture test. I understand the difference between them, but which one is best to use for stability testing? Which test do you guys use and why, that's basically my question. :)

[NookieN]

For general stress testing, Blend is the best option. If you just want to stress you CPU (and not memory or chipset) you could use small FFTs.

Since you have an HT-enabled chip, if you want to run two instances of Prime 95 (from two different folders), it's best to run one with small FFTs and one with either Blend or large FFTs (not in-place).

[substratus]

Couple questions...

-When running a custom test, what max FFT length should I use?

-Why shouldn't I run in-place FFTs?

[NookieN]

The max FFT size isn't too important. I find that it's best to use the widest range of FFT sizes. On a marginally stable system, some FFTs are more likely to fail than others. I once had a system that only failed on FFTs in the range of 128-160k.

You can run in-place FFTs if you want, but in that case you won't be stressing your RAM much. In-place FFTs use the same chunk of memory for all operations, so only about 10-30M will get tested. If you don't use in-place, Prime95 can use most of your system memory (up to about 85%) for testing.

[substratus]

Excellent, thanks :D