Torture test best practices

Hello,
I was hoping someone could shed some light on the usage of Prime95 as a tool to test for system stability. I know that different FFT sizes stress different aspects (e.g. smaller sizes stress the CPU, FPU, and L1 cache, while larger sizes may stress the memory, memory controller, and higher level caches more). Therefore, if a comprehensive test is desired, a variety of FFT sizes should be tested.

My question has to do with the "time to run each FFT size" setting available when running a custom test. Is there any significance in how long each size is tested? I know that it will cycle through the various sizes, and restart once it's done all of them. So if one is going to stress test for a specific amount of time (say, 12 hrs), would it make any difference if each size was tested for 1 min vs. the default setting of 15 mins?

I am thinking there could be a benefit in testing each size for 1 minute, in that the more frequent size switching would ensure that each size is given more equal test time. If it waits 15 mintutes before switching sizes, some sizes might not get tested as thououghly (or at all) if the total test time isn't long enough to cycle through all of them.

But is there any downside to a shorter time before changing sizes that I'm not considering?

Thanks!

[cheesehead]

It's called a "torture" test, not a "quick breeze-through" test.

Why would you want to test for such a short total time that it couldn't do each FFT size for 15 minutes? Concentrating on doing the same computation for that long allows heat to build up, to give a good chance of detecting whether a circuit is going to fail when it gets hot.

[LaurV]

Best torture practices:
- tarring and feathering,
- wooden horse riding
- judas cradle
- etc.

[chappy]

Those old fashioned forms of torture just aren't as effective as the more modern psywar stuff. For more information see: http://en.wikipedia.org/wiki/Justin_bieber

Originally Posted by cheesehead

Why would you want to test for such a short total time that it couldn't do each FFT size for 15 minutes?

Because, as I mentioned in the original post, you might not get all FFT sizes tested equally, or at all, with longer test times. For example, if you test for 12hrs, that may not even be enough time to run through all FFT sizes @ 15 mins each, so you end up testing some for 15 mins, and some not at all.

[Dubslow]

Originally Posted by Darin

Because, as I mentioned in the original post, you might not get all FFT sizes tested equally, or at all, with longer test times. For example, if you test for 12hrs, that may not even be enough time to run through all FFT sizes @ 15 mins each, so you end up testing some for 15 mins, and some not at all.

Perhaps, but Prime95 does an excellent job of "mixing it up" so that even if a particular size has not been tested, then another similar size will have been tested. If you really want, you can set it to 5 mins per FFT size, but it won't really make a difference as to whether or not an error is found.

[cheesehead]

Originally Posted by Darin

Because, as I mentioned in the original post, you might not get all FFT sizes tested equally, or at all, with longer test times. For example, if you test for 12hrs, that may not even be enough time to run through all FFT sizes @ 15 mins each, so you end up testing some for 15 mins, and some not at all.

Then test for longer than 12 hours!!!

[gamer30]

Originally Posted by chappy

Those old fashioned forms of torture just aren't as effective as the more modern psywar stuff. For more information see: http://en.wikipedia.org/wiki/Justin_bieber

Simply reading the link sent me into convulsions for a few minutes.