How to set up stress test?

Could I get any tips how to set up Prime95 for stress testing for various purposes please? I saw a few articles over the years, but they seem to be pretty outdated. I understand lots of changes were done to the software so things probably changed.
I read the Small FFTs is best way to test just the CPU, but for memory it's better to set up a custom test. I could use some help with that.

[Dubslow]

Have you read the "stress.txt" file that comes with Prime95?

There are three pre-defined tests you can run. "Blend" is best for testing memory, though honestly if you're testing only memory, then consider Memtest86.

Here's the description from the Linux (command line) version (which is mathematically identical):

1 = Small FFTs (maximum FPU stress, data fits in L2 cache, RAM not tested 
much).
2 = In-place large FFTs (maximum heat and power consumption, some RAM 
tested).
3 = Blend (tests some of everything, lots of RAM tested).
11,12,13 = Allows you to fine tune the above three selections.
Blend is the default.  NOTE: if you fail the blend test, but can pass the 
small FFT test then your problem is likely bad memory or a bad memory 
controller.

If you're ambivalent about what to test, then do the Blend test, which tests "some of everything".

Um, I suggest you re-read my post. I specifically said custom. Why? Based on http://www.playtool.com/pages/prime95/prime95.html this guide. It's pretty old, but some basics didn't change I assume.

[kladner]

Originally Posted by Octopuss

I read the Small FFTs is best way to test just the CPU, but for memory it's better to set up a custom test. I could use some help with that.

I have experimented a bit with the Custom options for torture test. I don't remember specific limits, but I tried increasing the max FFT and Memory to Use sizes. I think I got away with 8192 for the FFT. The amount of RAM in your machine will determine how high you can set Memory. Too high, and the test will crap out. I think (on a Win7-64 box) I had to leave about 2 GB for the system.

If RAM testing is a major concern I strongly recommend Memtest86+. http://www.memtest.org
Download the ISO and burn it to a CD. Boot from the CD and the program starts automatically. For certainty you should run it for at least an hour. More is better.

[Dubslow]

Originally Posted by kladner

I have experimented a bit with the Custom options for torture test. I don't remember specific limits, but I tried increasing the max FFT and Memory to Use sizes. I think I got away with 8192 for the FFT. The amount of RAM in your machine will determine how high you can set Memory. Too high, and the test will crap out. I think (on a Win7-64 box) I had to leave about 2 GB for the system.

If RAM testing is a major concern I strongly recommend Memtest86+. http://www.memtest.org
Download the ISO and burn it to a CD. Boot from the CD and the program starts automatically. For certainty you should run it for at least an hour. More is better.

Well yes, but how do you actually set up the custom test? I couldn't figure it out.

[kladner]

In Windows go to Options>Torture Test and tick the Custom radio button. I don't have a Linux box anymore, but it seems that this would be a series of choices in the menu system.

Originally Posted by kladner

I have experimented a bit with the Custom options for torture test. I don't remember specific limits, but I tried increasing the max FFT and Memory to Use sizes. I think I got away with 8192 for the FFT. The amount of RAM in your machine will determine how high you can set Memory. Too high, and the test will crap out. I think (on a Win7-64 box) I had to leave about 2 GB for the system.

If RAM testing is a major concern I strongly recommend Memtest86+. http://www.memtest.org
Download the ISO and burn it to a CD. Boot from the CD and the program starts automatically. For certainty you should run it for at least an hour. More is better.

I am not sure memtest is of any use by itself. I could loop that for months and not get a single error, and Prime95 would crash within five minutes.

The main question is which FFT sizes to use and what exactly it does from stress-testing perspective?
I currently have 4GB memory, so I told the program to use 3 of those.

[Dubslow]

The smaller the FFT, the less memory it uses. I'm not sure of the exact conversions, but FFTs in the <10K range should fit into the L2(?) cache, and thus not stress the memory at all. "In-place" means not moving the data around, putting the results in the same place as the start values; so if you want to test memory, don't use in-place, and use larger FFT sizes.

The flip side is if the small FFT lengths (< 10K) crash, then it's probably not your memory, especially since you can pass memtest.

[axn]

Originally Posted by Dubslow

so if you want to test memory, don't use in-place

I am not sure of the logic behind this. Isn't the memory more likely to fail if the same bits are being repeatedly hammered?

[Dubslow]

Originally Posted by axn

I am not sure of the logic behind this. Isn't the memory more likely to fail if the same bits are being repeatedly hammered?

If you're hammering section A but there's an error in section B, in-place won't find it, but not-in-place will eventually find it.

[LaurV]

Originally Posted by Octopuss

I currently have 4GB memory, so I told the program to use 3 of those.

This may be the boomer. On win7-32, that is certainly too much. Use 1024 for the program, there would be about 1.8Gigs for the OS and another 500-800k for the "kitchen" (video, perif, etc), from a maximum of 3.5G you can access (the rest is not accessible under win32 OS). On win7-64 you are a bit over the limit, the OS would need about 1.2G to survive (depends, it is pro, ulti? what do you have else in the memory?). Why don't you try with 1024 first, and increase it slowly. If you do not plan to do a lot of P-1, then you don't need to allocate so much mem to P95. To stress-test the memory, P95 is not the best choice.