Quote Originally Posted by NEOAethyr View Post
I take back my words on the 1st test being useless.
Been having troubles with my system lately and the other day before I went to goto bed I was failing y cruncher instantly.
I got her stable the next day by adjusting the drive strenths a little.

What I'm happy about is that if I would of went to bed, and if it failed the 2nd test, I wouldn't of been happy.
So that 1st test is cool in that respect, I didn't waste any time.
You might still fail in a half hour or an hour, but still..., when it was failing almost instantly, I knew more work was ahead of me and that I should just goto bed and deal with it later.

I've got her to run an 8hr run at 1.5gigs(I have 2gigs of mem).
And that's pretty much what I try to check for now.
If I'm going for a quick run, it's 2 hours, pretty close to where it repeats the 1st set of tests on a 3.5ghz setup with 6 cores.

I like the ability to choose the memory limit from inside the program.

I really like the talk of cache misses and fpu usage.
I defently think that these are the key in checking stability:
Fpu, sse, cache missing, and large workloads to work the ram and keep the cpu with fresh, "real" data.
Same goes for ram in that respect, can't test ram with generic values, you gotta use somewhat random values.

With the cache misses you force the cpu to re-read the mem, stressing the cache and the mem more.
Might take a few ns's of stress off the cpu though but I'm sure that would be fine as long as the cpu is being fed a decent workload to continuesly go through.

I'm defently keeping this program as one of my main stress testing verification apps.
Along side prime and s&m.


So far to be onest.
I haven't found any use for linpack or linx.
I always seem to pass those pretty easy.
I don't think they stress the fpu or see unit's enough or something.
So far, if I am interpreting your post correctly amidst all of the spelling errors and stuff, LINPACK should actually stress the FPU the most.

IntelBurnTest does it a little better in the sense that it's parallel, whereas the original LINPACK is just single processor. There is a LINPACK HPL, but I've never learned how to use it or use it properly.

And I don't know enough about coding/programming to rewrite LINPACK to create LINPACK MPI.

If there was a version, I think that it would be very useful.

And if you want it to use SSE, that will be compiler dependent.