The issue I see is that most of these semi "Burn-in" tests that max out a
CPU don't stress other subsystems at the same time (such as Disk IO). A real
burn-in test exercises the entire system. Back when I used to build
computers we would do a full 12-24hr burn-in test using specific software to
max out all the components we could think of before we would ship it.

While memtest86 can get the CPU heated up a few degrees higher, it isn't
touching the disk or GPU. I think that if you had something like Prime95
running with something else maxing out the GPU (Maybe something w/ BrookGPU)
while trashing the HDs would be a good test (because these are the systems
likely to run the hottest)...

Actually, on one of my systems, the chipset on the motherboard would
overheat if you used the onboard sound (Asus nForce board, I forget the rev
A7N8X). After plopping in a different sound card, my overheating issues went
away...

Plus, I've run into weird problems before due to crappy power supply units
(PSUs) that were feeding some badly fluctuating voltages that manifested in
the most bizarre problems. Those were always the most frustrating problems
to diagnose.

Anyway, I stick by my original assessment of Prime95 pushing the hardware
and the flaw showing up then. For all the people suggesting it was WinXP or
SP2, there is no reason to suspect that because the problem went away when
the guy stopped using Prime95, which leads me to think it is a heat related
problem.

- Jeremy Blosser

-----Original Message-----
From: Brian J. Beesley [mailto:[EMAIL PROTECTED] 
Sent: Wednesday, March 23, 2005 5:31 PM
To: The Great Internet Mersenne Prime Search list
Subject: Re: [Prime] Stress test value of Prime95 vs Boinc/Folding etc

On Wednesday 23 March 2005 22:12, Greg Edwards wrote:

> Does anyone have a handle on which of these DC apps would load/overheat
the
> machine the most ?

I'd put my money on Prime95, it's optimised in a way that few if any of the 
others are. However it tests CPU and memory only, not the I/O subsystems, 
which could be a stability issue to gamers (especially the graphics card).

Note, experience with systems with a front panel CPU temp display suggests 
that memtest86 (during testing cached memory) actually gets the CPU a few 
degrees above what Prime95 manages. The problem with Prime95 is that, in 
order to do useful work, pipeline and cache misses do occur in a way which
is 
not neccessary to a memory tester.

> I don't think any of the ones above have a torture test where they detect
> wrong results being computed, like Prime95 does.

Which is worth a great deal. A few compute errors might sneak through on a 
marginally unstable system if there is no checking.

Prime95's checking in torture test mode is pretty near perfect, we _know_ 
what the result should be so any mismatch... Small errors may however be 
disguised by rounding to nearest integer so there is no guarantee that 
_every_ glitch will be detected. That's why I reccomend to overclockers that

they use the speed just below the speed at which Prime95 appears to be 
stable. This safety margin also allows for high ambient temperature etc.

Regards
Brian Beesley
_______________________________________________
Prime mailing list
[email protected]
http://hogranch.com/mailman/listinfo/prime


_______________________________________________
Prime mailing list
[email protected]
http://hogranch.com/mailman/listinfo/prime

Reply via email to