Make sure that you actually enable round-off checking. The 25.x versions have it turned off by default.
I'm still a fan of Prime95--specifically, the 64-bit version 25.x series, which loads one worker per thread. I also use small FFTs and run it for at least 72 hours w/o error (along with a 24+ hour Memtest run, and 24+ hours of Bart's Stuff Test on HDs) before I consider the machine stable. There are a few other tools now that use a bundled Linpack library. Many have reported substantially increased temperatures with them, suggesting that it may apply more stress than Prime95...but results seem to be mixed, especially on non-Intel processors. Greg > -----Original Message----- > From: [email protected] [mailto:hardware- > [email protected]] On Behalf Of Winterlight > Sent: Sunday, May 17, 2009 11:07 PM > To: [email protected] > Subject: Re: [H] testing overlclock > > > >> > >prme95's torture test is VERY good for such tests, as it not only > >loads the thing, but also makes sure its doing its sums right :) > > That is exactly what I wanted to check.. to make sure there are no > errors
