Prime95 Round Off error...

[netarc]

So I was all excited to have my TBird 850 running @ 1050 ... and stably, at that, from what I could tell running SETI, SciSoft & T2.

However, I thought to try out Prime95, since I understand that's one of the best stress-testers...imagine my surprise when I run the Self-Test, and get the following error:

[Thu Apr 26 09:17:05 2001]
ERROR: ROUND OFF (0.4999996096) > 0.40
Possible hardware failure, consult the readme file.


Doh! There go the good times :(

I proceeded to try different combinations of FSB/Multiplier/voltage, and now the fastest I can get the system (and not experience Prime95 errors) is ~1007Mhz ... bugger!

My question is this - what exactly does this error imply? From what I could tell, even running @ 1050 everything seemed to be a-ok - Win started ok, T2 ran fine, 3Dmark was good, as was SciSoft Sandra's burn-in test. So what does the Prime95 error mean?

Also, I'd like to try out BURNK6, which I understand is also good at stress-testing Athlon Tbirds ... anyone know where I can find this util?

[Big Mike]

It means you were in fact pushing the chip too hard, Prime tends to pick up on minute errors, it might NEVER make a difference running an application but it could well crash at an inoportune moment or at the very least mess things up, id suggest you keep it down around 1ghz for now, better cooling could let you push the chip further though.

[Nagorak]

Prime95 errors out much earlier than any other apps...It's your choice if you choose to run it at a faster speed than P95 runs stable, but personally I wouldn't trust it. It's nice to have speed, but IMO, it's not worth it if your system is not 100% reliable. I want to be confident that I can throw anything at my CPU and not have to worry about any errors occurring. That's why my CPU is running at 1400 MHz instead of 1470 or so that it will run at, but not be completely stable.

In reality that 50-70 MHz is not going to make a big difference. Sure, the difference between a Pentium 50 and a Pentium 100 was pretty big, because the P100 was twice as fast. The difference between 1000 MHz and 1050 MHz is only 5%, so I wouldn't really be too concerned about the slower speed.

Also, keep in mind that you might not be helping the SETI program out much by running it on an unstable CPU. If your CPU is returning errors, then any data you are analyzing is worthless. Sure, they'll check it against other people's results and just throw out any erroneous data, but that doesn't make it any more worthwhile.