Asrock K7VT6 (1.3 bios)
3200+ XP (Barton)
2 x 512MB DDR400
500W Super Flower PSU
When I run prime95 benchmark it always fails with an application error before finishing it's short 40 second run.
Using the bios I increased the CPU core voltage by +5%, from 1.84v to 1.92v, and immediately the prime95 benchmark runs through every time. When I switch back to default voltage it fails agagin. So now at the +5% I've tried the prime95 torturetest and it runs for about 90 minutes before reporting that I have a hardware error as an inaccuracy in a calculation has been found.
So now I'm currently trying the torturetest at +10% voltage (2.00v) and the CPU is running at 60'C.
My question is:
Should I consider this CPU faulty and in need of replacement, or is this sort of increase in voltage to achieve stability acceptable and part of life?
<P ID="edit"><FONT SIZE=-1><EM>Edited by marky1124 on 10/27/04 05:03 AM.</EM></FONT></P>
If you paid for a new CPU and it is failing then you should RMA. Does it fail on other benchmarks like 3dmark or pcmark or is it just prime95? You shouldn't need to increase the voltage for stock speed stability. Also if you do end up returning it, don't forget NOT to mention you did the voltage because it voids most warranties.
Thanks Blasphemy. Without the +5% it fails on 3D Mark 2003 and 2005, sometimes with application error, sometimes with blue screens. Game benchmarks fail as well.
With the +5% I've not witnessed a failure of one of those 3D benchmarks. That said I've only run them each a couple of times. I've concentrated more on the prime95 since I felt it focused in on the CPU/Memory/Temperature/PSU and ruled out the GFX/Drivers etc...
I tried Sandra burn-in wizard and it ran the CPU/Memory tests for 7 hours straight with no problems. I guess if there were tiny problems they weren't enough to cause the application/OS to crash. I like the fact that prime95 knows the results of the calculations and consequently is verifying the results not just running until the application dies.
Is there another application like prime95 that perhaps I should use to test the machine as well?
Thanks Blasphemy. I forgot to mention I have tried the Sandra burn-in wizard which ran successfully for 7 hours.
I was thinking the key difference might be that prime95 knows and double checks the results whereas other burn-in programs may just perform work but not validate it against known good results. Hence I'd like to try another program that does validate the results, if I can find one.
Well if I were you, I would just RMA because Prime95 is a well respected burn in for systems, and if yours isn't passing it then it's a real possibility your CPU is a lemon. I mean, you could go try and find a similar working program similar to Prime95, but you may just find the same results in the end that you started off with in Prime95. Good luck.