Computer is failing Prime95 torture test even when not overclocked!!!

[mh454]

I've had my new system for about a week now and began overclocking right after I built it (3000 Winchester with 1 gig of Corsair value RAM). I worked my way up to 235x9 with DDR333 memory setting and thats the only settings I changed. System seemed to be stable but heard about this Prime95 program that tests system out. I downloaded and ran the blend torture test and it failed after about 3-4 hours. The error is the same every time it fails:
"FATAL ERROR: Rounding was 0.5, expected less than 0.4"

I then started working my way back down in 5 mhz increments but it would still fail but would go a little longer. Finally went back down to all factory auto settings and it lasted 7 hours. Why is it failing on stock settings? Temps were around 105F max regardless of settings.

 

[CheesePoofs]

Whats the rest of your computer?

If it fails in the same place everytime then it is probably a bug in prime95. If not, try running it with just 1 stick of ram. If that still doesnt help, it could be that you have a defective CPU.

 

[GuitarDaddy]

I have heard reports of some of the earlier Winchesters that wouldn't prime at stock settings. There have even been reports that Prime95 doesn't work well for Winchesters. In my experience overclocking, I can reach a point where Prime fails, but everything else, and I mean everything Benchmarks, Games, etc.. work perfectly. However my week 41 Winchester 3200+ will prime overnight at anything up to 2565mhz.

I would first eliminate ram as a source of concern by running memtest for a few hours.
Running the prime blend test, has a ton of memory test, so at this point you don't know if the problem is ram or CPU, and it could be something as simple a loosening the timings on your ram or upping the DDR volts a little bit. If ram tests out okay

I would then test the CPU with other benchmarks, Super PI 32m test is a good one, PCmark04 also has a pretty heavy CPU test, and do the Prime95 small files test (#1)
And if you determine that it is the CPU will not pass prime at stock settings, I would RMA that sucka a get another one.

 

[cockeyed]

I have a P4-3.0E that I used Prime95 on and when OC'd to 3.9ghz, I get the exact same error message as you do; "FATAL ERROR: Rounding was 0.5, expected less than 0.4". Seeing your post, I thought that seemed strange, that I get the same error. Just running my regular sw, I don't notice any problems at 3.9ghz, but Prime95 keeps saying there is a problem. AT 3.0-3.6ghz I don't get the error. If your system runs fine otherwise, maybe Prime95 has a bug.

 

[gobucks]

on my system, i get that error when I've clocked my memory too high, but lately, I also get it whenever i try to run at 1T timings, regardless of frequency. I am using a Chaintech VNF4/Ultra, and even at DDR333, it can't handle 1T timings on my corsair value ram, which should be able to hit 1T timings at least at DDR400, maybe higher. Not sure what the problem is, though. Anyways, i suggest trying to run your memory at 2T command rate and see if that passes. I'm thinkin there might be a bug with Prime95 accepting 1T timings from a winchester, but that's just a guess.

 

[mh454]

Well last night I started the test again (still at factory auto settings..no overclocking) but set it too small FFT settings (less RAM) and it went all night and I just stopped in now (12 hours total) without any errors. Now its looking like its the RAM thats giving me the trouble. I think I might bump up the FSB to 240 and put the RAM divider on 5:6 (DDR333) and run the same small FFT test again. If it is the RAM that is giving me the trouble, what could I do to try to fix that? Someone mentioned loosing RAM timings but I have no idea what to try. I have an ASUS AV8 mobo and I think I found where I change those settings (says stuff like TRC,TRFC,TRCD, TRWT, TRAS, TRP, TWCL and some other stuff like CAS and Asynclat). Oh and the 1T/2T setting is on auto right now. Does anyone know which one it goes to when its on auto?

 

[rdegler]

This is not really an answer, but I get the same error on the blended test. and can run small FFT for hours.

There are many posts on the prime95 forums with the same error. from my reading about 40-50% solve the problem with RAM changes and the remainder try known good RAM and continue to have the problem. I appears in AMD64's, faster AthonXP's, and faster P-4's.

the majority like myself pass MEMTEST and everything else runs fine. Since I am not interested in OC'ng and my system has yet to give me a BSOD I am OK with this error in Prime95.

My goal with running PRIME was watching CPU temp under load. I thought the 59-64 degrees was too high.

I might replace my RAM with higher quality when I move from 512 to 1 GB but will live with this in the mean time. That is what many other with this problem posted as well.

Good Luck

Read STRESS.TXT it gives good advice and the PRIME95 forum's URL if you want to learn more

roy

 

[Toro 45]

Sounds like your ram timings, if you have ram timings on auto it may be setting it to aggresively. Corsair Value ram runs good at 2.5-3-3-7or10 on the last one. Start @200 for your ram speed, then run memtest on test 5 for an hour or two and see if you get any errors.
The Prime blend test does stress the ram, small fft more stress to the cpu which held up fine in your testing. So I think you just need to get your ram timings setup better.

 

[Avalon]

Originally posted by: Toro 45
Sounds like your ram timings, if you have ram timings on auto it may be setting it to aggresively. Corsair Value ram runs good at 2.5-3-3-7or10 on the last one. Start @200 for your ram speed, then run memtest on test 5 for an hour or two and see if you get any errors.
The Prime blend test does stress the ram, small fft more stress to the cpu which held up fine in your testing. So I think you just need to get your ram timings setup better.

Agree, it sounds like it's your RAM. I'd suggest 2.5-3-3-10 timings for your value ram, and see if you can't get 2.7v running through it just for a little bit of extra stability.

 

[sangyup81]

I only use the first Prime95 test. All the other prime95 tests use your RAM. I would rather test my RAM and CPU seperately using memtest86 for the RAM.

 

[imported_Kensai]

Don't use Prime95. Go with SuperPi.

 

[richardrds]

Originally posted by: Kensai
Don't use Prime95. Go with SuperPi.

Stupid advice!!! You are telling him to ignore the stability pblm that Prime95 found and use a differant test that might not show the pblm exists. If you would have said use SuperPie along with Prime95 to make sure your system is stable then that would have been acceptable.

Bottom line is somethink in your system is causing the mathamatical equations used in Prime to fail, you PC is not stable with your current hardware, period!!!!!!!

 

[slash196]

To OP: Are you using the Ultra-D? Because me Corsair Value Select was fubared on that board. It wound up shorting it out. So if that's the board you're using, try some different RAM.