Hardware Problems--Please Help.

[TrueWisdom]

I'm running the following machine:
P4 2.4C at stock speed
ASUS P4P800 Deluxe (865P chipset)
1 GIG OCZ PC3700 Gold Rev.2 Matched Pair (Running at PC3200 settings 2-2-3-5)

I cannot get the blend test to pass--it crashes under 10 minutes every time I run it. It first crashed around 2 weeks ago, and to follow it up I ran memtest and discovered errors in my RAM. (I used to have Corsair XMS PC3200 1GB Matched Pair) I sent that back and purchased the OCZ. As soon as I got the OCZ, I underclocked it to PC3200 and set it 2-2-3-5. (The settings OCZ recommends.) I then ran memtest86 for exactly one full cycle (running all tests, not std, for ~1 and a half hours) and received no errors. I ran it again to make sure and got the same results. The other two Prime95 tests both return successful first rounds. I plan on running each overnight to see if either fails, but here's my question:

I know Large FFT tests CPU and Blend tests memory--why would Large FFT pass the first round of tests, and Blend fail? The memory isn't bad; I've tested it repeatedly in memtest and no problems pop up. Any suggestions? I monitored my temps and the hottest the P4 ever got on the Blend was 96 Fahrenheit (39 C), and it gets much hotter on the Large FFT test (104 F, 40 C) and no problems occur. What is unstable in my system? My chipset temps are constant too. (31-33C) Any help would be greatly appreciated. Thanks!!!

 

[TrueWisdom]

Ok, I'm an idiot. The tests I am referring to are those in Prime95.

Sorry if I caused any confusion!

 

[vietofmars]

http://www.mersenneforum.org/printthread.php?t=1588

 

[TrueWisdom]

Thanks for the thread!

Now that I'm running the latest version of Prime95, I'm getting fatal errors. It occurs within 10 minutes on the Blend test, and in one hour and 33 minutes in the Large FFT test.
[Wed Mar 10 22:53:44 2004]
Self-test 8K passed!
[Wed Mar 10 23:22:55 2004]
Self-test 1024K passed!
[Thu Mar 11 00:20:32 2004]
Self-test 1024K passed!
[Thu Mar 11 00:36:57 2004]
Self-test 896K passed!
[Thu Mar 11 00:53:16 2004]
Self-test 768K passed!
[Thu Mar 11 01:09:45 2004]
Self-test 640K passed!
[Thu Mar 11 01:26:08 2004]
Self-test 512K passed!
[Thu Mar 11 01:37:08 2004]
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.

Temperature at time of crash was 41C. Does this mean I have a bad CPU? I'm going to burn in my memory overnight by running memtest for a while. Man, I hope I don't have to RMA this CPU...

 

[TrueWisdom]

Ok, so I tried running it at timings 2-3-3-6 and it still failed in 6 minutes. (Citing a rounding error.)

I tried running at 3-4-4-8 and it froze before completing the first test. (under 5 minutes)

I'm so confused. Why would memtest pass this memory without fail after an hour and a half of continuous stress testing, but Prime95's Blend test crashes within 10 minutes? Do you think it could be my chipset drivers? I'm using the latest 865 drivers from Intel.

Is there any good program to JUST test the CPU? The fact that I failed Large FFT in an hour and a half makes me suspect that the CPU is the culprit, but the fact that I froze after relaxing the memory timings makes me suspect something else may be up.

Ugh. I'm not even overclocking. What a crappy situation.

Thanks for all your help, and please keep suggestions coming! I think my next step will be to run memtest overnight to see if it can find anything at all. If that doesn't do it, then I'm going to have to assume my CPU is bad...

 

[Fern]

If I were trying to focus on my cpu, I would set my ram to "optical", or manufacturers specs for stability and make sure it was up to snuff by passing memtest86 overnight. If OK I wouldn't mess with again til I was happy the cpu was OK.

Only other thing I can think of is to check some mobo specific sites and see if that mobo is picky and doesn't play well together with OCZ

 

[Matthias99]

I'm so confused. Why would memtest pass this memory without fail after an hour and a half of continuous stress testing, but Prime95's Blend test crashes within 10 minutes? Do you think it could be my chipset drivers? I'm using the latest 865 drivers from Intel.

The memory in my current system is the same way when I push the timings too tight. Passes memtest86 just fine, but fails the 'Blend' torture test within minutes. I suspect it's because the northbridge is getting much more of a workout with Prime95 than memtest86, or because the temps are higher while Prime95 is running. The 'in-place FFT' test in Prime95 should pretty much JUST test your CPU/cache -- it uses so little memory that it fits wholly within the processor's L2 cache. If you can pass *that*, your CPU is probably OK. I suspect that either you're having MB problems, or the new RAM you got is a little flaky.