Failure on blend test after 1 minute!

Hi all, just thought I'd let you know I have found a solution....

The motherboard was actually giving out a lower voltage to the CPU than it thought it was, so I increased that to where it should be and hey presto.... No further errors!

I get exactly the same problem running the Blend Torture Test:

Test 1, 4000 Lucas-Lehmer iterations of M19922945 using 1024K FFT length.
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.

Running Gigabyte GA-K8NSNXP-939 Mobo, Athlon64 3500+, 2GB of OCZ RAM. Not overclocking at all, running 200Mhz FSB.

I get the same error in blend but small FFT flies by itself and large FFT flies by itself (large FFT appears to be the same test as the first test in blend: 4000 iterations of M19922945 using 1024K) what gives? I want to know that my system is stable too but am I beating my head against the wall for naught?

[cartmanea]

I've been having a similar problem for a little over 6 months now

My system used to be perfectly stable, passing over 24 hours of the blend test. Now it randomly fails in blend, but running in place passes just fine. Also, sometimes it fails right away, sometimes after a couple hours, sometimes after 10 hours

I've gotten the same results on two totally different setups, one an AMD Barton System, and the other a p4 2.8c system. The only thing similar between the two setups was that I used the same ram, 2x512Mb Mushkin BH-5.

However, the ram can pass all the tests in Memtest86 looping all night without errors. Even slowing the ram down and loosening the timings doesn't help. One thing I have noticed, is that in BOTH setups, if I take one stick out (either stick) it seems to run fine. Also, if I limit the amount of ram used by Prime95, say to 768Mb instead of all 1024Mb, then it also passes.

Oh yeah, and I've tried another 2x512Mb set of ram - same results.

Seems like this is a software problem to me. I get no errors in any other Windows stress tests, only Prime95. I've tried older versions too with the same results. Sometimes instead of Prime95 throwing an error I actually get a Windows Fatal Error message that says "Prime95 has a fatal error and needs to close". That happens even when my entire system is underclocked and has happened on both my p4 setup and my AMD setup. I've also tried a fresh download of Prime95 in case my copy was corrupted....same result.

I've been waiting for a new version to come out so I could try that, but it seems 23.8 has been out about the entire time I've had the problem.

I've also tried Safe Mode with the same results, and have reformatted my hard drive countless times and done a full reinstall. Also have used both ATI and nVidia video cards on different installs, so I know its not the video card or video driver.

I'm going insane....WTF is the problem!?!?!?!?!?!?

Had all the same problems with asus A7NX deluxe mobo and crucial 3200 ram. I had the same problems with my first athlon 2600 and my upgrade to 3200.
I had every error in the books, hardware, user errors,application errors etc.
I had a hunch it was the memory. I tweaked the spd settings to no avail.
I reduced the fsb lower and it ran fine. However, It was slow and I couldn't stand. I did alot of reading on Tom's Hdwe and it confirmed in my mind that it was the memory. I even tested the cucial memory I had and it tested ok in microsoft tool and always failed in prime 95. I took a chance and put in corsair xms ddr 2 memory and it solved all the problems everywhere and it's fast.
I will no longer buy regular memory again. It's worth the additiona cost. Besides the fact that corsair is recommended by asus and Amd. Now it is possible that the faster mem from crucial could work as well as the corsair.
Some memory producers tell you what to set the spd at in the user mode.
Corsair does and cucial doesn't . Crucial just lists the class 3, maybe on the faster memory they do too.
ae: 2.5 3 3 3 5T

[cartmanea]

Well, as it turns out, one of my sticks of BH-5 had gone bad. I tested them individually with Prime95 and one passed over 24 hrs, the other failed within minutes. Funny thing is, memtest86+ didn't find errors in either stick. I guess it doesn't fully check every aspect of memory stability. I even ran the extended tests on them and it found nothing :(

I just wish I would have tried them individually 9 months ago :( Memtest passing made me totally write off the fact that it could be a memory problem.

I'm having the same problem - Getting this error after less than a minute:

"FATAL ERROR: Rounding was 0.5, expected less than 0.4"

It has to be a software bug,

I had put my RAM through overnight tests both in memtest86 (v1.55) and Microsoft Diagnostic Tool - the extended test and extended memory map.

Spec:

Athlon64 3500+
GV-K8NSNXP-939
Corsair TwinXP1024-3200XL

[Mystwalker]

It's interesting that a lot of people with this error think that a software error is the only possible solution. After all, this kind of test (which has next to no variation in it) has been run millions of times on a lot of different architectures, and only a tiny fraction doesn't pass it...

Did you consider that the CPU could be the source of the error? In that case, neither memtest nor MDT would find anything.
Even in case of a memory issue, history has shown that prime95/mprime is a tougher test then memtest. IIRC, Memtest+ shows better performance when it comes to error detection...

I'd propose taking out one memory module and try again. With non-overclocked systems, DualChannel access is one of the most likely sources for faults.

[cartmanea]

Unregistered Guest, my ram passed every Memtest I could throw at it, basic and extended for hours on end. Prime95 found an error with one stick though, so I wouldn't be so sure its a software bug.