AMD Athlon 64 X2 4200+ unstable @ stock

[Crunchy]

Hi, I need a little help figuring why my CPU is unstable.

My AMD Athlon 64 X2 4200+ is unstable when running two Super PIs, they always spit out an error before finishing. Everything is stable until the CPU is put under stress: one Super PI is ok but two won't run. I can play Sims 2 for a couple of hours and then suddenly get a crash / BSOD.

I've run memtest86 and let it do a couple of sucessful passes to rule out a memory problem.

I have a decent cooler attached to the CPU: at idle its about 30C and under load it doesn't go higher than 45C - one core is usually a few degrees warmer than the other.

I'm running Windows XP SP2 x86 and tried with and without AMD drivers / dual core optimzer / windows dual core fix.

My thoughts:
a) I might take out the CPU, reseat it and reattach the cooler. I don't think this can be the problem though as the temps are low?
b) The PSU might be at fault, in CPU-Z the voltage always seems a bit lower than what was set in the BIOS and can fluctuate a bit (+/- 0.05v). I tried overvolting and even undervolting and it doesn't improve things. In the past I've had problems with the PSU turning off in hot temps even though theres plenty of cooling in the case. I even exchanged the PSU and still had the turning off problems some times.
c) The mobo could be at fault, I was thinking of switching to an Asus A8V anyway because of the better driver support from Via.

Can anyone give me any advice on further tests, possible fixes etc?

CPU-Z 1.39 report file
Processor(s)

Number of processors 1
Number of cores 2 per processor
Number of threads 2 (max 2) per processor
Name AMD Athlon 64 X2 4200+
Code Name Toledo
Specification AMD Athlon(tm) 64 X2 Dual Core Processor 4200+
Package Socket 939
Family/Model/Stepping F.3.2
Extended Family/Model F.23
Brand ID 5
Core Stepping JH-E6
Technology 90 nm
Core Speed 2210.8 MHz
Multiplier x Bus speed 11.0 x 201.0 MHz
HT Link speed 1004.9 MHz
Stock frequency 2200 MHz
Instruction sets MMX (+), 3DNow! (+), SSE, SSE2, SSE3, x86-64
L1 Data cache 2 x 64 KBytes, 2-way set associative, 64-byte line size
L1 Instruction cache 2 x 64 KBytes, 2-way set associative, 64-byte line size
L2 cache 2 x 512 KBytes, 16-way set associative, 64-byte line size

[josty2]

One of the most common mistakes when testing dual superpi.
The exe writes its results to a data file and when you run two instances of superpi, you have both exes writing to the same file, which will make 1 of the 2 spit out an error. Make 2 different maps with exes in there, and they will run perfect... :toast:

[Crunchy]

D'oh! Your right! More testing is needed, I just ran Orthos StressCPU for 5 mins without errors. I'll try SuperPI again. Thanks.

[Crunchy]

I've been doing some more testing this morning. I tried Super Pi again but it was still spitting out errors after running two seperate copies, I also tried without the dual core driver/optimizer/windows fix.

I ran Orthos on small FFTs for 20 mins with no probs but switching to Large it failed fairly quickly. So even though memtest86 did a couple of passes fine it does appear to be a memory problem.

I took out 2 of the 4x 512MB sticks so I'm back at my original 1GB at 2-2-2-5 and 1T command rate. Large is now running ok for 20 mins! I've now also run Blend for 20 mins or so as well. I'll run dual Super Pi again and see if that will pass at 32M.

Really weird, I expected 4x 512MB to be ok as long as I was at 2T command rate. Well I now know to stress test as soon as I change hardware and to not assume too much :D.

[breakfromyou]

ahh, weak memory controller?

[Crunchy]

Hmm, could be, I'll try some overclocking with 1GB and see how I get on.

[Hemi]

Try running memtest for windows. It will let you know real quick if you have a bad memory module or if your timings are set wrong. Memtest 86 will pass errors and memtest for windows will catch them. Give it a try

[Repoman]

Try running memtest for windows. It will let you know real quick if you have a bad memory module or if your timings are set wrong. Memtest 86 will pass errors and memtest for windows will catch them. Give it a try

+1

Memtest86 is useless compared to Memtest for Windows. It might have a use pinpointing certain kinds of memory errors by running certain tests, who knows

[m411b]

Make sure your mem is supplied the correct voltage!!

[Crunchy]

I re-read the MSI K8N Neo2 Platinum thread and there was a new hint saying that if you set Tcrd in the BIOS to 2 that it drops the voltage by 0.1 when you go into windows. I set it to 3 in the BIOS and back to 2 with A64 Tweaker and it's stable!

I'm not sure I'll use the 2GB even so as it doesn't overclock half as well. I only got it to about 241 FSB, where as 1GB gets to 280 FSB at 2.5-4-3-7 1T.