System freeze after hours of heavy load

[Fenton]

Hi All,
I would like to get some help, how to approach, how to find and localize my problem, which is the following :
I use MATLAB for heavy numerical computations, which really use all 4 cores (AMD Phenom ii x4) in 100 % continously, and deals with large amounts of data (matrices of 1-4 Gb).
The system freezes after a few hours (typically 1-4 hours) of this kind of load.
It doesn't seem to be software dependent, as I have windows and linux also installed on the same machine, both running the same code, and both freeze (altough not always after the same amount of time)
I logged the temperatures using everest, and they never exceed 50-52 C for the cores, and 37 C for the chipset.

setup:
AMD Phenom ii x4 945
ASUS M4A78-EM
8 Gb Corsair XMS2 6400 (5-5-5-18, 1.8V) memory in 4x2 Gb
LG DVD burner
Samsung F1 500 Gb HDD
OCZ ModXStream Pro PSU 500W.
ArcticCooling Freezer 64 Pro PWM CPU cooler

Nothing is overclocked.

Before installing the system, I ran memtest86+ for the RAM, and found no problem.
What I have additionaly tested by now, is that I ran the system stability test in Everest Ultimate Edition, stressing the CPU, FPU, cache, memory countinously. Voltages, and temperatures seemed to be rock stable, and the test had been running already for 15 hours when I stopped it.


hints, how to go on ? what to check ?
thx

 

[sminlal]

I'd try removing half your memory and and seeing if you still have a problem. If you do, then swap with the other half of the memory to see if that fixes the problem.

 

[Fenton]

I have tried it, but the problem doesnt seem to appear consistently.
I found a new code sample, which sometimes immediatley freezes the machine,
sometimes it can run continously for hours.
A few times, when the freeze occured, there were some weird patterns mixed with the original display.
As I use the IGP on my board, I dont know if it couldnt be an issue with shared memory, or whatever.

 

[orangegator]

Try raising the ram to 1.9V or even 2.0V. When using 4 sticks of ram, you may need a little extra voltage. However, are you sure that it's not just the Matlab code itself causing the the freeze? Can you run it on a different computer without issues?

 

[Fenton]

Yes, sure, the code runs without problems on other computers.
OK, I can try the trick with the ram voltage.
I also heard about the increase of VDDNB ... but I don't know what is it exatly, and what influence it has.

 

[Guest]

Sorry for resurrecting an old thread, but is there any change that the thread starter is reading this? Did you find a solution to this?

I have the exact same problem, but completely differenent setup, ASUS P5Q-Pro, Q9550 and 2*2gb mem. I am using matlab for factorizing large tensors and it seems that the bigger the tensor dimensions the higher the probability of crashing, so the problem might be memory related?

The cpu is sligtly overclocked to 3.4Ghz, and I tested stability using intel burntest. The temps with Matlab are not at all as high as with IBT, but it crashes quite often. I also tried increasing the cpu voltage, but at least sligth increase didn't help. I'm not very familiar with overclocking and voltages, so basicly I have everything else at stock voltages, just overclocked cpu and adjusted the cpu voltage to the point where it seemed to be stable. So raising ram voltage would be a possible solution, and it is completely safe right?


/ Joonas