I have a new home built system:
-AMD Quad 9950 Black Edition, with a Noctua NH-U12P heatsink
-2x 2Gb DDR1066
-Thermaltake 750W toughpower modular
-ATI X800XL (I ordered a HD4870, but it was defective..., waiting for a replacement)
Anyways, I wanted to test stability using OCCT stress test. The result is that it always fails after about 1-10 minutes, although there is no overclocking, and the temps are stable at 45 degrees (Should be stable...). I get error code 4: Calculation error.
Other stress test pass with high colors.
I have seen in other forums not to stress, as sometimes OCCT will bug, so it's a software, and not hardware problem.
What did bother me is that some time ago, my HD4870 fried after my PSU tripped, so I wanted to monitor my power supply, during the stress test. My Vcore and 5V is rock solid stable, the 3.3 is alright (bit of fluctuation), and the 12V fluctuates a lot.
So here are my three questions:
1 - Should I be worried about OCCT errors? Can I just shrug it off to software bug?
2 - Is my power supply stable enough for my components? Is what I am seeing alright?
3 - Even if what I am seeing is OK, should I expect this kind of behavior from a High-End PSU? It's supposed to be very good quality.
PS everything is up to date, bios & everything
That's only 100mv !!!
Even the best power supplies in the worild have a tiny amount of ripple.
Typically the 12V rails are 120mv to 150mv. So what you are seeing is that ripple.
So I would not be concerned with it.
As for software, Shadow nailed it! Use prime95 it's the best and will stress your system!
Hope this helps.
I ran prime95 in mixed mode, and after 30 minutes or so, one of the workers stopped due to a round-off error. This happened 3 times.
So I guess there is definitely something wrong. The test was in mixed mode (FPU+RAM), so I tried the small FFT test (FPU only), and the test ran overnight for 8 hours with no problems. Guess this probably means the culprit is the RAM...
I'm going to run a Memtest86+ tonight to see.
I'm not sure whether or not I hope my ram is defective
Could you link me to a topic or thread about that?
Why would I change the voltages if I am not overclocking or anything?
I know DDRII specs are supposed to be 1.9V, but PC8500 is usually 2.1-2.2V right? Do I have to do this myself, or is my mother board smart enough for that?
For info, I have this memory: OCZ Reaper, 2x2Gb PC8500 (rev2)
And my mobo is an Gigabite GA-MA790X-DS4
Is there anything else I should know about memory stability, other than the obvious? My system works quite well, except for an occasional stress test failure, so probably only tweaks.
JEDECs normal voltage for DDR2 is 1.8V. DDR2-800 is usually rated at 2.0-2.2V, DDR2-1066 is pretty much the same, maybe slightly more. But some DDR2-1066 modules can't even reach their rated speeds no matter what voltage is used.
and yes, you'll need to change it yourself in the BIOS. The mobo defaults likely to something slower and lower voltage, so it might cause some instability.
Several tests using memtest86+ revealed that one of my two PC8500 sticks couldn't take 533Mhz settings.
Running it in PC 6400 fixed the problem. Apparently, they were PC6400 sticks to begging with, but were supposed to be overclockable to PC8500.
I have a few questions about ram now, but I think I'll go create a new thread, To ask questions about timings and such (but if you have any clue now). My mobo sets all these settings itself, but I manually changed the frequency mode (533 -> 400)
In PC8500, memtest displays the timings as 5-3-3-3-26; BIOS Voltage:2.160
In PC6400, memtest displays the timings as 5-5-5-15-41; BIOS Voltage:1.960
Aren't the PC8500 timings WAY too aggressive?
Anyways, I want to thanks everyone. You've helped me pinpoint the problem and solve it. But more importantly, I sleep better at night now
well that's why! 5-3-3-26 (whats with the extra 3? oh do you mean 13?) is an extremely unbalanced timing to begin with, and nigh impossible even with 800mhz! manually set the timings to 5-5-5-18 and im sure it will pass memtest at 1066 or "PC8500" as you call it. then just lower it to 5-5-5-15 and one notch at a time till it fails stress testing...
Alright, I'll try.
As for the extra three, I'm really just reporting what memtest said (which looked odd, I admit), but there are indeed 5 numbers (instead of the usual 4), and it displayed:
I'll try to change the timings again, but I'd like to test the stability of the current 800, as OCCT still gave me an error, but prime95 seems stable. Guess it's all about tinkering now.
It shouldn't even boot with the 4th number as a 3. Something's definitely wrong there. Ignoring that, it is possible that they are almost stable at something like 5-3-3-26, as although those are odd, they are almost doable. I'd go for something like 5-5-5-18 though, as stated above.