Odd behavior testing ram with Prime95

[dderidex]

Question - I have some OCZ ram that is behaving...oddly:

2gb (1024mb x 2) OCZ Gold PC4000 (CL3-4-4-8) 2.8v

Here's the thing, I've am using a simplified test case after trying to isolate anything else that would cause a problem. I've tried using 3 different CPUs with this, too (AMD Athlon64 X2 3800+, Athlon64 X2 4200+, Athlon64 X2 4800+).

The BIOS of my Shuttle ST20G5 allows for the new dividers to be set. IE., I can set the ram to run at 216, 233, or 250 with the HTT link staying put at 200mhz and CPU at stock clocks.

At ANY of those settings, Prime95 (two instances, each set to test roughly 1/3 the system ram) will fail almost immediately on one core - always within one minute. If I set the ram to run in sync with the HTT link (200mhz), I'm Prime95 stable for 24 hours+ (I hit 25 hours and stopped it, no errors).

If I try keeping them in sync and *overclocking* the CPU - even to 220 HTT link, it fails in the same manner as above. If I set the ram to a lower ratio (166, or 5:4), the CPU can be clocked up to 240 HTT link without any problems (well, the CPU won't stay quite stable at this speed, but it stays up 4 or 5 hours or so). Given that 220 HTT with the ram in sync causes the memory test to fail under 60 seconds, I'd take this as a compelling argument.

So...does this sound like bad ram? But, why so specific a failure point? I mean, 200mhz and 200mhz ONLY...with DDR500 ram??

Alternatively, is there possibly something funny with how Prime95 works in conjunction with higher ram clocks?

 

[(cf)Eclipse]

weird. what ram slots are you using? did you actually set ram voltage to 2.8v? are the sticks hot?

 

[dderidex]

weird. what ram slots are you using? did you actually set ram voltage to 2.8v? are the sticks hot?

Hard to say on the sticks - they are right under the PSU (literally, like, mm away), so the whole area does get rather toasty.

I've set the BIOS to 2.8v, yes.