Memory Stability Testing Q and Heat Q

[LOUISSSSS]

i got my CPU OC stable and am trying to tighten timings now.

should i just use the Prime95 Blend test? and how long should i run it for? i already know my CPU oc is stable. so is like 30mins enough of testing?

or is there a better Memory stability test out there?

and i've set my VRam to 2.1v. when i touch it, its like holding a hot cup of coffee, is that alright? ram is ballistix ddr2 800

 

[BonzaiDuck]

Barring the use of other testing programs like "Serj's [he's Russian] S&M or SnM", I'd say that the Blend Test is the obvious choice, beyond MEMTEST86 or ____+.

To "find" a possible "setting," run it for two hours. If it fails, tweak it again and run it for two hours. After you've decided on a preferred setting, some would argue you should run it day and night for 24 hours, but I think 6 to 8 hours on the memory is sufficient.

Since I'm also using Crucials, I find your subjective "burnt-ma-hand" observations interesting. I'm pushing a lot of air past my RAM modules just to cool them down, but never had an idea of just how hot they got.

RAM is supposed to be one of the cooler components, but I can see where pushing the voltage a few notches higher would make them sizzle a bit. . . .

 

[BonzaiDuck]

Remain reluctant to do it, and do it carefully -- but as you tighten the timings, you may have to nudge up the voltage a notch.

I'm just discovering that I can run my DDR2-800's at 800 Mhz 4,3,4,8,2T (with a 5:6 divider ratio) at voltage lower than 2.1V, but I need darn near 2.2V to get them to run at 667 Mhz and 3,3,3,8,1T.

 

[LOUISSSSS]

i ran Prime95 Blend test for about an Hour and they were fine @ 395mhz and 4-4-4-12 (stock speeds) (Bios Vram @ 2.1)

they're not burning, but its warm like holding a cup of dunkin donuts coffee

 

[MadScientist]

Blend is the right setting but 1 hour is not enough, IMO, and there are plenty others, i.e., the Duck's, on how long to run Prime95, 6 hours minimum, 12 hours better. I have found ram overclocks and settings that were Memtest86 stable for hours but failed Prime95 blend.

 

[BonzaiDuck]

Yeah -- I once had a failure with ORTHOS on C2D E6600 after 9 hours.

I think that N of failures for any voltage setting V is Poisson-distributed over time. At the right voltage, the distribution flattens out or disappears.

With V at lower setting, the distribution bunches up closer to the start-time. So you KNOW if you find it occurring after 9 hours, that bumping up one notch, or maybe adjusting termination voltage, should eliminate the "probabiility" of it happening at all.

 

[LOUISSSSS]

well i am running my ram at 2.1V while the rated is 2.2V.

figured 2.1 is close enough for 395mhz @ 4-4-4-12. ballistix are rated for 400mhz @ 4-4-4-12.

 

[Idontcare]

Originally posted by: LOUISSSSS
i got my CPU OC stable and am trying to tighten timings now.

should i just use the Prime95 Blend test? and how long should i run it for? i already know my CPU oc is stable. so is like 30mins enough of testing?

or is there a better Memory stability test out there?

and i've set my VRam to 2.1v. when i touch it, its like holding a hot cup of coffee, is that alright? ram is ballistix ddr2 800

Memtest+ v1.70, loop test 5 for about 10 loops. If no errors reported then you should be golden.

http://forums.anandtech.com/me...=1901991&enterthread=y

 

[BonzaiDuck]

One way to look it it: less in the way of marathon PRIME95 run-times makes more sense if the OC you're trying to achieve is not a fixed setting you plan to make "default".

But it's still interesting that you can stress your system for 8-hours with "0 err, 0 warn" -- and an extra hour would turn up a failure. And as I said -- under those circumstances, a bump up in voltage to the component (CPU or RAM) -- or a slight tweak to the other mobo-component voltages -- might mean stable, perpetual operation. PRoblem is, re-testing is gonna cost time and patience.

That's why it's probably useful to methodically isolate -- through your settings and strategy -- all the causes of failure so that you know what caused something that took so long to find . . . .