Which stress test?

[Talrinys]

This is probably an extremely obivious question, so please gimme a chance:
Running Stress Prime 2004, and have two questions:
1: Which test do i use? I've always used the small FFT one, seeing as it usually makes my CPU fail, while blend always stays stable.
2: The other day i ran SP2004 in small FFT mode for 3½ hours before going to bed, with no problems, and a steady temperature. Now(without any kind of changes whatsoever) it fails about 15 minutes in, and that's actually with a 3 degrees lower temperature the other day due to the weather being horrible now, how can that be?

Thank you

 

[mk17]

This is probably an extremely obivious question, so please gimme a chance:
Running Stress Prime 2004, and have two questions:
1: Which test do i use? I've always used the small FFT one, seeing as it usually makes my CPU fail, while blend always stays stable.
2: The other day i ran SP2004 in small FFT mode for 3½ hours before going to bed, with no problems, and a steady temperature. Now(without any kind of changes whatsoever) it fails about 15 minutes in, and that's actually with a 3 degrees lower temperature the other day due to the weather being horrible now, how can that be?

Thank you

Now that's what I call flame-bait .

But seriously folks, that's very often why you see claims for longer and longer validation cycles. 8hrs, 12hrs, won't be long before 24hrs prime stable is called for. I always run 12 minimum on a new build, but I've seen people run overnight then restart in the morning and fail almost immediately.

I wouldn't classify 3½ hours as stable, but it's very often down to personal preference and usage profile. If its a browsing/gaming machine then a crash is a minor inconvenience.

If its for your work or your dissertation, then you may be looking for something more robust.

 

[Talrinys]

Now that's what I call flame-bait .

But seriously folks, that's very often why you see claims for longer and longer validation cycles. 8hrs, 12hrs, won't be long before 24hrs prime stable is called for. I always run 12 minimum on a new build, but I've seen people run overnight then restart in the morning and fail almost immediately.

I wouldn't classify 3½ hours as stable, but it's very often down to personal preference and usage profile. If its a browsing/gaming machine then a crash is a minor inconvenience.

If its for your work or your dissertation, then you may be looking for something more robust.

A crash is absolutely not a problem, rather trying to find out why it can be 3½ hours stable one day and not last 4 minutes the next on exact same voltage settings, speed and with actual lower temperatures. And it never crashes in games, haven't experienced any so far, and that's after about 8 months with various clocks on the CPU.
Also, any idea why my Tuniq Tower cools less than my previous Zalman cooler, we're talking 4-5 degrees?

 

[mk17]

I think normally the Tuniq's are very good, but you may need to lap the base of it. Whether you want to do that for 4-5 degrees is another question entirely though!

You can run some specific stress tests to evaluate exactly why you're crashing. For example memtest will specifically stress your ram, prime95 v25.4 on small ffts will test your cpu. Narrow down the component and take the appropriate action.

If its your ram, you might want to up the volts, or loosen your timings. If its cpu, you most likely want to boost volts, possibly also to the northbridge/mch.

If you decide to run the tests you can post the specific results and someone here would be able to help out.

If it were me, I'd leave small ffts running overnight just to really, really, rule out the cpu. My next step would be memtest - your ram shouldn't have any problems at 830Mhz, but you may want more volts going through it, or drop to 5-5-5-15.

What mobo have you got, current volts, full-spec?

 

[mk17]

fwiw, have a look at this thread for some great advice and a blow-by-blow account.

 

[Talrinys]

Have a P5B-E mobo, it sucks to put it nicely.
Has a huge vdroop, and i think that might be it, it seems like it varies how much it drops from boot to boot, up to 0.03 volts, which is quite a bit. I suppose i should just set the voltage even higher, to make sure it always boots up as stable.
And i'm 100% sure that it's not my RAM, haven't run memtest but i'll do that when i get home, i have however run the Rightmark Stability Test for hours, and it should run just fine with 2.1v.

My CPU is set to 1.58 volts, which is crazy of course, but with that setting it drops down to 1.42-1.44 which is extreme.

 

[mk17]

That's a massive vdrop - I think you've already isolated the weak-point . Any chance of upgrading to something P35-based?

 

[Talrinys]

Not just yet, but i'll probably just hold on on this speed, it hasn't crashed once in anything, just not primestable
Never once though of overclocking when i bought the system, so it doesn't surprise me that i didn't pick the best board anyway

Will get something when Nehalem comes along i think, or a Penryn quadcore.
I can probably just get the vcore well over 1.6v, with the vdroop that'll put it in the middle of 1.4v range, but that's just stupid really, no board should be that bad.

Thanks for your advice, suppose it's simply new-mobo time for me at some point