Difference Between Prime Tests

[imported_inspire]

Right, so this setup:

HTT - 300
CPU volts - 1.5250
CPU multiplier - 9x
DDR2 volts - 1.95 (board's max)
PCIe Bus - 100
HT multiplier - 4x
DDR2 - 400 MHz (1:2 Divider) @ 600 MHz
4-4-4-12 1T timings

Fails two instances of SP2004's blend test after 7h 10m 34s, but passes 24h of running two instances of SP2004's CPU stress test. Matter of fact, none of the OC's I've tried have passed SP2004's blend for more than 14h. Thoughts?

 

[krotchy]

check your ram timings and make sure they are correct using memtest. CPU test stresses your CPU, blend stresses your CPU and your RAM.

 

[imported_inspire]

Ostensibly, yes, krotchy - but, I've noticed in the task manager performance window that during a blend test, my CPUs don't ever really exceed 10% load after the test gets started. The Page File shoots up to 1.8 GB, though.

You're probably right though - my mobo can't give more than 1.95v to my memory, but sheesh - it runs at 800 MHz at stock just fine. I don't see why it should work at 600 MHz with a divider.

Maybe I'll check the baseline - drop everything back to stock and run a blend test and see if I pass...

Afterthought: That setup has no problem running two simultaneous instances of SuperPi's 32M test and finishing - doesn't that seem to imply that my memory is stable? (honest question)

 

[htne]

In my experience, getting the blend test stable is much more difficult than getting the "stress cpu" test stable. And the "large FFTs - stress RAM" is even more difficult. If the system will pass memtest for one hour, and SP2004 "stress ram" for one hour, then I am a happy camper. Trying to get your system stable under 100% load for more than 14 hours is (in my opinion) going a little too far....

 

[aggressor]

It's not going too far, it's to ensure your overclock is actually as stable as it would be at stock speeds, which is the goal for many people.

I wouldn't accept anything less than 24h prime stable.

 

[AdamK47]

Originally posted by: htne
In my experience, getting the blend test stable is much more difficult than getting the "stress cpu" test stable. And the "large FFTs - stress RAM" is even more difficult. If the system will pass memtest for one hour, and SP2004 "stress ram" for one hour, then I am a happy camper. Trying to get your system stable under 100% load for more than 14 hours is (in my opinion) going a little too far....

I would never trust a system that wasn't 100% stable. I usually run two Prime95 instances and set both to blend. If it passes that for 24 hours I move on to other test.

 

[imported_inspire]

Well, this may be my problem. I haven't done a baseline Blend Test, and I haven't even tried the RAM stress test. When I get home I may take some baselines and just see if that will run for 24 hours. If not, then I'll at least have something to compare my OC stability to my stock stability.

@adam - Jesus, man - what other test trumps dual Prime at blend for 24 hours?

 

[X2Ocer]

HT @ 1200 (4X300)...might be you need to drop it...just a thought

 

[imported_inspire]

Yeah, I thought about that, too, but... nevermind - it seemed like a good idea at the time... I'll dial it back and try again...

*grumble* Overclocking takes a long time.... *grumble*

 

[X2Ocer]

follow up and let us know how it goes

 

[imported_inspire]

Yeah, so I got impatient and after this setup ran two instances for 13 h of SP2004 Blend Test with no errors while AVG ran a full system scan in the background (my bad...), I called it 'stable enough'. I've had this box running for two weeks and I'm ready to use it. If if BSOD's I will let you guys know and endure the 'I told you so's til, then, I'm going to enjoy 2.7 GHz.

 

[aggressor]

Hopefully you dropped the multi on the HT bus so you aren't over 1Ghz!

 

[imported_inspire]

No, but I can - what specifically will that hurt?

 

[myocardia]

Originally posted by: inspire
No, but I can - what specifically will that hurt?

First, it usually makes the system awfully unstable. Secondly, and most importantly (to me anyway), it adds no performance, so you'd be much better of with it at 2x, than at 4x.

edit: Although 3x HTT multiplier would obviously be perfect. I'll see if I can find a link to those tests that someone did, with low and high multis, even around 1200. There was somthing along the lines of a 1/10 of a percent difference between 600 and 1200.

 

[imported_inspire]

I wasn't shooting for a performance increase - I just wasn't thinking... I'll change it before I power down for the night. I'm satisfied with this OC - maybe later I'll tweak with it (like when BIOS stuff gets fixed), but for now, I'm good. Thanks to all the guys here for the help and support - I definitely could not have done any of this without you guys.