Seeking advice from expert overclocker

Hi there,

I ran into a strange problem with my machine, which I don't understand.

I'm running a Q6600 on an Asus P5K Primium with 4x2 GB of G.E.I.L. DDR2/800 ram. OS is Vista64. The CPU is watercooled, the rest is on air.

Currently the setting is 8x400Mhz (1600 Mhz FSB and 3.2 Ghz CPU). Ram is 800 with [email protected] which is exactly the specified settings for this ram.

To test stability I used OOCT, Prime95 (64bit version) and Memtest86+ (booted from a CD).
"Round off checking" for Prime95 is enabled for all tests.

Test results:
OOCT for 2 hours - no errors, max core temp around 60°C.
Prime95 Small FFT for 2 hours - no errors, max core temp around 60°C
Prime95 Large FFT for 2 hours - no errors, max core temp slightly above 60°C
Memtest86+ in standard mode for 9 hours - no errors

Now comes the catch:
Prime95 in blend mode: rounding errors on multiple threads within the first five minutes of testing (core temp at that time slightly abouve 50°C)...

I'm slightly confused as in which direction to look now.
The CPU seems to be stable enough as proven by the Prime95 small and large FFT tests.
The memory seems to be stable as proven by 9 hours of Memtest.


But why do I get massive errors immediately when starting the blend test in Prime?

Any ideas in which area I should dig into?

Cheers,

Emacs

Memtest is not exactly good for testing ur memory for stability, but rather for whether it has errors or not...Blend tests a lot the RAM, and from my knowledge, it also stresses a lot the NorthBridge...Try giving a bit more V to ur RAM, (2V or 2.1V) and see whether u still have problems...Also, u might need to up ur NB voltage a little bit...

JCD37 said:

Memtest is not exactly good for testing ur memory for stability, but rather for whether it has errors or not...

Click to expand...

I was thinking into direction of memory as well. The prime95 small and large FFT tests are actually stable up to 3.6 Ghz on the CPU.

Is the built in Memory Diagnostic tool of Vista anything good?

I will try upping the NB voltages as well (have them still on auto)!

My approach would be to set the CPU on 6x400Mhz (2.4Ghz) to avoid any CPU problems.
From there I would start several tests with NB and DIMM voltages and use Prime95 blend test for measuring the results.
Once I should have a stable setting I'll up the multiplicator to 8 or if I feel adventurous to 9.

Does that sound reasonable?

Thanks,

Emacs

Emacs said:

I will try upping the NB voltages as well (have them still on auto)!

Click to expand...

There's your problem.

Is there something flawed with the 64bit version of Prime95 (25.6 build 6) ?

After I have spent the afternoon tweaking all kind of system settings, I switched back to the 32bit version of Prime which runs the blend test errorfree for two hours now.
Using the 64bit version, running the same blend test, I get rounding errors within the first minute.

I feel a bit silly, ramping through my BIOS and many forums...

Anyway, I got it now at 9x400Mhz (3.6 Ghz). Northbridge is pretty upvolted @ 1.55V, DIMMs at @2.0V (not the recommended 1.9V) and FSB Termination at 1.4 V. CPU Voltage damper enabled I get 1.39V on the CPU under load in CPU-Z.
Maximum core temp during two hours prime (32bit) was 66°C on two cores, and 61°C on the other two.

If I had never tried the 64bit version, I'd say, that is a rather stable system (yea, yea, just two hours of P95, I know. But what the heck).

Cheers,

Emacs