Originally Posted by colin_of_hunt
Well, d'oh! Installed BIOS F9, loaded optimized defaults and checked what voltage it suggested for DRAM. Wouldn't you guess it... 1.8V. Selected 2.1V (according to OCZ's website) manually and running memtest+ right now.

:)

Try 1.25V-1.3V northbridge too if you get trouble.

Got trouble already, in fact. Did some x264 encoding on XP and Vista. XP is fine, 3 hours of all cores at 100% without problems. Vista, however... five minutes of encoding and Vista froze up completely. Do I need to raise DRAM Termination voltage and Reference voltages as well to 50% of 2.1V?

OK, now I'm really in need of some good advice. The board appears stable, but only for a while. I'm scared of burning (i.e. wasting) Blu-rays on it since the rig has a nasty tendency of freezing up completely after being powered up for about 90 minutes. This is especially a problem with Vista, but it froze up once on the XP as well.

From BIOS F9's defaults I've raised the DRAM voltage from 1.8V to 2.1V as suggested by the memory manufacturer (OCZ), Northbridge (MCH Core) from 1.1V to 1.2V and MCH/DRAM Reference, DRAM Termination, Channel A Reference and Channel B Reference from 0.9V to 1.050V (i.e. half of DRAM Voltage).

The board is still not as stable as I would like. Richard, you recommended raising Northbridge voltage to 1.25V-1.3V; I'll try that next. Do I need to alter any of the other voltages?

Thanks for any advice.

Try 1.25-1.3 - it'll easily handle that. Check your temperatures because maybe something is overheating after a length of time.

Raised MCH Core to 1.26V and have now been running Burn In 2008 in CPU Stress Test mode on XP for 40 minutes. According to SpeedFan 4.35, Core 0 is at 51C and the rest are running somewhat cooler, from 44 to 46C. Core 0 temp hasn't budged in the last 30 minutes. Any suggestion on what app I should run to do a proper memory<>CPU stress test?

Put your finger on the MOSFETs and northbridge :)

Prime95 torture test and loop 3dmark06 over the top is what we do.

Hmmh. Prime95 starts only 2 of 4 worker threads and 3dmark06 gets stuck in the system configuration scanning stage maxing out one core. Man, I'm really feeling the love from this PC.

On Vista64, running Prime95 (64-bit), three of the four worker threads crap out within seconds of test start. "FATAL ERROR: Rounding was 0.5, expected less than 0.4. Hardware failure detected. Work thread stopped." Sigh...

You've got a front side bus problem - try increasing your VTT/CPU PLL slightly or losening the memory timings.

Thanks Richard, I'll try that.

Well, that didn't help. Noticed Gigabyte has released a new BIOS for the board (it's a beta, apparently) so I figured what the hell, I'll give it a shot. The only thing I did after loading optimized defaults was to change DRAM voltage from 1.8 to 2.1V. Prime95 64-bit has now been running on all four cores for a minute and all worker threads are still at it. I'm carefully optimistic.

Now running Test 8 of 4000 iterations each. All threads still running, core temps at 51-58C.

1024K and 8K self-tests both passed OK. Core temps are at 51-56C after 31 minutes of Prime95. Memory timings, 5-5-5-15, were selected by BIOS.