Suggestions for fine tuning a Q6600 @ 3.2G w/ EVGA 750i SLi FTW

[susofseattle]

Soo- I'm getting 12hrs small FFT stable @ VCC 1.275v & 3.5hrs stable large FFT's in prime95 with the following:

VCC: 1.275 @ 53C
FSB: 1.3 @ 33C (wouldn't post at anything lower including my best attempts at fine tuning the GTLREF lanes from +60mV to +110mV in all 4)
DRAM: 2.0 @ {
tCAS 4
tRCD 4
tRP 4
tAS 12
CMD 1T
-
tRRD 3
tRC 22
tWR 5
tWTR 9
tREF 7.8uS
(5 passes, no errors in MemTest86+ 2.11) }

SPP: 1.55 @ 55C (again, won't run prime95 lower)
NF200: 1.3 (raising this from 1.20v seems to have fixed the errors on core 0 within 1 min, in prime95)
GTLREF: all @ +80mV

I know I'm getting Large FFT errors @ VCC 1.275 and a friend recommended 1.375, but doesn't 12hrs of small FFT's mean the CPU core voltage is stable and secure? I was thinking that either the NF200 voltage was too low or the GTLREF's are out of spec, to cause the fail on the large FFTs..

Appreciate the help and I will change and test the specs as you think best. I just think the lowered vcc means a huge drop in heat?

 

[susofseattle]

Prime95 froze the PC @ 3.5hrs. Adjusted to +85mV all lanes...

 

[susofseattle]

Again frozen at 3.5hrs. adjusting to +90.. kinda weird it freezes on schedule..

 

[susofseattle]

At +95mV it froze at 7.5hrs on Large FFT's. Rerunning at 320K FFT's to adjust the lanes.. Still looking for helpful suggestions.

Is 4-4-4-12-1T @ 712FSB 1:1 too tight or should I consider just doing unlinked, unsynced 1066 FSB on the DRAM?

 

[techgeek]

I would think that small FFT's would indicate a stable CPU. What sort of problem are you having with Prime 95? Re-boots, rounding errors, lock-ups, BSODs. I know that you checked your RAM with those timings in Memtest, but if you are stable with small FFT's, but get problems with large FFT's, it really sounds like a memory issue because large FFT's put a lot of stress on RAM and cache and occasionally the storage subsystem if it starts swapping. How many DIMMS do you have? If you have the option in your BIOS to force 2T, I would try that. Also it couldn't hurt to loosen your timings up a little. The easiest way to troubleshoot these things is from a completely stable base rather than an unstable one. Meaning you need to find settings that are consistently stable with large FFT's and then adjust one thing at a time until it becomes unstable.

Also I see that you are running your RAM at 2.0V. You neglected to tell us what motherboard and what RAM you are using . This could help if someone has a similar setup.

For now, my money is on it being border line RAM settings. I could be wrong.

Lastly if you are using a good cooler with your Q6600, 1.35V shouldn't be a problem as far as heat goes. Nominal VID for the Q6600 is as high as 1.3V. So in the grand scheme of things 50mV isn't a whole lot. Don't get me wrong, I am not suggesting that your problem is Vcore related. I am just assuring you that with a decent cooler, 1.35V is long term doable. If you look at my signature, I am running my Q6600 overclocked at 3.2GHz and it was dead easy. I am running Vcore @ 1.325V. We have different motherboards, but 3.2GHz is the really just an average overclock for the Q6600. I use a TruE and it keeps my cores in the high 40's to low 50's at full load.

The last thing I notice here is you aren't running your RAM in sync with your bus speed. What memory divider are you using that gives you 3.2GHz core clock and an FSB of 1424. I used a straight multi of 8 with a clockspeed of 400. This gave me an even 3.2GHz while running my RAM at it's full rated speed and keeping the RAM and FSB in sync (no dividers).

 

[susofseattle]

Getting lockups only, no errors, BSODs and only 1 reboot in 7 tests. hardware as follows:

EVGA 750i SLi FTW (most recent bios)
3x G.Skill 1G DDR2-800 (#4 is in RMA) underclocked @ 712Mhtz 1:1 on 1T
FSB: 1424 (356x9 multi.)

As far as cooling goes, I have lapped both the CPU and heatsink using an Arctic Cooling Freezer 7 Pro w/ Arctic Silver #7 & a 11"x11" box fan taped to the open side, forcing air thru the hardware, out the PSU fan hole, unused PCI slots and open DVD bay. Its ugly and loud, but for $15 dropped my CPU temps by 15-18C. ^.^

PC is currently on hour #7 of prime Blend tests w/ GTLREF lanes at +100mV

 

[techgeek]

Which G. Skill kits do you have (I am assuming you have two identical dual channel kits)? Is i this one : F2-6400CL4D-2GBPK? It seems to match your settings that you are using (4-4-4-12 @ 2.0V). One thing about running more than two DIMMS, is command rate. Usually with more than two DIMMS it's advisable to run the command rate at 2T. Your RAM will take a slight performance hit, but it's worth it to gain stability. This overclock (the settings you are using) seem overly complicated for such a conservative overclock. That said I am not familiar with nVidia chipsets. My overclock was dead easy with a P45 chipset. There were a few voltages I specified manually besides Vcore simply because ASUS boards are known to really crank the voltage on certain things when left set to auto. I never had to add voltage to anything other than the CPU and that was only .25V off of the default.

Until I know for sure what G. Skill kits you have, I can't say for sure whether loosening your timings will make a difference. Although it couldn't hurt either way. Again I would focus on the command rate first.

What sort of load temperatures are you seeing and what software are you using to read them?