This motherboard is driving me nuts. I suspect that I'm missing something really obvious, but I can't figure out what it is.
In a nutshell, I can overclock the heck out of the board and it runs Prime, BOINC clients, etc. perfectly (see below for some excpetions), but the stress test seems to be 3DMark06. Once I fire that up, even moderate overclocks crash on me. Here's the situation. If anyone can offer any insight, I'd appreciate it.
CPU : Q6600 SLACR (Stock 2.4GHz - 9*267)
MB: P5N32-E SLI
Cards: EVGA 8600GTS, SB X-Fi
RAM: Supertalent PC6400 1GBx4 T8UX2GC5
(settings detailed below)
From 1066-1200Mhz QDR FSB (266.5-300Mhz), I can increment by 1MHz and it runs like a champ at every speed. No hiccups at all
At 1200-1466.7 (300-366.65), I can only use the following settings. Anything else either doesn't bring up the BIOS or locks up at the CPU speed display.
1280 (320x9 = 2.8GHz)- Completely stable until running 3DMark
1353.8 (338.45 x 9 = 3.04GHz) - Very unstable
1440 - (360 x 9 = 3.24GHz) Moderately stable.
1466.6 - (366.65 x 9 = 3.3GHz) Completely stable until running 3DMark
**multiplier has no apparent affect on stability or ability to bring up the bios at other clocks in this range*** Tried 6-9 (board allows me to select 10+, but it never sticks since the chip is capped at 9)
At 1466.6 (366.65), I can start incrementing by 1MHz again. I've had it all the way up to 1680 (420 * 9 = 3.78Ghz) with decent stability until I run 3DMark or any other graphically intense program. I can beat the hell out of the board with any other kind of stress/benchmark tool and it rarely flinches. (except as noted). The sweet spot seems to be 3.3GHz. I haven't had a single hiccup at that speed until I kick in DirectX. I ran for 72 hours straight running 64Bit BOINC apps. I then fired up WoW and "blerf". Blue screen.
Dropping the LDT from 5x to 4x improves the stability in 3DMark and lets it run longer. Dropping it any lower makes it worse and causes Windows itself to crash. I can't see anything changing in NTune, SANDRA or PCProbe when I tweak this. So I'm assuming that it strictly a NB to SB setting that doesn't impact anything else.
Memory speed can be changed to anything. I've run all tests at 400, 667, 800, 900 and 950 and a number of oddball settings to try syncing the memory to the FSB. The memory isn't stable at 1GHz. But otherwise, there's not a huge difference at any of the speeds. (no more than ~10-15% performance difference. Probably because I'm letting the timings self adjust to more aggresive settings at lower clocks). There is no apparent impact to stability until I try to push the memory above 950Mhz. It doesn't require a voltage tweak until it gets above 900MHz.
BIOS Version 1205 (shipped with 902, but I couldn't run 4 sticks of RAM until I flashed to 1205. It would just lock up at the BIOS when I threw in the 4th stick)
Scythe Infinity with a single 120mm fan (29-60 degrees depending on the OC)
All Spread Spectrum disabled
All PCI clocks - 100MHz
SPP <-> MCP - 200MHz
Enhanced C1 - Disabled
Speedstep - Disabled
Virtualization - Enabled (need it for Xen. Doesn't seem to make a difference to stability)
Execute Disable Bit - Enabled (again, doesn't seem to impact stability and both my Linux Kernel and Windows support it.)
RAID Enabled - (No impact to stability whether I run single drive or mirrored.)
VCore - 1.25 - 1.45 (no tweaks required up to 1280 [320x9=2.8GHz]. Oddly, required increase in the 1200-1466.7 (300-366.65) range, but then ran decently at stock voltages at 1466.6 (366.65). Only a minor tweak wa required to make it rock solid.
VTT - 1.55 (no apparent downside to running it high and it seems to have the highest impact on stability. Just pegged it here and left it.)
1.2V HT - 1.2 - 1.35 (Only apparent impact is to increase instability when set above 1.35. Left at 1.25 for most tests)
SB Core - 1.5 (no apparent impact to stability)
NB Core - 1.35 (Mild overclock. Possible stability improvement. Higher voltages increased instability)
Memory - 1.8-2.1 (Only req'd at 900MHz+. Performance gains were minimal, so left at 800MHz and 1.8v for most tests.)
Coretemp reports the CPUs running from 29-65 degrees (no apparent correlation between CPU temp and stability. Runs rock solid at 65 running Primes and then crashes at 41 at a lower clock when I fire up 3DMark)
So, my questions are
1) What's a likely cause of this "forbidden block" of frequencies at 300-366? Has anyone else experienced this? I've seen postings about other boards that talk about "blackouts" around 300-350Mhz, but I haven't seen any explanations or theories.
2) I suspect that there's some PCI-E timing issue causing my instability, but I have no clue how to fix it. Any suggestions? The card and PCI-E speeds aren't changing. So it's not the card itself. But there isn't a whole lot to tweak on the PCI side in the BIOS.
3) Should I be considering a BIOS downgrade? Is it possible or likely that this is a 1205-specific issue? What's the best OC BIOS for the P5N32-E SLI right now?
4) Any value to trying a different set of video drivers? I suspect that it's a long shot and I'd prefer not to put any additional unneccessary variables into the mix. But I'll defer to the experts.
Any help would be appreciated.