Odd OC situation, help/advice needed

[shiranai]

I first posted this on the AT forums here:

http://forums.anandtech.com/messageview.as...6&enterthread=y

 

But didn't receive many responses, so I'm trying here. The following is copied verbatim:

 

Relevant system info:

Venice 3200+ 10x270 @1.4v*110%

OCZ Value VX 5:6 divider (225Mhz), 2-2-2-6-1T @3.0v

DFI nF4 Ultra-D 510-2f BIOS

 

Problem:

The machine passes memtest (20+ hours), prime95 (60+ hours), superpi, etc. However, she will sometimes reboot and/or BSOD when loading windows (BSOD has errors of IRQL_NOT_LESS_THAN_OR_EQUAL, and PAGE_FAULT_IN_UNPAGED_(something), or no error at all in some cases). However, there is no consistency in the rebooting or errors - sometimes I will be able to boot and run stress tests, other times it will restart repeatedly once reaching the windows login screen, and sometimes it will BSOD once I log in, with different errors every time (although the IRQL errors shows up most often).

 

I have tried reformatting and reinstalling windows, thinking I had a corrupted installation, but that did not fix the issue. I then thought it might be temps, but the CPU never goes above 48C and the case ambient temp not above 31C even after almost 3 days of 100% load. And again, I'm able to run days of stress testing without ever rebooting, with no errors, lock-ups, or anything.

 

So basically, sometimes I'm unable to boot into windows, but the times when I am able to boot windows, I can run whatever I want, as long as I want, without any issues. Dropping the divider to 3:4 (202Mhz) resolves the issue, as does dropping the command rate per clock to 2T, but I really want to keep settings at the current level if at all possible, because I can pass all of the stress testing with no problems.

 

Any idea what could be causing this?

 

All else fails, I'm thinking of flashing to the 510-2 pre-fix version BIOS and seeing if it'll help.

 

---

 

Damn, knew I forgot something

 

OCZ Powerstream 520W. Amps are spec'd at: +3.3V(28A), +5V(40A), +12V(33A) according to their website here: http://www.ocztechnology.com/products/p......am_power_supply

 

I also adjusted the rails to within +0.02V of the listed voltage.

 

And in case it matters at all, I have the Vdimm jumper on the mobo set to the 3.3V rail (as opposed to the 5V rail).

 

Also, the memory is in the yellow slots (dimm1/3), since the guys over at dfi-street claim that those slots are better for high voltage memory like UTT/BH-5 (which VX is supposed to be, yes?)

 

---

 

So, any ideas? Thanks in advance for any help.

[Guest jot]

Hey those are def utt moduels and u def have a mem prob. Are u runn'n off the 3.3 rail or the 5 to get that 3.0 vdimm? I think i would use the 623-2 bios been use'n the 623's since they came out without a prob.

[chieferich]

that irq not equal message points to memory. i don't think your going overboard with the oc, so try out some different settings/timings. run memtest 5 (loop) one night, then #8 another night. finally, loop 3dmark03 for a night. if you get a problem from this, you're not stable. hit the drawing board again.

 

*go to 3.1v. use 4708.

[shiranai]

I originally thought memory problem too, but as I said I've run memtest several times for 20+ hours, all tests, with absolutely no errors. Similarly, cleared SuperPI 32M many times, and Prime95 with various FFT size ranges (small, large, blend) has run for 60+ hours without errors. I've also looped 3dmark benchmarks without issue.

 

Also, loosening the timings to SPD (2.5-4-4-8) didn't change the situation at all.

 

I'll try fiddling with the settings. I'm somewhat leery of higher voltage to the RAM, since I don't have active cooling directly on the sticks (although the XP-90 does hang over one of them). I'll zap it with 3.1v and seeing what happens.

 

I'm actually currently trying FSB at 265 (RAM at 5:6 = 220 Mhz). If it works out at that speed I think I'll just take it. Some guys over at AT recommend taking the CPU down about 5% from the max 'stable' overclock just to be safe. My CPU stabilizes around 10x280 standalone, so 265 would be about right according to that rule.

 

I still wonder what kind of memory error could cause random boot errors, but still allow me to run all of my stress tests for days without error ...

 

Thanks again for the help. I'll see what kind of results I get in the morning. And if anyone else has ideas, lemme know?

[Fire-N-Ice]

Use the search function! This question is asked like every other day! Most of the posts I've seen regarding this, people say that you need to up the voltages a little. Don't know for sure, as I've never had this problem.

[shiranai]

I'm familiar with the search function, thanks. I've been lurking around here for the past month or so reading up on OCing and the various BIOS options, and I haven't seen anything that addresses this particular problem. If you think otherwise, mind providing some linkage?

 

I wouldn't be asking this question if I couldn't pass stress testing, because then the problem would be obvious. My issue right now is that all stress testing passes, for several days at a time, but the computer will randomly have boot errors anyway.

 

Oh, I also tried increasing the voltage and the refresh timing to 4708 as suggested (I tried up to 3.2v), as well as loosening all of the other timings, and there was no change. Still get boot errors. Memtest came up with nothing when I checked this morning; left it running and will check when I get off from work. Given past experience though, it's not going to return any errors regardless of how long I leave it on. I've been running various tests literally 24/7 for almost two weeks now with absolutely no errors of any kind whatsoever. But sometimes I still can't boot.

 

Will try flashing to plain 510-2 or 623-2 tonight and see if that fixes anything.

[CJ4]

I had some boot problems that the 618 bios fixed. Haven't tried the 623 ones yet.

 

CJ

[Fire-N-Ice]

Angry Games says

"IRQL less than or equal blue screen error, and then immediately rebooted.

this is almost always a memory error.

I've had memtest pass with flying colors and 3dmark pass easily but the instant i ran a 'real' game like UT, Chronicles of Riddick, Doom3, HL2, etc, it would crash instantly.

overclock was the cause of it one time

memory timings not being correct was another cause another time

incompatible memory was another cause....

it always went back to the memory being incorrect/incompatible"

 

From what I've read, up the voltage on the ram a little (you're using the vx so goto 3.2volts-not on the 5v rail), then loosen up your ram timings. You're at 2-2-2-6, try 2-3-2-6, or 2-2-2-7. Change each one, one at a time and try. If you get an error again, put that one back and try another. If you try loosening each one and nothing, then try loosening 2 of them up, like 2-3-3-6, or just try 2.5-4-4-10, and see what happens. If this last one works then you can work your way down, till you hit the errors again.

[shiranai]

Upped to 3.2v and 2.5-3-3-7 (10x270 with 5:6 divider) and I no longer get any reboot errors. Will try running 'real' games (I'm embarassed to say I can't remember where I stashed all of my games ...) and seeing what happens. If it works out, I'll post results.

 

Thanks for the help.

[shiranai]

Hmm, quick question. I'd always assumed it was an unspoken rule that tRCD and tRP had to be higher than tCL, but I'm having no problems running 2.5-2-2-7. Is this unusual/impossible?

[Guest jot]

One more quick idea if u got a meter check your psu. Check 3.3 5 and 12 volt and make sure u are not gett'n any kinda funky spikes. Then check your mobo voltage reg and make sure u don't have any leak cause'n the voltage to ramp up and throttle back. I never trust what the bios tells me about voltage (even though DFI does a wonderful job) thats why man made (fluke meters)

[Fire-N-Ice]

I'm not sure about the "unspoken rule" you speak of. As long as it works for you it probably don't matter. Try pm'ing one of the ocz guys here and ask them.