Prime/Orthos fails after 8hrs (UD3P and E5200)

[francisA]

So I've had my UD3P with E5200 for almost a week now and been trying to push fsb further (not winning, only maxing out at 350fsb) so I settled for 333 x 11 (3.663GHz). But now. I'm just looking for the lowest vcore I can get away with.

Here's the setup:

Gigabyte GA-EP45-UD3P Rev 1.0 F6
E5200 (333x11) (vcore: BIOS set to 1.3375, CPU-Z shows 1.312)
2x2GB G.Skill PI Black 4-4-4-12 (running at DDR800 5-5-5-15)
Windows XP SP3

CPU PLL: 1.5v
MCH Core: 1.22v
CPU Termination: 1.3v
DRAM: 1.9v (may be able pull it off at 1.8v)

Tests:
Orthos (Blend) - fails after 8hrs
IntelBurnTest - 20 passes (PASSED)
Prime95 (Blend) - fails after 8hrs
Prime95 (Small FFTs) - 12hrs and going (no failures)
Memtest+ - Test #5, 30 passes (PASSED)
OCCT - 2hrs (PASSED)

Now, the failure on Orthos and Prime is not the rounding error (0.5 expecting 0.4 or something like that). In my case XP shows an error popup stating that "Prime95 or Orthos encountered an error and needs to close".

That error prompted me it might be memory-related and not oc-related thus my testing on Small FFTs and Memtest (#5) which incidentally, reported PASS.

Does this mean that XP is having trouble working with the 4GB of RAM that's in there, even though Memtest's Test #5 tested the 2-4GB block and reported no errors? Or is it that Prime95/Orthos cannot address anything over 2GB?

XP reports 3.325GB of RAM which I expected but was hoping it won't have any issues using what it think is there.

Is my analysis sound? Or should I run Memtest+ much longer that 30 passes?

Thanks in advance.

EDIT: As for next test, I'm thinking of pulling one of the 2GB modules and see if that'll pass. Whatdya think?

 

[VirtualLarry]

I think that your CPU vcore sounds low, actually. I needed 1.4v actual (CPU-Z, load), to reach 3.75GHz. 3.66Ghz is pretty close to that.

 

[francisA]

Thanks. Maybe it really needs more juice ........ sigh More juice coming up

 

[aigomorla]

wait it makes no sense how it will pass Linpack, but it wont pass OCCT?

Oh wait.... what max cpu temps were you seeing?

OCCT has a function on it where default, set to 70C. if your occt temp probes go above it, sometimes it will stop, or in your case, it could crash.

 

[francisA]

Exactly the point of my post aigomorla. It just doesn't make any sense. {Sorry, forgot to post about OCCT - updated the post above now}
OCCT also passed 2hr test.

Only Prime95 and Orthos are failing. I'm perplexed, to say the least.

Max temps I'm seeing are 60C when running IntelBurnTest and 54-55C when doing Prime/Orthos/OCCT.

I've gamed on it for 2hrs (Crysis) then asked my son to quickly play again for an additional hr (COD5) and it didn't crash nor hang.

As for OCCT, I didn't mess with any of it's settings, just changing the sensors to monitor. I'll look into that but then again, OCCT actually passed.

 

[allies]

The UD3P should have no problems hitting an FSB of 400mhz since it's RATED to that speed and after that speed, it becomes overclocking the FSB.

Try dropping your multi to 9.5 and running at a FSB of 400mhz... make sure the correct memory divider is set.

 

[JAG87]

E5200 in most circumstances wont reach 400 fsb.

OP honestly... fuck it. If you are 2h OCCT stable and 20 passes of IBT stable, you are fine. Don't waste any more time.

If, and only if that 8h crash means anything, I'd be surprised if your pc crashed once a year. I think anybody can live with that, unless you need 100% reliability, in which case your dumb for overclocking.

 

[aigomorla]

yeah like my pal jag says.

dude if you passed linpack 20 passes, and Orthos and prime is failing... That is uber bazar.

Dont worry, id go with linpack.

Did you do max stress test? or did u set a custom memory?

My next guess is your south bridge or northbridge is wacked, and yeah i can see why your looking at your ram.

 

[francisA]

Well, that sums it up pretty well, JAG87 Honestly, I was just perplexed as to "why". I've been playing Crysis, COD4, COD5 on it for days without any issues. It's just force of habit that if something doesn't add up, I can't help but try and figure it out. And also, maybe make a good reference for someone in the future just in case they encounter the same issues.

aigmorla:
I did max stress with IBT and yeah, that makes me confident that the system is working fine. With just that little bizarre thingy.

allies:
400fsb is a no go with my E5200. It hits a wall around 360fsb but with a 47% increase in Freq and 75% increase in FSB @ 350, I'm not complaining. 400fsb will be the job for an E8400, whenever I get around to getting one.


Thanks all for the input.


CPU-Z Screenie
IBT Screenie

 

[allies]

Originally posted by: JAG87
E5200 in most circumstances wont reach 400 fsb.

OP honestly... fuck it. If you are 2h OCCT stable and 20 passes of IBT stable, you are fine. Don't waste any more time.

If, and only if that 8h crash means anything, I'd be surprised if your pc crashed once a year. I think anybody can live with that, unless you need 100% reliability, in which case your dumb for overclocking.

Does the chip itself play a limiting factor to FSB? I was under the impression that the board is the deciding factor for FSB...

 

[mhouck]

When I was working on getting my q6700 stable at 3.6 and working voltages back down I would get the same error with Prime 95 but I'm running vista x64 bit. I used OCCT and memtest too without error.

Anyhow I stepped the voltage back up one step and everything ran fine. I think the cpu vtt setting is the one that I needed to step back up to eliminate the problem.

 

[Fox5]

If something fails, ever, it means your system is not stable. 100% stability may not matter to you, but I'd rather be sure of stability.
That said, I probably would have ended the test well before 8 hours (unless I had nothing better to do), called it stable enough, and just gone with the overclock until I encountered problems.

 

[VirtualLarry]

Well, if it's not stable, it's not stable. On my friend's E5200, it seemed like it would do 3.8, with everything else, but it just wouldn't pass Prime95 25.6. So I had to clock it back down to 3.75Ghz.

He's only running at 300FSB, btw, to get that speed, using the max 12.5x multi.

 

[ACSanchez]

francisA - did you ever reach any conclusions? I have almost the exact same setup and problem!

UD3P, E5200, I am at 3.75 GHz (12.5 X 300).

10 passes of Intel Burn Test pass 100% at max of about 70-71C, so I'm quite happy. But Prime Blend has failed the last two times I've run it. It's running right now at about the 90 minute mark (just bumped my VCore to 1.30V).

VCore 1.30V
DRAM 1.9V
MCH Core 1.16V

What do I increase next if it fails again? Try 2.0V on the RAM?

 

[Idontcare]

Originally posted by: ACSanchez
But Prime Blend has failed the last two times I've run it.

Failing the blend test unfortunately doesn't really give you the info needed to determine the likely culprit of the instability.

What you need to do is run small FFT and Large FFT separately and determine which one fails.

If small FFT then your likely culprit is CPU/Vcore, if large FFT then it is likely to be memory/NB/FSB related.

Blend is really a crappy "stress test" because it alternates between stressing the CPU versus the memory/FSB and never really stresses either component for too long of a duration to really uncover those errors that only reveal themselves under intense 8-12hrs of continuous small or large FFT processing.

 

[n7]

Idontcare is bang on, of course.

Small FFTs tests the CPU, & does it well.

Large tests the motherboard, & if FSB or NB is unstable, it'll fail.

IBT/LinX stresses the CPU & RAM, but misses things that P95 Large finds in terms of board stability.

HCI Memtest is by far the hardest RAM & somewhat NB test (particularly if you have a lot of RAM); it's more sensitive that Memtest86+ is.

 

[fpsdean]

I'm just seeing this now, but I have a super weird issue that's related.

I pass IntelBurn Test. I normally find that if it's going to find a problem, it will do it in the first 5 passes, but I gave it 20 just to be sure.....

Prime95: Small FFT & Blend pass no problem.

Orthos Small FFT? Fails inbetween test 1 and 2. Tried increasing CPU voltage a bit, and it's a no go.

Blend in Orthos passes.

OCCT passes.

Normally I find if anything is going to detect an error, IntelBurn test will get it first, then OCCT and finally Prime95. Why is it the least stressful fails and the most stressful tests don't?

In this case, I'm going to chalk this up to some sort of bug in Orthos...

For the record, I've been running these settings since October and have never seen a crash.