i7 920 overclocked to 3.8 GHz - Prime Stable -- Random Restarting?

[ItsJono]

I have an i7 920 overclocked to 3.8 GHz. I get some random restarts which I don't like at all but I am prime stable? I'm rather confused...

CPU - Intel Core i7 920 D0 @ 3.8 GHz - 181x 21 (Turbo ON, HT ON) running at 1.25xx v (high end of 1.25xx volts because I figured it was a lack of voltage)

MOBO - ASUS P6T Deluxe V2 (Used to idle at 50c! Now at a comfy 35c thanks to the GentleTyphoon side intake)

MEM - OCZ Gold DDR3 3x2GB 1600 (155x~ mhz) @ 1.66v

GPU: 4870 X2 (light OC)

PSU: HX750W Corsair Power Supply

Sorry I don't have exact features, i'm watching Seinfeld but I really want these random restarts gone!

-

CPU Cooling - Noctua NH-U12P SE 1366
Case: Antec 300

2x 1850 RPM GentleTyphoon Intake
1x 1850 RPM GentleTyphoon Exhaust
1x 1850 RPM GentleTyphoon Side Intake
1x Stock Antec 3 speed exhaust fan on medium (top exhaust vent)

Please help!

 

[MagickMan]

Prime isn't a very good test, get Linx and OCCT.

 

[lopri]

First thing to determine in situation like this is whether it occurs when the system is not overclocked. Chances are it doesn't. Then you may want to see at what point it starts happening by gradually raising the bus frequency.

 

[Idontcare]

but I am prime stable? I'm rather confused...

Are you small FFT or large FFT prime stable?

If you tested with "blend" then the results don't really mean much.

Small FFT tests the cpu, but you could still be unstable wrt ram or the QPI/memcontroller/L3$ (the uncore on your i7 920).

Large FFT tests your memory, the QPI, and your L3$.

You can't claim prime stable unless you run both the small FFT and large FFT tests and passed both. Recommended test time is 3-6hrs for each FFT type.

 

[jvroig]

Small FFT tests the cpu, but you could still be unstable wrt ram or the QPI/memcontroller/L3$ (the uncore on your i7 920).

I echo this.

When I overclocked a Phenom II X4 to 3.7 core and 2.6 CPU-NB, it would pass hours of small FFT. Doing in-place large FFT test, however, results in rounding-off errors just after a couple of minutes. I needed to increase the CPU-NB voltage to get it stable.

 

[Makaveli]

What about folding @home for stability tester?

 

[StrangerGuy]

What about folding @home for stability tester?

That is just as stressful as Prime.

Might as well use IntelBurnTest maximum load, 10-pass. If you cannot pass this don't even bother using Prime.

 

[Makaveli]

Good to know i've been priming about 5 days straight at 100% load for the chimp challenge 2010 and machine has been rock steady!

 

[SHAQ]

20 pass LinX and 5 hour Prime is my minimum testing. How long did you test Prime? You can try upping CPU voltage one notch and/or QPI voltage one notch.

 

[Suspicious-Teach8788]

20 pass Linx may not be enough. I failed 3 hrs in the other day. Linx 0.6.4 has some known issues where your CPU doesn't work as hard as it can if there were some problems and maybe you readjusted freqs/voltages in Windows. I tried yesterday after a long set of benches and Linx was giving me only 36 GFlops. Not only that temps were a good 6-7 deg lower than they were in the morning. I fired up OCCT and it errored in about 1 hr in the Linpack test. I think for now I'll stick to OCCT.

Linx works fine on a reboot though. I'll use both and see!

 

[BTRY B 529th FA BN]

One of your component parts is degrading. I had a DDR2 kit that did 500fsb all day long on a Gigabyte UD3R P45 board. As I tested the kit before I sold it, it was only good for 400fsb at 1.8v

It may possibly be your RAM kit. Test it with Memtest HCI Link

 

[ItsJono]

One of your component parts is degrading. I had a DDR2 kit that did 500fsb all day long on a Gigabyte UD3R P45 board. As I tested the kit before I sold it, it was only good for 400fsb at 1.8v

It may possibly be your RAM kit. Test it with Memtest HCI Link

I have a feeling you're correct with RAM but if my idea of testing OC's is dated for these intel CPU's (first time Intel owner, long time AMD fan) then there might be a better testing methodology for these with a better tool... or is MemtestX86 still relevent? haha

 

[BTRY B 529th FA BN]

Memtest HCI > Memtest86. Vista and Win7 also offer a memory diagnostic tester.

 

[Idontcare]

Memtest HCI > Memtest86. Vista and Win7 also offer a memory diagnostic tester.

what about memtest+ v4.10? works great for me but what does memtest HCI do above and good ole memtest+? (is it faster to cause errors, please say yes, memtest on 8GB of DDR2 takes fooooorrrrreeeevvvvvveeeerrrrrr).

 

[BTRY B 529th FA BN]

In my experience, I could run many passes with Memtest86 and not get errors. People kept telling me Memtest HCI was better. I didn't believe it. I tried it and it's a pretty solid tester.

I can't give you a yes or no if it's faster but I do recommend it.

 

[Suspicious-Teach8788]

Yeah. I agree about HCI. I easily got 7-8 hrs of stability on Memtest 86 and I was convinced I should go higher (this is in the midst of overclocking, not trying for a 24/7 stability yet... was seeing how high I could push my mem at 8-8-8), and while I was confident, HCI errored out on me in a mere 5 MINUTES!!!

Yeah. That stuff works.

 

[caboob]

random restarts in my experience indicate psu going bad or not up to the task. if you were getting bsods id say memory. 750 should be enough but you've got a power hungry quad and an x2 gpu.

 

[Meghan54]

random restarts in my experience indicate psu going bad or not up to the task. if you were getting bsods id say memory. 750 should be enough but you've got a power hungry quad and an x2 gpu.

I'd say it's anything but the power supply. And as an i7 920 OC'd to 3.33GHz along with a 4870X2 barely gets over 450W of draw under Furmark, his system certainly isn't stressing his Corsair unit at all, even with a mild OC of his video card.

I'd tend to much more lean towards a bad stick/IC of memory, esp. given OCZ's downward trend in reliability of their memory products of late.

 

[biostud]

turn off turbo, and 2nd on the blend prime95 test, I got all my workers to crasg within 10 minutes running the blend test, due to memory problems.

 

[Mr. Pedantic]

That is just as stressful as Prime.

Might as well use IntelBurnTest maximum load, 10-pass. If you cannot pass this don't even bother using Prime.

More. Especially -bigadv. I've had my system overclocked so that it's 24 hours prime stable but the SMP client stopped the WU after 20 seconds because it was deemed unstable. LinX and IBT are really the best, though. Run them for about 6 hours each and if you pass, you're golden.

 

[GLeeM]

I'd tend to much more lean towards a bad stick/IC of memory

I have the same system (mobo, CPU, PSU, maybe RAM).
The only spec I see that might be off is 1.66v on the mem. Is your OCZ of the 1.5 or 1.65 variety? I think max recommended for the system is 1.64? I have mine at 1.60.

Could having too much v on ram be causing this problem?

Seeing as how my rig is 24/7 Folding @ Home, with talk of degrading, I am wondering if I should back off my 3.99 @ 1.264v (temp low 70s) OC?

 

[Idontcare]

In my experience, I could run many passes with Memtest86 and not get errors. People kept telling me Memtest HCI was better. I didn't believe it. I tried it and it's a pretty solid tester.

I can't give you a yes or no if it's faster but I do recommend it.

Ah, testing with HCI now...one thing I already don't like is that because it runs in windows (or linux) it can only test the unused ram. If your bad bits happen to lie in a memory region that is currently allocated to the OS then HCI won't catch it. Memtest+ tests all the ram.

That said, for the ram addresses that are testable by HCI the ability to run multiple HCI instances in parallel certainly speeds up the process considerably.

Currently testing ~6.8GB (1.2GB I can't test because OS has it locked up) via four instances of HCI.

Not gonna throw out my memtest+ bootable cd just yet...

 

[SHAQ]

20 pass Linx may not be enough. I failed 3 hrs in the other day. Linx 0.6.4 has some known issues where your CPU doesn't work as hard as it can if there were some problems and maybe you readjusted freqs/voltages in Windows. I tried yesterday after a long set of benches and Linx was giving me only 36 GFlops. Not only that temps were a good 6-7 deg lower than they were in the morning. I fired up OCCT and it errored in about 1 hr in the Linpack test. I think for now I'll stick to OCCT.

Linx works fine on a reboot though. I'll use both and see!

Interesting I had that at times and thought my memory OC wasn't stable although reboot didn't solve it half the time. I still think it was a weak memory controller. I get about 55Gflops consistently lately. I do 20 pass LinX, or actually 10 pass LinX initially. My final test is 40-50 passes which takes it about 2 1/2 hours and another 8 hours of Prime Blend. That has been enough for it not to crash in games/programs. Any more than that is academic because if it passes 3 hours how do you know it is stable at 4,5,6,10,100,1000,10000 hours? You gotta draw the line somewhere.

 

[ItsJono]

Update -

2x 1HR OCCT Tests - No Errors

Will proceed to test memory soon.

Interesting I had that at times and thought my memory OC wasn't stable although reboot didn't solve it half the time. I still think it was a weak memory controller. I get about 55Gflops consistently lately. I do 20 pass LinX, or actually 10 pass LinX initially. My final test is 40-50 passes which takes it about 2 1/2 hours and another 8 hours of Prime Blend. That has been enough for it not to crash in games/programs. Any more than that is academic because if it passes 3 hours how do you know it is stable at 4,5,6,10,100,1000,10000 hours? You gotta draw the line somewhere.

I have the OCZ Gold DDR3 sticks. They, just like a lot of other memory modules are rated at 1.65v. I went to 1.66v as per the recommendation of some overclockers.

 

[Suspicious-Teach8788]

Interesting I had that at times and thought my memory OC wasn't stable although reboot didn't solve it half the time. I still think it was a weak memory controller. I get about 55Gflops consistently lately. I do 20 pass LinX, or actually 10 pass LinX initially. My final test is 40-50 passes which takes it about 2 1/2 hours and another 8 hours of Prime Blend. That has been enough for it not to crash in games/programs. Any more than that is academic because if it passes 3 hours how do you know it is stable at 4,5,6,10,100,1000,10000 hours? You gotta draw the line somewhere.

Certainly you gotta draw the line somewhere, and 24 hours Prime used to be the standard right? I'd probably do at least 12 hrs of Linpack tests or so before calling it quits.

I've seen OCCT Linpack fail on me at 2 hrs in or 3 hours in even. I can claim 3 hour stability, and then what? My chances of failing at 3:01 are plenty higher than 24:01 (if I claim 24 hour stability). It's simple statistics I guess, and I know plenty of friends who run around claiming OCCT passed 1 hr or Linx passed, and in their rush to get their systems up and running we played plenty of DoTA games where they just kept crashing out.

I'd thoroughly test things to make sure it's completely clean. I Memtested 24 hours stable with my OCZ VX years ago, and I Primed 36 hours stable also. But somehow my OS would corrupt every few months. I swapped out the HD but that wasn't it.

I later dumped the RAM and switched away from BH-5 and high voltages and voila, problem solved. I was always puzzled, but there's plenty of ways you could still be "unstable," but I'd rather have my system very thoroughly tested before I decide to do everything with it. 3 hour stability would keep me wondering.

But for my most recent overclocks I determined 36 hour prime stability and then I tacked on a 0.02 volt bump and called it a day. Been fine for years now.

I'll see how my i7 holds up under my strategies. LOL.