Prime95 freezing

[Hogan773]

I've been stability testing some 44x OC voltages today after living well with a 42x OC at 1.22 using offset minus .05 for the past couple months.

I get thru LinX 15 passes on maximum (6000ish memory) just fine.

When I let Prime just go, within a few hours it must freeze up and reboot, because I'd go do something else or head out shopping or whatever, and when returning after at least 2 hours the computer would be on, Windows desktop idling there, but no Prime and no HWMonitor open as it was when I left.

So do I just need a tad more Vcore? Or could there be something else that is causing Prime to freeze after a couple hours that is not related to the CPU or VCore? My temps are fine.....sitting in the 53-55 area with an initial spike to about 59-60 while the Auto Fan kicks in at the first minute or starting Prime. Voltages around 1.26/1.27 under load. To be fair, I haven't run LinX for 2+ hours straight, just 15 passes which was like 1.5 hours. But I thought LinX was "way more stressful" than Prime and could ferret out problems more quickly, so when I'm passing LinX then I'm surprised that Prime is freezing up.

Anyone else have this happen? It's never an Error on Prime, just a freeze and automatic reboot.....

 

[Idontcare]

Are you running small FFT or large FFT in Prime95?

 

[jvroig]

Usually it's vcore when I get that in Prime95 (freeze or reboot; usually in small FFT, but just recently I've also experienced the same symptoms in IPL FFT when vcore needed to be increased by a notch).

When it's something else (like RAM or IMC) it just usually gives me rounding errors in IPL FFT or Blend.

 

[Hogan773]

Are you running small FFT or large FFT in Prime95?

I'm running the "blend" which is the default that pops up.

I guess its possible that all of my various OC attempts are really inherently unstable over many hours of chugging, but I just find it curious that they all pass on LinX. I guess I should try a pure default settings (stock multiplier and voltage) and see if Prime can run all night. Its just suspicious that Prime can run for a good hour or more but then all of the sudden something happens that makes it freeze or BSOD.

 

[Hogan773]

Usually it's vcore when I get that in Prime95 (freeze or reboot; usually in small FFT, but just recently I've also experienced the same symptoms in IPL FFT when vcore needed to be increased by a notch).

When it's something else (like RAM or IMC) it just usually gives me rounding errors in IPL FFT or Blend.

These rounding errors - those are reported by Prime I assume? I've never ever had an "error" in Prime - it just runs until it freezes/BSODs after a couple hours.

 

[jvroig]

These rounding errors - those are reported by Prime I assume? I've never ever had an "error" in Prime - it just runs until it freezes/BSODs after a couple hours.

Yes. It will notify you quite clearly. The thread icon will become red, and the last message will inform you of the error that happened (FATAL ERROR, etc), and the thread will then stop working, but the rest of the threads will continue until they also encounter an error of their own.

 

[BTRY B 529th FA BN]

Give your RAM a test using memtest hci

 

[Idontcare]

I'm running the "blend" which is the default that pops up.

I guess its possible that all of my various OC attempts are really inherently unstable over many hours of chugging, but I just find it curious that they all pass on LinX. I guess I should try a pure default settings (stock multiplier and voltage) and see if Prime can run all night. Its just suspicious that Prime can run for a good hour or more but then all of the sudden something happens that makes it freeze or BSOD.

Blend is not helpful when it comes to actually diagnosing errors, nor is it helpful for actually gaining confidence that your system is "stable".

I have long wished they'd remove that option.

Blend alternates between running smallFFT's and largeFFT's.

Small FFT's are needed to make your CPU hot, it doesn't involved the memory controller or the ram.

Large FFT's are needed to make your memory controller hot and really stress test the memory subsystem.

By alternating between small and large FFT's, Blend never get either set of components very hot and thus it never really stress tests the components as intended by the application in the first place.

What you need to do is run smallFFT and largeFFT separately.

If you get errors while running largeFFT then your ram (or memory controller) is too hot or not high enough Vdimm (or simply to high of clockspeed).

If you get errors while running smallFFT then your cpu (or L1$) is too hot or not high enough Vcc (or simply to high of clockspeed).

Running blend is not helpful because you have gained no info regarding the hardware you need to work on (you just know something is not stable) and it is not helpful because passing Blend doesn't mean anything since you haven't actually stress tested the system with smallFFT or largeFFT for hours on end.

 

[BTRY B 529th FA BN]

^ thank you, bookmarked to save myself the typing to explain this, hah

 

[Hogan773]

Blend is not helpful when it comes to actually diagnosing errors, nor is it helpful for actually gaining confidence that your system is "stable".

I have long wished they'd remove that option.

Blend alternates between running smallFFT's and largeFFT's.

Small FFT's are needed to make your CPU hot, it doesn't involved the memory controller or the ram.

Large FFT's are needed to make your memory controller hot and really stress test the memory subsystem.

By alternating between small and large FFT's, Blend never get either set of components very hot and thus it never really stress tests the components as intended by the application in the first place.

What you need to do is run smallFFT and largeFFT separately.

If you get errors while running largeFFT then your ram (or memory controller) is too hot or not high enough Vdimm (or simply to high of clockspeed).

If you get errors while running smallFFT then your cpu (or L1$) is too hot or not high enough Vcc (or simply to high of clockspeed).

Running blend is not helpful because you have gained no info regarding the hardware you need to work on (you just know something is not stable) and it is not helpful because passing Blend doesn't mean anything since you haven't actually stress tested the system with smallFFT or largeFFT for hours on end.

This is an extremely clear and helpful explanation. Thank you. Now I wonder why the makers of Prime95 can't update their interface to match your approach. Many incoming users don't know what the heck an FFT is anyway (I have read the definition but I still don't know what it means...I just know its math which is exercising the CPU). Should be two buttons that pop up for torture test 1) CPU test (small FFTs) 2) Memory/Chipset test (large FFTs) and then maybe the blend option but blend shouldn't be the default.

I ran Memtest 86+ last night and I passed all 8GB with no errors. Actually I went to bed while it was still running, assuming that it would stop after 1 pass, and when I woke up this morning it was still testing.....probably was on its 7th iteration of the full test. ooops. Sorry memory for giving you an all-night workout.

I saw somewhere else that sometimes the PSU can be responsible for the hanging in Prime. ie when the PSU is running hard consistently for hours, it can get hot and maybe it gives an erratic voltage that hangs the computer. I have a brand new Antec Earthwatts 380 which I think is a good manufacturer. I have no powerful GPU or anything so I don't think I'm at all close to maxing our 380 watts.....but maybe the PSU gets touchy after running full bore for hours.

 

[BTRY B 529th FA BN]

Now try Memtest HCI

I've found errors sooner with this than Memtest 86

 

[kevinsbane]

My home computer was an i5 750 running some corsair ram when I built it. I could run Intel Burn Tests, Memtest, and Prime95 small FFT's with no problems; but within seconds of running large FFT/Blend tests, it would freeze solid; hard lock and unresponsive. This would also happen on occasion during normal use: hard lock and unresponsive. I hadn't even overclocked it.

I tried many many different configurations, and couldn't figure it out. I ended up swapping my ram out for some other (GSkill) ram, and all my freezes went away. I would check out your ram; if you have some other ram lying around you can test, pop it in and see if it makes a difference.

 

[Diogenes2]

...Now I wonder why the makers of Prime95 can't update their interface to match your approach. ...

...Because the makers' of Prime95 main goal is is not stability testing of overclocked PCs...

 

[tweakboy]

All this blend talk, wants me to make myself a fruit juice with my blender.

Best is setup do your thing and run it and dont touch it again.

 

[Hogan773]

My home computer was an i5 750 running some corsair ram when I built it. I could run Intel Burn Tests, Memtest, and Prime95 small FFT's with no problems; but within seconds of running large FFT/Blend tests, it would freeze solid; hard lock and unresponsive. This would also happen on occasion during normal use: hard lock and unresponsive. I hadn't even overclocked it.

I tried many many different configurations, and couldn't figure it out. I ended up swapping my ram out for some other (GSkill) ram, and all my freezes went away. I would check out your ram; if you have some other ram lying around you can test, pop it in and see if it makes a difference.

I dunno. I have GSkill and it passed Memtest just fine. I haven't had the instant lock - more like after 2 hrs.

I'll try setting my RAM at 1333 (even though it is 1600 RAM) and see if that makes a diff. Or maybe a little more VCore.

 

[Hogan773]

Strange as I'm running Prime w/small FFTs tonight....for first 20 mins voltage is at 1.21 steady and temps around 55 C.

Then within like 1-2 minutes the voltage rises to 1.26 and temps up to 60 C. And its been sitting there since.

Why the heck would you get such a quick change in either Vcore or temps after Prime has been running for awhile?

EDIT - WHOOPS. Good old ASROCK Tuning utility. I had it open to check voltages and somehow when I went to the OC tab it must have applied 0 offset voltage vs the minus 5.5 I had in BIOS. I never hit "APPLY" but somehow it did it anyway. I just moved it back to minus 5.5 and hit "APPLY" and now its back to 1.21 and 55C.

Alright I need a drink now

 

[tweakboy]

Good, as long as it doesnt hover around 80's and high 70's then your ah right. thx,, gg

 

[LiuKangBakinPie]

Increase the voltage by 0.025 till it pass or you reached 80 degrees