What's your Prefered system stability test?

[devilchrist]

just wondering what people use.

I try to test all three for at least 8 hours to confirm that the system is stable.

I see a big difference in stable small fft voltage settings to large fft voltage settings.

I read the sticky post on how 5 passes on memtest86+ test #5 and 24 hours on blend would make a system be considered stable for some.

but for me what's stable in those tests are not stable in large fft test.

Please post what's your flavor of test or tests and why you use them.

 

[AVP]

Custom, 1gb ram usage, 20 minutes, and then I go have fun with my computer. If issues arrive I change it. Back to fun.

 

[bailw]

I use OCCT 30 mins or 1 Hr test.

 

[Thor86]

24 hours+ LargeFFT for max heat/cpu load. Then 48+ hours Blend Custom using all available physical RAM. Never failed me ever. Memtest and SuperPi tests are only useful for finding max OCs settings, as Prime95 Blend will fail far faster than those two tests.

 

[BlueAcolyte]

Doesn't small FFTs stress the CPU most?

 

[Shortass]

A full day of Folding@Home usually does the trick.

 

[Markfw]

Originally posted by: Shortass
A full day of Folding@Home usually does the trick.

W week in my case...

 

[Tempered81]

For stress testing, I usually run 3d flower box for 24 hours or more

 

[MadScientist]

Here's a good article on Prime95. Link. I use small FFts for intial testing of my CPU oc, then Memtest86+ to test my ram oc, and then 8 hours or more on the Blend setting to test both my CPU and ram. I've had ram overclocks that passed Memtest86+ but failed Prime95 Blend.

 

[imported_Scoop]

Can someone explain why you should run these tests for like 12 hours? What does it prove? Shouldn't 2 hours be enough if you never ever will do anything that stresses the CPU for that period of time.

 

[Anubis]

i normally do prme blend and small FFTs as well as memtest also run 3dmark and sandra, and then general use things like gameing and such

ran a 6 hour compression render in sony vegas on my new one, figured if any real world app crashed was gonna crash it it woudl be that, but it did fine

 

[Idontcare]

First I try and find BIOS settings necessary to get the system stable at a particular FSB frequency.

For this I set ram time loose (usually loose just means SPD) and memory ratio to 1:1 (called synch on some mobo). I also kick the CPU multiplier down to 6X to eliminate CPU stability from the equation. And I turn off EIST/C1E and virtualization.

Then I overclock the FSB and use Prime95 Blend as my stability test, upping volatages as necessary on NB and MCH to make the system stable.

Then I work on fine-tuning the ram, tightening the timings and testing with memtest+ 1.70. If memtest is passing then I do large FFT until I am satisfied (usually overnight 12+hr runs).

Then I work on clicking thru the CPU multipliers and upping Vcore while keeping check with small FFT to ensure the CPU is stable at each frequency vs. Vcore.

I actually intentionally click thru each multiplier in iterative fashion, one at a time, and determine minimum stable Vcore for small FFT for each.

This allows me to make a plot of the data which (1) allows me to pre-determine the Vcore needed for the next higher multiplier and (2) allows me to determine when I have exceeded standard Vcore scaling and entered into an unsafe Vcore regime for the specific chip I am overclocking.

When I am all done getting my CPU up to max desired multiplier for a given FSB then I retest memtest, large FFT, and small FFT to confirm all the system's components are happily cooperating.

 

[VirtualLarry]

Originally posted by: MadScientist
Here's a good article on Prime95. Link. I use small FFts for intial testing of my CPU oc, then Memtest86+ to test my ram oc, and then 8 hours or more on the Blend setting to test both my CPU and ram. I've had ram overclocks that passed Memtest86+ but failed Prime95 Blend.

That's a good point. I hadn't thought of that. Especially with the multi-threaded Prime95 25.5, it probably hits the memory controller more than the single-threaded Memtest86+.

I usually use Prime95 small FFTs to test my CPU OC (24hr), and then run Memtest86+ (24hr) to test memory stability. I don't really use Blend, perhaps I should.

 

[MadScientist]

Originally posted by: Scoop
Can someone explain why you should run these tests for like 12 hours? What does it prove? Shouldn't 2 hours be enough if you never ever will do anything that stresses the CPU for that period of time.

For most probably 2-3 hours is sufficient, usually if your oc is not Prime/Orthos/Memtest stable it will error within the first 15 minutes, but I've had errors running Prime95 Blend setting after 2 hours.
My last build was for my son, a graphic designer, so in his case I made sure his oc'ed Q6600 sytem was 24 hours stable. His 3D renderings can take many hours to complete and are very CPU and ram intensive.

 

[jameswhite1979]

I run Prime95 and at the same time SuperPI M8 with CoreTemp and CPU-Z

 

[devilchrist]

Looking at my test It seems on the large fft you can get teh max temp by doing custom in-place large fft with 128k min and 320k as max.

this when when I get the hottest cpu temps recorded

 

[devilchrist]

I'm tring a new setup where i run 2 instances of older versions of prime95, no processor affinity selected. one is running Small FFT and the other is running Blend.

At the same time I'm running Coretemp, SpeedFan and HWmonitor.
Coretemp and speedfan gives a nice log of the temps and voltages
while HWmonitor gives a quick glance at the max temp/voltage on of the component has reached.

I'd guess that this would be the real world situation at max load of running different programs.

 

[nyker96]

I like to test thing separately first then combine test. 2-3 hours of memtest86+, small fft 1-2 hrs, then either OCCT/ortho belnd for 8 hrs.

 

[Big Lar]

I usually run Memtest in Dos for 2 hours, then start to OC reverting back to memtest when upping the fsb in large amounts. Mostly because I hate re-installing windows.

Then once I settle on an OC I think is high enuff for me,I run OCCT for 30 minutes, then I run memtest in windows for 2 hours, then Orthos Blend for 4 hours, then Orthos Small fft for 4 hours and finally Large fft for 4 hours. After all that, I'll run some bench's with SuoerPi. If all that is good, I am done.

 

[Rubycon]

Quickie - OCCT standard 30 min or 1 hour.

Long run - Intel Linpack. Text

 

[Duvie]

For cpus

I do 24 hours of small fft minimum. Then I run F@H for a week or 2 non stop......

For memory

I run memtest over night if 1gb or less....2gb I may let run longer. I like lots of multiple passes. Then I run F@H some more....


MOst things i run are no where near as intensive as prime95 or F@H. Prime will usually stress my power to limits I wont achieve in my actual uses of a PC. F@H seems to be a great rpogram for weeding out those uncommon but may eventually happen stability burps....

If I do those things, my PC will run effortlessly in temps and power until I am ready to start changing it all up again....

 

[Zenoth]

I always start with Memtest86+ for at least 3 passes (minimum), if not more than 6 or 7 passes (a good 5+ hours usually does the trick, but the longer the better). Then I go into Windows (if Memtest passes had not a single error) and try a solid 3+ hours of Small FFTs, and if that works I immediately switch to Blend Test (Orthos or Prime95). If that works I then switch to OCCT for more or less 8 hours test.

In the end I combined:

Memtest86+
Orthos
Prime95
OCCT

 

[Thor86]

Originally posted by: Scoop
Can someone explain why you should run these tests for like 12 hours? What does it prove? Shouldn't 2 hours be enough if you never ever will do anything that stresses the CPU for that period of time.

If you care about data integrity over time, then you know running these test for 24-48 hours without errors will not contribute to data integrity loss.

I've had Prime95 Blend tests fail on me at the 42nd hour of testing.

 

[superstition]

I've had Prime95 Blend tests fail on me at the 42nd hour of testing

How do you know it's not a random variable issue, then? Why not test for 200 hours?

 

[Idontcare]

Originally posted by: superstition

I've had Prime95 Blend tests fail on me at the 42nd hour of testing

How do you know it's not a random variable issue, then? Why not test for 200 hours?

You never know the precise reason why, but data is data and with more data you can proceed to iterate thru another round of tweaks.

Stability is such a user-defined subjective concept based on experience and expectation.

There's not much point to get passionate about it (not saying you are), but it does help in general for a community to have some loosely accepted terms and conditions behind the term.

A 42hr prime test would be extreme to me, but after a 12hr prime test I do run my computeres 24x7 at full load with my application of interest and if reboots happen then I do tweak up the overclock until it doesn't happen...be it for weeks or months between system instabilities.

How useful is this info for anyone else here? Not very useful unless someone else runs the exact same applications as I do. Just as your exact methodology won't do me much good either, beyond a certain point. But up to a certain point your method will work for me just as well as for anyone else.

The spirit of the methodology is all we should really hope to agree on. I.e. "Prime for something >2 hours and <24 hrs, how about we agree on 12 hrs?"

12 hrs is kind of an easy target as it is over night, just setup the run and go walk Jay Leno before heading to bed. Then check in the morning. Done.