Your Stability Testing Routine

smeaney said:
This is true. I run prime for as long as possible to spot these mistakes but usually I want my PC back after a day or two. Usually I have found that if it will last 24hours, it will last anything but there have been some expections to that.
Click to expand...

You have to remeber that there is no such thing as 100% stable overclock system. Running any stress program for a period of time successfully just ensures that our pc is good enough to handle demanding tasks which obviously won't be stressing the cpus as much as the stress programs (prime 95, IBT etc) themselves. :)
 
5x IBT for finding max oc
when happy with oc 25x IBT to confirm stability, rinse and repeat whilst finding lowest vcore
once happy with all settings 50x IBT followed by 24x hour prime95 blend
maybe a few passes of memtest aswell if the ram is running out of spec
 
Justintime said:
My own system: current testing method
600 loops of IBT
20 loops of Linx
48 Hrs prime 95 Blend
1hr OCCT gpu test at native res and max shader complexity, 1hr power supply test.
few loops of 3dmark

Might be overkill but its rock solid. I don't depend on the system so i have time to test, i use two pcs mainly.

Clients system: 100 runs IBT
24 hrs prime 95
If its a gaming rig i'll do OCCT GPU for an hour, same for psu test.
Few loops of 3d mark
Click to expand...


overkill!!! 3 days testing.

i play games on my computer.
 
!/ Reset to stock, remove half the ram (12gb @ stock isn't stable), install 64bit vista on a spare hdd

2/ Overclock in a large jump to settings which are likely to work, e.g. bsck = 200, vcore = 1.35, qpi = 1.35, vdimm = 1.65, ram = 1200 cas 8.

3/ IBT for a couple of hours on maximum memory, if passes refine overclock

4/ When at whatever target I'm aiming at, IBT overnight, then prime blend for 24 hours, then prime small for 24 hours.

5/ If that went well, turn off some of the fans and repeat the overnight ibt to check for thermal headroom

Even then the bloody thing can still crash in use.

If that seems excessive to anyone, I spent the best part of a day debugging (matlab) code which sometimes failed on my desktop. I'm pretty sure the code is fine but the overclock isn't, on the uni servers it's slower but it didn't fail. It also worked consistently at stock settings. The overclock was not stable enough.

edit: Installing a new os to begin with is an alternative way of dealing with potential corruption during overclocking
 
Last edited:
20 passes of LinX/IBT
8 hrs prime 95 small fft
8 hrs prime 95 blend

For gpu's.
Few loops of heaven bench
3d mark 06/11
Games

Graphics cards are ran on a mild oc of 751/1502/1684 daily, though ive had them at 800/1600/1700 for a few benchmark runs, though they need a voltage increase at theese clocks, which tends to increase temps a lot.
 
Why is everyone doing long runs of both Prime small and Prime Blend? Blend will also repeat your small tests. You might as well just do Prime long after a small run.
 
In my case its just habit i guess, saying that though, its ages since i last ran any stability tests on my cpu oc, had it clocked at the speed in sig for over a year now.
 
1/2 hour of prime small fft stock ram
2 hours of prime blend stock ram (popping to shops)
12 hours of prime blend stock ram while I'm at work
12 hours of prime blend o/c ram while I'm at work.

For graphics card, as per the recommendation of some lads on here, I just run a couple of Unigine demos looping for several hours. Used to use furmark, but it would just cause the card to get hotter and hotter till it crashed. Doesn't represent proper stresses in my mind.

My system is stable using this routine.
 
Jokester said:
Why is everyone doing long runs of both Prime small and Prime Blend? Blend will also repeat your small tests.
Click to expand...

Ignorance in my case, i've never looked into what the various options are beyond what the program states. What approach would you suggest for a >24 hour prime?
 
JonJ678 said:
Ignorance in my case, i've never looked into what the various options are beyond what the program states. What approach would you suggest for a >24 hour prime?
Click to expand...
Well, it used to be suggested that if it failed on small FFT it was a CPU issue, but if it failed on large FFT it was a RAM/NB issue.

From memory Blend will run large FFT for a 2-3 hours, then small for 2-3 hours then large again etc etc.

Might be a case of leaving Blend to run until it fails and if it fails on say a small FFT test, do a small FFT for a couple of hours once you've adjusted your CPU voltages to see if that cures it, then back to blend for 24hours.
 
I just used the Prime 95 small fft.

Uping fsb and volts as needed, testing using prime 95 small fft for 30 mins, if pass, up fsb, carry on like this untill, max volts reached or cpu too hot, back down slightly then test for stability, for about 8 hours of prime 95 small fft.

Then just game on the machine, rock solid since June.
 
You must log in or register to reply here.
Back
Top Bottom