How long to stress test for?

spleenharvester

spleenharvester

spleenharvester

Associate
Joined
4 Dec 2008
Posts
1,812
How long do you lot stress test for? Is 2 hours OCCT good?

I'm undervolting atm and have gone from 1.4v -> 1.18v, has passed an hour and a half of OCCT so far. 1.13v failed after 40 minutes.

I'm not doing 24 hours of prime or any crap like that 'cause it just isn't worth the effort. :p

Cheers
 
Most overclocks will fail within the first 5-10 minutes of stress testing so 2 hours is pretty solid. Most would say 8 hours to be sure :)
 
some will say a 4-6 hours of prime or 50 passes of intel burn test,
but personally, i do the same as you
a couple of hours of OCCT or intel burn test, i have never had any problems
 
spleenharvester said:
How long do you lot stress test for? Is 2 hours OCCT good?

I'm undervolting atm and have gone from 1.4v -> 1.18v, has passed an hour and a half of OCCT so far. 1.13v failed after 40 minutes.

I'm not doing 24 hours of prime or any crap like that 'cause it just isn't worth the effort. :p

Cheers
Click to expand...

Is that OCCT Linpack? If so I wouldn't recommend using it more than 2 hours as it is equivalent to LinX or Intel Burn Test as all three use the same Linpack algorithm. However is there any way of knowing what GFlops values you get in OCCT?
 
WingZero30 said:
Is that OCCT Linpack? If so I wouldn't recommend using it more than 2 hours as it is equivalent to LinX or Intel Burn Test as all three use the same Linpack algorithm. However is there any way of knowing what GFlops values you get in OCCT?
Click to expand...

Why not more than two hours just outta interest? The CPU's absolute max is 35*c undervolted, should be fine right?

And nopes, though 10 tests of IBT show a consistent amount.
 
I tend to go with 20 runs of LinX/IBT, followed by 8hrs of prime blend. Saying that though, ive had blend fail after 5 hours or so on occassion.
 
spleenharvester said:
Ah right, would 2 hours OCCT -> 100 Passes IBT on high do the trick ya'think?

Cheers
Click to expand...

spleenharvester said:
Why not more than two hours just outta interest? The CPU's absolute max is 35*c undervolted, should be fine right?

And nopes, though 10 tests of IBT show a consistent amount.
Click to expand...

When you use LinX or Intel Burn Test, the main factor you are looking for is the GFlops values i.e. how fast cpu is processing/executing the matrix equations stored in the ram. In other words can your cpu cope with high GFlops values and pass the stress test because at high GFlops values, cpu requires more vcore and it also outputs more heat.

The theoretical maximum GFlop value for your cpu is :

no of cores x no of Flops/per core/per cycle x cpu speed = 4 cores x 4 flops x 2.8GHz = 44.8GFlops.

In IBT or LinX if you can get 70-80% of the theoretical maximum or atleast 32GFlops and run the stress program for 50 passes and are successful, then you will have a stable IBT cpu:)

Edit: One hour of IBT/LinX is stressfull enough provided you get high GFlops as mentioned. You will soon see the load cpu temps reaching high values.
 
Last edited:
spleenharvester said:
GFLOPs have not dropped from the CPU's stock, still at ~35.75. :)

Gonna start some prime blend now for 2 hours, then maybe an hour of gaming and then a few IBT passes to check GFLOPs again.
Click to expand...

That's good.

You are getting atleast 78% of the theoretical maximum or 35.75GFlops out of possible 44.8GFlops. If you can run IBT for 50 passes and it completes the test successfully, then you are IBT stable:).

However in my experience stock cpus pass IBT any way as they are generally slightly over volted by manufacturers to ensure you don't crash at stock speed.

High GFlops IBT stress testing becomes more crucial when your cpu is overclocked. That's where it can crash your system easily.
 
spleenharvester said:
Might just do 1 hour of prime, and if it finishes stable here just put the volts up a notch to be rock stable lol. Wanna play JC2.
Click to expand...

Prime95 is usually conducted over a period of 8 hours to ensure long term stability and also to pick out any errors which couldn't be detected at earlier stages of stress testing as the loading is consistent.

In the case of IBT/LinX, as these stress your cpu harder than prime and are cyclic in nature, only 20-50 passes are usually recommended for stability.

So IBT/LInX are short stabiliy stress tests which stresses your cpu harder.

Prime95 is a long stability stress test which is much better at picking out any errors in later stages or in a long term running.


Saying that I have never stress tested my cpu at stock speed:p.
 
setter said:
I tend to go with 20 runs of LinX/IBT, followed by 8hrs of prime blend. Saying that though, ive had blend fail after 5 hours or so on occassion.
Click to expand...

I've had it fail at 23 hours 52 minutes :(

What people forget though, is blend is cycling memory therefore it's most likely to have a failure caused by the memory. Issues like yours should only require a bump of the nb volts to stabilise where as my hiccup ended up pointing to flaky ram, which failed 9 months later :D For cpu work small/large fft is better.

I've actually become fond of AOD. Seems more effective than running prime/linpack simultaneously.
 
If I have a flaky OC it'll show up within 10 minutes or less on a small fft Prime run. Once I'm Prime stable and within temps at about an hour I'll do 20 runs of LinX to see if we're OK (although I'm a total LinX n00b so I have no real idea what I'm looking at).

Then when I'm confident I have something pretty solid I'll do a 8 hour small fft Prime run, and if it's OK after that I sometimes carry on to 24 hours if I'm really pushing it. To be honest though, if I fail after 12 hours I'll just drop the OC down a smidge and just live with it - I want to use my bloody computer, not stress and bench it!

I've not undervolted yet so I don't know if my testing will be appropriate.
 
Think I've managed to get a good split of 3.12GHz/1.25v, which is well under stock voltage. Time to test ;)

The longest I've had between failure is 2 and a half hours. Mine never crashes on clockspeed but voltage, the most I've ever hit is 3.9GHz on 1.6v briefly
 
Ive had 24hr prime small fft settings fail within 10 mins under blend, even a few nights ago, i was clocking my cpu to 4.2ghz, failed LinX with bsod, upped qpi a notch, LinX then stopped again on error after 3 mins, (no bsod though), upped vcore a notch, passed fine on the next run.
 
Indeed, 'tis why once it's 2hr stable, I'm going to up the voltage to the next setting for extra stability's sake.

Luckily, mine'll lock up with an 0x0000124 if it's unstable since I hit a voltage wall not a clockspeed wall, and if RAM's unstable it'll fail LinX, so it's not too hard to get sorted. :)

Lol feel sorry for this poor board, it's only been outta the box for half a day and it's taking an absolute battering :p the undervolting means it's still 95W though, so considering it's a 4+1 125w board, the VRMs'll cope fine.
 
Last edited:
I guess it depends on what you are doing with the overclock - maybe for bragging rights there should be some agreed standard such as x hours using x test ? For real world use you may never stress the CPU to the extent of some of these tests so if it hasn't failed after x time it's good to go for real world use. For my use I am happy that 1/2 hour on Prime 95 is enough to saturate the cooling system and is way more CPU load that I would encounter in real world application stress.
 
A full loop of Prime95 lasts about 10 hours. I've had overclocks fail after 7-8 hours.

It's down to how stable do you want it to be? I for one like to know I've reached rock solid stability hence the long testing.

I recommend:
- Prime95 for 10 hours
- LinX for 25-50 loops (all memory)
- OCCT for 2-4 hours
- 3DMark or games for 2-4 hours a pop
 
You must log in or register to reply here.
Back
Top Bottom