What's your Prefered system stability test?

[Gary Key]

Crysis....... :laugh:

 

[amenx]

Crysis? really? I dont think it can stress the CPU as much as orthos nor the GPU as much as ATiTool.

 

[devilchrist]

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?

Or another question would be is it possible for a test to fail at 42nd hour without overclocking? It's entirely in the possibility region.

when i run blend test, i custome set it to run from 8k to 4096k

 

[geokilla]

I usually run Orthos Blend or Small ffts for 12 to 24 hours. Then I run Memtest test #5 for 2 hours.

 

[VirtualLarry]

Originally posted by: Thor86
I've had Prime95 Blend tests fail on me at the 42nd hour of testing.

That's an interesting data point.

 

[Idontcare]

Originally posted by: devilchrist

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?

Or another question would be is it possible for a test to fail at 42nd hour without overclocking? It's entirely in the possibility region.

when i run blend test, i custome set it to run from 8k to 4096k

That is akin to a "burn-in" test done to ensure all components are properly functioning as-received.

All new components should be tested for "burn-in" when first purchased as manufacturers stocks just to ensure you didn't receive defective or near-death (walking wounded) components.

Doesn't everyone do this? It would be a real PITA to seperate typical hardware faults from overclock induced faults if you don't first determine that hardware faults are not occuring

 

[CatchPhrase]

Originally posted by: VirtualLarry

Originally posted by: Thor86
I've had Prime95 Blend tests fail on me at the 42nd hour of testing.

That's an interesting data point.

If you tested it stock it would give in at some point also.

 

[LOUISSSSS]

12 hrs on Memtest86+ v2.00 (dos based?)
20mins-1hr of HCI Memtest (windows based)
12-24 hours of P95 v25.6 Blend (windows based)

keep in mind that windows based tests can fail due to a corrupt operating system from prior failed overclocks and unstable ram/cpu/drivers.
example:
MB: DFI LANPARTY NF4
CPU: AMD X2 4800+
RAM#1: Kingston Valueram
OS: XP 32

fresh install of XP, no initial stress testing (memtest/p95 etc), loaded up CS:S, BSOD. reboot etc etc SAME PROBLEM
buy new ram (RAM#2: OCZ platinum) replaced kingston valueram and get same problems under same OS.
Problems magically went away after a fresh install of new ram.

Solution: kingston valueram was incompatible with DFI lanparty boards which corrupted the OS feeding it corrupted data. even with new ram installed under same OS the corrupted data was already saved in the hard drive.

 

[lopri]

I run quick Memtest (Test 5) first whenever I feel the chance of OS corruption by memory overclock. After that, LargeFFT all the way. 3DMark05 CPU test is a very good quick test for dual-cores before wasting too much time in the process of overclocking (but not for quads).

 

[Capitalizt]

I do Orthos priority 8 for 20-30 minutes. This pushes the CPU to the limit, far beyond what any real world game or app is going to do. If the system passes that, it is stable enough on my view. There is no point running it 12 hours+ like some crazy people. This just causes the CPU to run extremely hot for a long period of time...and we all know heat = CPU death. I never understood the point of very long stress tests...It is only reducing the life of your CPU.

 

[bunnyfubbles]

Originally posted by: Capitalizt
I do Orthos priority 8 for 20-30 minutes. This pushes the CPU to the limit, far beyond what any real world game or app is going to do. If the system passes that, it is stable enough on my view. There is no point running it 12 hours+ like some crazy people. This just causes the CPU to run extremely hot for a long period of time...and we all know heat = CPU death. I never understood the point of very long stress tests...It is only reducing the life of your CPU.

Right, so you don't use your computer for anything serious, just a toy. There are many situations where an app will require far more time than 20-30 minutes to complete an intensive task. And as far as that whole BS about reducing the life of the CPU, a run of 12 hours one night to make sure your system is stable is a mere drop in the bucket at most. In fact I'd say its far worse to be running a potentially unstable system for months on end than to really push it one night to make sure everything is good to go.

(then you have to look at the guys who do DC such as F@H where the DC clients are very stressful on the CPU/system yet they run these things 24/7 for months on end - years even. Simply put, the argument about stress on a CPU is extremely overemphasized. The only major factor you should have to worry about in terms of reducing the life of any of your components is voltage. Keep your volts at safe levels and you can run your parts 100% all day, every day and you'll almost always be able to still have a running part by the time you're ready to upgrade.)

I'd rather not have my system crash on me one day that reveals a corrupted OS; even with stringent backup routines I'd rather not put up with any unnecessary risk of lost data because I'm too lazy to do something properly.

Oh and overclocking can kill other parts of your computer, not just the CPU. Parts can/will degrade and overclocks that were once rock stable can become unstable over time. If your system can last a torture test overnight the chances of it remaining "stable enough" for the entire life you intend to use it for go up quite a bit.

You think we're crazy for running 12+ hour tests...I think you're crazy for not being able to do something as simple as run a test over night while you're sleeping, or perhaps during the day when you're at work/school...the only reason I can think of not to do something so simple is for the people who don't want to find out that their system might not be stable, that their bragable overclocks are too high...

 

[brencat]

I did the following for my new build this past December:

1. CPU @ stock with RAM 1:1 and loose 5-5-5-15 timings, memtest86 (all tests) overnight (8hrs minimum), then Orthos Small FFTs 2 hrs.
2. Overclocking begins....CPU first, with RAM still @ 1:1 and loose timings. 3 - 5 hrs Orthos Blend at each FSB bump adding vcore as necessary.
3. RAM clocking next...via dividers, one at a time, then methodically tighten timings. Memtest86 overnight (all tests) at each divider setting and/or timing adjustment, followed by 5 SP1M runs to verify which divider/timing combo gives best speed boost. I try to minimize the vDIMM necessary to achieve stability.
4. Final OC stability check...Orthos blend 8hrs, then 10 loops of 3DMark06

Total process took about 2 weeks.

 

[nullpointerus]

I do the following for a given target OC:

-- Test desired FSB clock w/ Prime95 Blend 12 hr., raising vmch, vfsb, and/or vdimm as needed
-- Test desired RAM clock w/ Memtest86+ 12 hr., raising vdimm and/or vmch as needed
-- Test desired CPU clock w/ Prime95 Blend 12 hr., raising vcore as needed
-- Test whole system w/ F@H for one week

If my system passes those tests, I can keep the overclock indefinitely w/o any problems.

Those who balk at 12 hr. runs ... can you not set the system to run overnight?

Originally posted by: LOUISSSSS
12 hrs on Memtest86+ v2.00 (dos based?)
20mins-1hr of HCI Memtest (windows based)
12-24 hours of P95 v25.6 Blend (windows based)

keep in mind that windows based tests can fail due to a corrupt operating system from prior failed overclocks and unstable ram/cpu/drivers.
example:
MB: DFI LANPARTY NF4
CPU: AMD X2 4800+
RAM#1: Kingston Valueram
OS: XP 32

fresh install of XP, no initial stress testing (memtest/p95 etc), loaded up CS:S, BSOD. reboot etc etc SAME PROBLEM
buy new ram (RAM#2: OCZ platinum) replaced kingston valueram and get same problems under same OS.
Problems magically went away after a fresh install of new ram.

Solution: kingston valueram was incompatible with DFI lanparty boards which corrupted the OS feeding it corrupted data. even with new ram installed under same OS the corrupted data was already saved in the hard drive.

Excellent post! :thumbsup:

Generaly, I don't like memory overclocking because (a) I don't understand it as well as other kinds of overclocking, (b) it has little real-world performance benefit, and (c) I've corrupted two Windows installations by (naively) assuming Memtest could prove memory overclocks stable. It can't.

The only thing Memtest86+ can prove conclusively (in conjunction with other tests and proper test methods) is that a memory overclock is NOT stable.

To be safe, I created a second partition w/ clean installation of Windows. It only takes about 30 minutes. There I can boot, set it as the default OS, overclock, and run all my stability tests without ever touching my main OS/data partitions. When the overclock is stable as I can make it with synthetic tests, I switch back to my main partition for F@H 24/7 running in the background while I'm using the computer normally.

(If you do use Vista for a dedicated overclocking partition, I recommend disabling Superfetch/ReadyBoost, Automatic Updates, and automatic defragmentation. Windows attempting to defragment the main OS partition in the middle of a 12hr. Prime95 stability test strikes me as being a REALLY bad idea...)
:Q

If you look at people who experience constant video driver related BSODs etc. -- how many of those people have tested their overclocks well? Even if their BSODs persist when clocks are lowered back to stock, are they trying to run their games with driver files that were already corrupted by overclocking? For that matter, how many have tested the stock components well (as Idontcare suggested)?

Defective hardware, corrupted OSes, and unstable overclocks -- 3 things I never want to experience again!

 

[Martimus]

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.

I can tell you that my system would run fine for approximately 6 hours at 2.75GHz, and then have an error using Blend on Orthos. I had to back it down to 2.7GHz to keep it stable for 12 hours at voltages I thought were acceptable.

If you don't stress test it for a long period of time, it can randomly crash on you when doing something intesive, like playing a new game. I had this happen with the above overclock, because I figured 6 hours of stability with Orthos was probably good enough for my everyday tasks; but it would crash - albeit infrequently - when playing TF2. Now it never crashes, so I use a longer stress test when I test my system for stability.

 

[ttnuagadam]

i usually leave prime95 running overnight.

 

[spinejam]

Memtest 2 passes
Blend for 8 - 12 hr's
3DMark05 benchmark
3DMark06 benchmark

Increase the system's heat a little more by stressing the videocard with a few hours of:

Crysis
COD4
UT3

Then the finale -- let the wife use it for a few days and see if she can tank it!

If no problems arise, then you're good to go!

 

[LOUISSSSS]

Originally posted by: spinejam
Memtest 2 passes
Blend for 8 - 12 hr's
3DMark05 benchmark
3DMark06 benchmark

Increase the system's heat a little more by stressing the videocard with a few hours of:

Crysis
COD4
UT3

Then the finale -- let the wife use it for a few days and see if she can tank it!

If no problems arise, then you're good to go!

i laughed, but thats a very bad idea. too many unknown variables in that test

 

[skillyho]

I'm not nearly as thorough as some of you....my machines aren't essential to my work (that's what my laptop is for) so I do the basics and then play.

1) MemTest 3-4 Passes
2) Orthos Blend for 1 hour

That's about it....I'll then put 6-8 hours of BF2 and UT2k4 (cause I'm old school like that) and call it stable.

 

[aigomorla]

Originally posted by: jaredpace
For stress testing, I usually run 3d flower box for 24 hours or more

lmao!

the screensaver? or is there a program called flowerbox? This is a real question, because im assuming the flowerbox is both a gpu + cpu option? and its a new program i never heard besides the screen saver?

Originally posted by: LOUISSSSS

Originally posted by: spinejam
Memtest 2 passes
Blend for 8 - 12 hr's
3DMark05 benchmark
3DMark06 benchmark

Increase the system's heat a little more by stressing the videocard with a few hours of:

Crysis
COD4
UT3

Then the finale -- let the wife use it for a few days and see if she can tank it!

If no problems arise, then you're good to go!

i laughed, but thats a very bad idea. too many unknown variables in that test

LMAO! ditto on that. who knows what she downloads off those myspace/facebook sites?

Originally posted by: Rubycon
Quickie - OCCT standard 30 min or 1 hour.

Long run - Intel Linpack. Text

Ruby why occt? i never really liked it, had too many bugs, and also wasnt very quadcore friendly at first. After that i never gave it a second shot seeing how prime was multi core optimized.

 

[VirtualLarry]

At one point, before I had heard of Prime95, I used to use the OpenGL "pipes" screensaver as a stress test, on my Athlon XP CPU. It stresses the FPU quite a bit, so in that sense it was a sucessful stresstest for that CPU. Ran it for a few hours until visual anomolies did or did not show up.

 

[Modelworks]

15 million polygon scene rendered in max or maya, left for 8 hours.
If the system doesn't crash the render, then I know its stable.
Accesses hard drive, ram, cpu, video.

Purposely rendered on a system with only 4Gb ram, when 8GB is required, so it uses paging file heavily.