How do you stability test?

[Smartidiot89]

Just as thread title says, how do YOU stability test your AMD CPU for 24/7 usage? I am talking about your everyday clock, which you use when you game, fold, etc. etc.

Usually, I clock in steps and use a simple Linpack called "LinX" and run about a 100 small problem sizes (takes about 2 minutes), then I do a few Large ones for 10 minutes. If that passes I continue.

When I think I found my stable sweetspot, start LinX and go to bed, a few hours of Linpack is enough, then I do memtest 500%. Don't tell me it's overkill to run Linpack for so long, I've had this CPU in particular tbh, being LinX stable for 2 hours but yet giving me rare - but random crashes. If I find any trace of instability I decrease my clock a step and try again.

[chew*]

This is how I do it......IBT and the other programs are great if you want max heat ( more than any bench or any realworld program will produce ).........Me personally I think Prime blends stresses the IMC more.....Just my preference......no right or wrong way.....I just got a PM from someone that can pass IBT and fails prime so to each his own.........I use what works for me.

To be truthfull there is no subtitute for 24 hours straight stress testing.....Even I have gotten lazy in recent times...I remember a time when 24 hours was the minimum I would run....

http://www.xtremesystems.org/forums/showpost.php?p=3771025&postcount=240

[pinoyxpryde]

I do stuff on my comp till it crashes, then I turn things down a notch.

[CobraXP]

Is OCCT good? Or Everest? I have used both of those in the past along with Prime 95 also. I have never used LinX or IBT.

[chew*]

Is OCCT good? Or Everest? I have used both of those in the past along with Prime 95 also. I have never used LinX or IBT.

Sure any of them are good, but I still say that time matters most.....people have gotten lazy over the years......devised quicker methods for stress testing......20minutes of this is equal to an hour of that.....eh to each his own.....

Still no substitute for 24 hours straight IMHO.

Seems I'm really lazy now when I dig through old screenies.....

http://members.cox.net/wmdieselmc26/eventighter.jpg

[CobraXP]

Sure any of them are good, but I still say that time matters most.....people have gotten lazy over the years......devised quicker methods for stress testing......20minutes of this is equal to an hour of that.....eh to each his own.....

Still no substitute for 24 hours straight IMHO.

I have seen systems crash after 5 hrs of stress testing. I agree with you 100%. 24 hrs will also get the weird smell off the parts while being stressed too :)

[Mescalamba]

OCCT 35 - 65 mins. Usually enough. Especially Linpack is good killer. Ofc 24hour testing is good, but its more cooling test than HW test.. at least it worked for me.. and yes Im lazy. :D Plus I don't have much computers.. so if it pass 35 mins OCCT, its usually fine.

[Smartidiot89]

This is how I do it......IBT and the other programs are great if you want max heat ( more than any bench or any realworld program will produce ).........Me personally I think Prime blends stresses the IMC more.....Just my preference......no right or wrong way.....I just got a PM from someone that can pass IBT and fails prime so to each his own.........I use what works for me.

To be truthfull there is no subtitute for 24 hours straight stress testing.....Even I have gotten lazy in recent times...I remember a time when 24 hours was the minimum I would run....

http://www.xtremesystems.org/forums/showpost.php?p=3771025&postcount=240
I do agree. Remember running LinX(basically OCCT and IntelBurnTest, it's a Linpack :P) for 2 hours and wow the clock is stable!

But as for real world usage, it was stable, except for 1 crash on avarage every 24hours, so I have had to decrease my clock a notch.

According to what I've read, stability testing with Linpack needs to be done 2 ways;

A huge amount of Small TTFs to stress the Cache of the CPU, and Large TTFs to stress the CPU/Memory to the max, but as you said there is no exact science.

Btw Chew* do you know a good way to stability test my NB clock? Should I just run a Linpack, or memtest perhaps for that?

[chew*]

Prime blend for NB seems to be the best for me....3 hours seems to be minimum though......

I recall it stressing the IMC quite well on 754 and 939 and it seems to do the same with PH II to this very day.

I found memtest pass = not stable.....drop 20 mhz from memtest pass and thats usually the ballpark......I only run it so I don't hose my OS though, I hate reinstalling...about as fun as watching paint dry.

Memtest minimum 1 full pass and 10 loop test 5.

Boot can run PI 1m, everest........fail 32m.....

Back off 20 Mhz pass all benches.....

Right now i can pass memtest 7-7-7 1704.....cant pass 32M 1688mhz. 1680mhz passes......So it seems consistent across a wide range of boards.

Back off just a tad more for prime stable.

[Hemi345]

How do I stability test? 10-15 minutes of Orthos/SP2004. If temps are okay and no errors, I start gaming. haha

I used to run all those tools for hours and hours but IMO, they're no substitute for testing stability with how you're gonna be using the machine. I've seen memtest86/orthos/etc pass after running 24hrs yet a game keep crashing because the overclock is too high.

[Oldguy932]

I ran the Gimps portion of prime 95 for like 2-3 months. If that's not stability I don't know what is. Not to mention I played games on top of that running.

[Daveburt714]

Usually, I clock in steps and use a simple Linpack called "LinX" and run about a 100 small problem sizes (takes about 2 minutes), then I do a few Large ones for 10 minutes. If that passes I continue.

I do pretty much the same thing except, I use "Wprime" to get a rough idea of what works. 32M first, then 1024m, when I get that stable, I'll run Cinebench.
If all that goes without problems, I start a 10 pass IBT run, turn on some tunes and surf the web... :D



When I think I found my stable sweetspot, start LinX and go to bed, a few hours of Linpack is enough, then I do memtest 500%. Don't tell me it's overkill to run Linpack for so long, I've had this CPU in particular tbh, being LinX stable for 2 hours but yet giving me rare - but random crashes.

Maybe I'm a little skiddish (and boards have gotten better) but I've lost an MSI DKA 790GX and an Asus M3A78-T by stressing things overnight (both due to PWM failures). Other than Folding/Seti or other distributed computing when are you ever going to push things that hard? I DON'T run stress programs overnight! Of course if DC is what you intend to use the machine for you definetly need to keep that in mind... ;)


If I find any trace of instability I decrease my clock a step and try again.

I do the same thing... After I've found the Max the CPU can do using Multi's, I push my HTRef/Mem clocks, then the IMC and the HTLink speed. Then if I do get instability, one bump down on the HTRef lowers everything a little bit... :p:

It may not be the best way, but the process has served me pretty well.. :D

I've been playing with these chips since the 9600std and I'm not sure if you'll ever be done tweaking, but that's what makes'em fun.. :yepp:

[HuffPCair]

intel burn test 10 or 15 times and then game and if it last its good enough for me.

[Smartidiot89]

Maybe I'm a little skiddish (and boards have gotten better) but I've lost an MSI DKA 790GX and an Asus M3A78-T by stressing things overnight (both due to PWM failures). Other than Folding/Seti or other distributed computing when are you ever going to push things that hard? I DON'T run stress programs overnight! Of course if DC is what you intend to use the machine for you definetly need to keep that in mind... ;)
I remember using LinX for 17 hours on my AMD Athlon x2 4000+ 2,1ghz clocked @ 300 x 10, with 1.41v i think. The CPU was at 67C under Linpack that was fun :ROTF:

But yeah i'll think about what you said, I guess a Quad-Core is abit more demanding on the PWMs then a old low-end Athlon x2 :rolleyes:

[JoNo216]

I usually just run prime for 5 min and then bump it up retest until it fails then work backwards slowly to find stability. After 20 min I consider it stable enough for normal usage. Then i just game with the overclock and if I notice hiccups, hangs, or crashes I will back off the overclock and retry it.

Once it's good I haven't had any real problems with it...:up:

[i found nemo]

for cpu i run prime for about a night ( 6-8 hrs )

for mem/imc i run memtest for a night ( same as above )

last but not lest, for vid card, max mem test for 1 hr and max core for 1 hr. with ati tool

[Entropicity]

For me, I run prime blend and if it passes the first 10 tests, it should be stable for 24/7 usage.

Heck.. I've even had systems work rock stable for gaming and work the entire month despite failing at test 7.

I'd say 99% of the time if your system can cross the first 10 minutes of prime blend without issue, it should run just fine for daily usage. I havent seen anything contrary otherwise.

Some others might disagree with me.. but to each his own.

[Plan.B]

if I can play Oblivion for 4 hours and crunch for 2 months with no issues I call it stable

[AgentGOD]

Try IntelBurnTest (http://www.xtremesystems.org/forums/showthread.php?t=197835), it brings many "prime stable" rigs to their knees in seconds. It is recommended to test on XP/Vista 64-bit.

[chew*]

Try IntelBurnTest (http://www.xtremesystems.org/forums/showthread.php?t=197835), it brings many "prime stable" rigs to their knees in seconds. It is recommended to test on XP/Vista 64-bit.

Funny becasue I got a PM from someone that passed IBT and failed prime instantly ;)

[wez]

Prime, memtest and long gaming sessions. Thats what I use :)
I don't leave prime unattended for long times anymore, perhaps these new boards can handle 24h full load runs, but why take the risk. A couple of years ago I could run prime for days, and sort of it was needed. If it didn't pass you'd be sure you would see an occasional bsod/crash. :shakes:
Today I could stretch myself to running three blend threads over night, but four threads and 100% load for 24h is just ridiculously overkill for this system. I can sleep well with memtest running over night though. :p:

Oh, and while doing ongoing oc in win I usually use 3 blend threads for this board. Perhaps it takes a minute longer to fail, but it wont hard crash/bsod as it tends to do if things go wrong while running 4 threads. And yes I blaming you, Asus.

[Newblar]

10 runs ibt + 5hours prime 95 + a few hours memtest
kekkeke

[Tom128]

Depends on the stability I am looking for. For my main rig that I game on if it passes an hour of prime blend I consider it fine for use. If I need it to be 100% stable for say crunching, I would run all 3 prime tests for 18+ hours each before I felt comfortable.

[Razrback16]

I've never been too crazy in my overclocks -- usually something solid but nothing record breaking. I continue bumping clocks up once per day (I rarely shut my rig down) and game randomly just like I normally would -- if I get a lockup/reboot I make a change, whether it be more volts, back the clocks down, etc. and try again. Once I have made corrections for these types of lockups in the real world apps like games and meaningful apps and feel I have a solid clock, I will pull up Prime95 and run a 4-core stress overnight at the minimum and sometimes if it's a work day the next day, let it run while I'm at work and see if it's still alive when I get back. If it is, it's stable.