Three Levels of Stability Methodolgy?

[EarthDog]

Heh... this is all subjective... Here is my thinking:

1. What is the recommended quick test method to check an overclock for stability (less than an hour of testing)
Linx/Linpack/IBT/OCCT - Those are massive overkill applications. To me its akin to driving your car past redline to the rev limiter. So that is quick, dirty and way more stressful than P95 and a decent way to quick check stability.

2. What is the recommended good test method to check an overclock for stability?
I use Prime95 Small FFT for 8 hours to test CPU, then another 8 hours on blend.

3. What is the extreme recommendation to ensure you almost definitely won't see a crash under almost any condition
24 hours P95 Small FFT, than 24 hours blend.

To add to that, using the application/game and it doesnt crash.

EDIT: Just read the entire OP - IMHO Prime is still a contender in the game. Like I mentioned above, those programs, to me, appear to run your CPU past redline. Im at 100% load with P95, so what does Linx/Linpack/IBT/OCCT do more EXACTLY?? Integers? FPU? What does it do OVER Prime that makes it better? I cant get my PC to reach P95 temps, none the less those applications temps which are a few C higher. Thats why I believe that although Prime95 may be slower at finding errors, its no less effetive, hence my method.

 

[Automata]

Level one stability should be rock solid both crash and error free and this is really impossible because even stock clocked crashes from time to time and this is due to the software having a conflict or hickup. I have had plenty of new setups and when I was setting my baselines and updating everything the drivers may crash or the scary update may cause problems.

You are combining hardware instability with software instability. For the purposes of the thread, we should only be discussing hardware instability, since we are overclocking. We can't guarantee the stability of a system forever, but we can use probability to reduce instability to near zero.

 

[RollingThunder]

My methods, hardware only, no graphics card tests, very basic and suitable for a beginner.

Programs needed:

• Memtest86 v2.1 on a boot disk
• SuperPi
• Prime 95 or Orthos (same as Prime 95 but with a nice GUI)


New Assembly, all default voltages and timings (set manually) and assuming all tests “Pass” with zero errors. This just checks new RAMs / hardware on a new system, nothing more.

• Memtest86 from a boot disk: 5 – 6 full passes



Then I start my overclocks in small incremental steps while checking each step below and watching your temps:

• Memtest86: 5 - 6 full passes
• Superpi (32mb test only for now)
• Orthos for a few hours


I do this for each incremental overclock until I reach my temperature limits or have a “fail.” You must pass each program’s test 100% to advance.

When I am at the point of max temps and stable overclocks so far, I’ll run Orthos / Prime95 for 8 hours. If it passes, I’m done. If not, then I’ll back it off a “tick” and repeat Orthos / Prime 95.

Keep in mind this takes some time. The more advanced guys using advanced cooling, nitrogen, dry ice or heavy experienced graphics / gamers can and do skip the lower incremental overclocks and go right to the heart of it. They are in another advanced world and they often use other programs and methods. My basic method, which was taught to me here on OC Forums, works for daily working computers and it puts me right near my maximum stable overclock.

I’ll do this for each new build with just the operating system installed. Serious crashes really won’t hurt anything because this is the time for a beginner to familiarize himself with his BIOS and overclocking methods. The worst that can happen is a corrupted operating system, just reinstall it.

I may spend a week on my new system until I have learned a new BIOS and know when my max temps / overclocks are reached. In my opinion and for the use my working machines get, arguably I have reached a “Level 1” stable overclock. I don't do this every week or every month so I too need to refresh myself with a new build or hardware upgrade.

As long as a new overclocker realizes you have to walk before you run and has a little patience, it’s quite an enlightening and enjoyable experience. Of course it helps to have another machine available to reach OC Forums.

There is always someone here to help.

 

[Archer0915]

You are combining hardware instability with software instability. For the purposes of the thread, we should only be discussing hardware instability, since we are overclocking. We can't guarantee the stability of a system forever, but we can use probability to reduce instability to near zero.

It must be considered. That is why I said it the way I did. If your system crashes from time to time it does not mean the system is unstable.

 

[hokiealumnus]

1. What is the recommended quick test method to check an overclock for stability (less than an hour of testing)

For very quick stability testing, I use LinX (thread to download it from). 10 passes using 2048MB of memory. That will ferret out most instability issues very quickly.

2. What is the recommended good test method to check an overclock for stability

LinX for longer and using the maximum available memory - 20 passes maybe. Prime95 blend for 24 hours minimum; but as stated earlier in the thread Prime95 seems to have lost its efficacy over the years.

3. What is the extreme recommendation to ensure you almost definitely won't see a crash under almost any condition

Folding for more than a week is the most sure-fire way to test for stability I've ever tried. I've crashed Prime and LinX "stable" systems with F@H. It might take a while, but if there is instability there, F@H will find it.

Note: This applies only AFTER stability testing with LinX/Prime95, because you want to avoid crashing if possible so you don't mess up the test results.

/Slightly OT/

Dear Moderators:

You may feel free to position this thread in a more appropiate area if you so deem. Thank you.

Right now viewers per forum:
Cooling 75 viewers
Cpus 175 viewers
Motherboards 127 viewers
Video Cards 111 viewers
Storage 65
Memory 40
Display and Sound Technology 65 viewers
The rest of subforums under "Hardware" are on average showing less than 30 viewers.

I think this is the place for it, but to add eyes it has been added to the CPU stickies in both AMD & Intel CPU forums. That should cover most of the audience this seeks to address. If you'd be so kind, please keep the OP updated with as much of a consensus as the thread comes up with.

/Slightly OT/

 

[storm-chaser]

There are good threads in this forum for sure. Wheels already do turn.


Big three in and of "Stability".
Of further importance, is the factor of general system stability at manufacturer settings. Before overclocking or tweaking your machine, one should ensure that all hardware is operating properly and with integrity. Overclocking an unstable machine then stress testing it, finding errors, and spending hours troubleshooting a problem inherent to some faulty component, isn't something anyone wants to put themselves through! Sometimes a component is simply faulty to begin with, and needs to be replaced or switched out.

Holy hotrod it is so hard for me not to interject any car corrolary here. Just a minute one then. How the heck does it perform before I start beating on it. There got that out the way.

When I saw "felinusz" speaking to establishing a baseline...I felt kinship for what I had been taught to do in anything I intended to mod. For my part I am convinced that is the first step to accomplish before twisting the wick up.

The basis for my first post is in a sense to raise awareness (thank you Rgone, for your efforts here) over the years I’ve seen newbies amateurs and hardcore pro's alike struggle to make sure their glorious overclock passes 20 hours of prime or more. AGAIN, in my opinion I think this is overkill (for years and years we always used prime95 to test out new server builds before we wound put them into production or implement on a client’s network so I really have nothing against it. I think the base indicator of system reliability, aside from the obvious torture test, is daily use. If you are a system builder or a network administrator (like me) you spend enough time in front of said pc to get a sense of how well its running, just as a mechanic knows how well an engine is running by driving the car, running through the gears, taking it up to speed, basically just listening to the engine over the course of a couple days/weeks.
Much like Spybot search & destroy and hijackthis are long since obsolete, It’s just my feeling that prime95 is past its “prime” as well.
Prime95 is a very aggressive torture test. When I was tuning my Q9550, I was getting errors in prime after running for about an hour. However, after I torture tested with AIDA64, it would run all night no problems. I never had any blue screens or stability issues, despite the prime failures. I didn’t change a thing and its been running fine ever since (gaming, 3d rendering, etc) So the reality is that my overclock was in fact stable and prime95 wanted me to hunt down a problem that didn’t exist.

 

[bigben6]

ANy computer can crash under the right circumstances.... accept that firstly.... when consider your intended usage and determine what stable means for you, if your using secret gov programs and top secret data... dont OC, if you gaming and listening to music an OC is likely more annoying that any real detriment to the system.. I always wanted P95 short FFT for 2-4 hours then blend for 12+, that was good enough "for me" to call it stable, and with my usage, it never crashes...

 

[RGone]

I am looking at posts with an eye to upating the first post to chronicle what is going on in the thread.

If we can the post and reflect the 1 , 2 and 3 levels if possible it maybe a better display of a methodology. I know I said in the OP that there would be some overlap, but the closer we can come to a 3 level consensus for stability, maybe the better it will be for the newer overclocker.

Of course we can view stability objectively or subjectively. There can be a place for both views. If we say " I " did it and back it up with results then we have been ojective.

I am still suspicious that there are those that should not overclock. Are there those types of situations?

Thank you.
RGone...

 

[lilg_06]

Heh... this is all subjective... Here is my thinking:

1. What is the recommended quick test method to check an overclock for stability (less than an hour of testing)
Linx/Linpack/IBT/OCCT - Those are massive overkill applications. To me its akin to driving your car past redline to the rev limiter. So that is quick, dirty and way more stressful than P95 and a decent way to quick check stability.

2. What is the recommended good test method to check an overclock for stability?
I use Prime95 Small FFT for 8 hours to test CPU, then another 8 hours on blend.

3. What is the extreme recommendation to ensure you almost definitely won't see a crash under almost any condition
24 hours P95 Small FFT, than 24 hours blend.

To add to that, using the application/game and it doesnt crash.

EDIT: Just read the entire OP - IMHO Prime is still a contender in the game. Like I mentioned above, those programs, to me, appear to run your CPU past redline. Im at 100% load with P95, so what does Linx/Linpack/IBT/OCCT do more EXACTLY?? Integers? FPU? What does it do OVER Prime that makes it better? I cant get my PC to reach P95 temps, none the less those applications temps which are a few C higher. Thats why I believe that although Prime95 may be slower at finding errors, its no less effetive, hence my method.

I totally agree with this, of course not everyone will, but it works for me.

1. I use IBT max ram settings on 10 loops, takes 20-25 minutes with 8gb of ram. If I can get through this, I know that the system is very unlikely to stall and crash when I'm multitasking, because this puts loads of strain on both the cpu and ram and draws its maximum power output.
2. Then I run prime blend for 12-24 hours....
3. Then I run AIDA64 stability test with Heaven Benchmark simultaneous for about an hour just to ensure it will hold out.

AIDA64 is good for testing all the CPU's instruction sets, remember prime and IBT only use prime integer calculations, AIDA64 tests pretty much every instruction set your cpu is capable of...

 

[Dlaw]

Sorry for the Necro, but this thread looks to me like it's far from solved.

In my opinion, here is a good layout for instability:

Level 1 ("least" stable): run the most intense program you have; video editing, BF3, whatever you have that pushes your system. Intended for those who can afford a crash, maybe it's a spare system that's not used much, etc.

Level 2 ("average" stability): Run a program designed to stress your system, preferably one that "loops", or keeps repeating. It doesn't really matter which one, as none of the real world programs you will ever use come close to the load that you will be putting on your system with it. The longer the run, the more stable you can consider it. Intended for power users and enthusiast gamers; those who can afford a crash, but would prefer to not deal with one.

Level 3 ("most" stable): run your system at stock with the most advanced cooling you can afford, and a voltage tick above default. If you absolutely cannot afford a crash, then you absolutely can not afford to overclock. Besides, if you absolutely cannot afford a crash on your system, your system probably doesn't need an overclock anyway.

I do feel that some people shouldn't overclock their systems. If your system handles other people's data, it shouldn't be overclocked. If it's operation affects other systems, it shouldn't be overclocked (servers). If it isn't yours (kids...), it should be left the H-E-double hockey sticks alone!

 

[RGone]

Dlaw, I think in the pure sense of what you said for your response #3 above, you are absolutely correct. I worked day in and day out for 6.5 years with DFI and had to be able to contact them in CA and Taiwan and the Netherlands everyday. Emails had to remain readable and other customer information had to remain viable. I followed the advice of two old IT buddies that said leave it stock if you need your data. I took that to heart and basically used the same mobo for that 6.5 years without issue. Oh I had to change some sticks of ram out that failed and I had one hard drive failure in the system, but no loss of data.

If you don't have a 'play' system and you have data you want at all cost...I would leave the overclocking to others. IMO. Now the play system when I benched; it was all bets off.
RGone...

 

[RJARRRPCGP]

Lately, Prime95 can't test core stability as effectively as Linpack.

I have evidence that even with an AMD, Prime95 is not utilizing as much of the processor as Linpack, even when the OS reports 100 percent CPU usage.

Even with tweaks I did to Prime95, still inferior to Linpack when testing core stability! Linpack makes Prime95 look like "CPUBurn"!

Prime95's defaults suck:

1. Round off checking is disabled.

2. FFT settings not optimal for finding instability.

 

[EarthDog]

Lately? Has it changed?

Amazing how times change... my post from 9 months ago.. now aida64. Confirmed it does use the instruction sets vs others that do not.

 

[RJARRRPCGP]

Lately? Has it changed?

Amazing how times change... my post from 9 months ago.. now aida64. Confirmed it does use the instruction sets vs others that do not.

For very quick stability testing, I use LinX (thread to download it from). 10 passes using 2048MB of memory. That will ferret out most instability issues very quickly.


LinX for longer and using the maximum available memory - 20 passes maybe. Prime95 blend for 24 hours minimum; but as stated earlier in the thread Prime95 seems to have lost its efficacy over the years.

Well, at least since roughly 2008.

Looks a lot like, with the Vcore amount that Prime95 passes on, still then need at least one more Vcore bump.

But, Prime95 still seems to excel in bus stability checking. (High FSB OC'ing)
It's been looking like I'm gonna have Prime95 error out with the FSB beyond 363 Mhz on my Asus P5QL Pro. (P43)
IIRC, a Linpack test passed, before running Prime 95 for 5 hours. ( Prime95 errored out at 5 hours.) Then it got strange, Prime95 then errored out at 2 hours, then possibly less!

(Actually, Prime95 was failing faster, even when turning down the OC!)
Maybe it's just the darn core. That was with my E2180. It looked like it was gonna be fine at 375 Mhz FSB.
That was with my Thermaltake V1 heatsink, which seems to have crappy contact.

And hokiealumnus is right on!

 

[EarthDog]

There are stories ally over that go both ways... p95 stable but failed X.. X stable but failed p95... whatever works!