What do you (in your opinion) consider a stable OC? Discuss.

[Idontcare]

So it is not only a variety of opinions as to what is stable, but really a mathematical probability of how long between crashes. And you can adjust the probability of variance of time between errors - but in theory, even at the "best" settings, that could be years or, even if extremely improbable, 5 minutes.

Exactly.

In the professional realm we don't speak of "stable" versus "unstable", we speak of probabilities and "reliability".

Reliability is defined as the probability that a device will perform its intended function during a specified period of time under stated conditions. Mathematically, this may be expressed as,

,where

is the failure probability density function and

is the length of the period of time (which is assumed to start from time zero).

The connection between reality (reliability and probabilities) and the layman's take on stability (is it stable? did it pass 50 cycles of IBT?) completely breaks down upon inspection.

What we do as enthusiasts is we OC our hardware to the point that the reliability is so bad that failure is assured within an extremely short timeframe such that we detect it before we get bored and move on.

And when we tweak the parameters such that our perception of reliability is that the machine doesn't crash in a matter of minutes or hours (or even a day) then we unjustifiably feel confident in concluding we have restored a degree of reliability that will span months or years.

In other words, ignorance is bliss

 

[Markfw]

My distributed computing software appears to be checking for data corruption, and stops when it is detected. Does anybody have any evidence that F@H does NOT detect this correctly ? I am very curious. 4 of my 5 systems have been running it forever, and never had a bad unit. (even some of the video cards)

 

[Idontcare]

My distributed computing software appears to be checking for data corruption, and stops when it is detected. Does anybody have any evidence that F@H does NOT detect this correctly ? I am very curious. 4 of my 5 systems have been running it forever, and never had a bad unit. (even some of the video cards)

I have no doubt it is performing error-checking correctly. But do recognize it is only checking for errors that would be generated by the likely small subset of instructions that DC is using out of the entire ISA.

When AMD or Intel validate the processor for binning purposes they have to verify that each and every instruction in the ISA is operating error-free at temperatures all the way up to TJmax and set Vcc accordingly.

When we enthusiasts overclock or undervolt we are usually deciding for ourselves that we haven't undermined our reliability based on the feedback of a handful of applications which in reality are only testing a narrow subset of instructions.

It is very easy to lull ourselves into a false sense of having attained/retained reliability and system stability. Confirmation bias also clouds this situation because it is human nature to stop testing as soon as we get the result we were looking for.

On the other hand if we are taking a general-purpose processor and basically using it as a poor-man's fixed-hardware solution to run a specific application which will only use a limited set of instructions then we probably don't care if some unused portion of the ISA is completely unstable and generates corrupted results when used (because we have no intent to use it).

 

[VirtualLarry]

Didn't someone report last year or the year before, that their Core2Quad 45nm CPU, would overclock to 4Ghz fine, and pass Prime95 testing, but at anything over 3.6Ghz, their SSE4-using app would fail massively?

I wonder how many people that overclock, test their SSE4 functionality for correctness? Those instructions are fairly complex, and yet, barely any software uses them.

 

[moonbogg]

I'll leave prime 95 run on blend overnight. If I come back in the morning and its still humming along, I am a happy man. Then I start using the PC as intended (gaming) and observe its bahavior and adjust the OC if I get blue screens, lockups etc.
For instance, my rig was at 4.5 and i got blue screens like twice a week. Went to 4.3 and it fixed it.

 

[Markfw]

2600@3.6, 3930k@3.8, 920@3.6, 950@4.0 (yes I need to update my sig)

Pretty mild IMO

 

[Soulkeeper]

I've never owned a cpu that clocked higher than 3.2GHz fully stable now that I think about it.

Damn I feel slow ... all that wasted money

 

[AdamK47]

For my current system, 8+ hours of LinX 11 threads, 28GB allocated. Followed by 8+ hours of Prime95 with 8192 problem size and 10GB allocated. All is fine and dandy like wine and candy.

 

[AtenRa]

For my system, I consider a stable OC when it runs all my apps, games etc without crashes, BSoDs etc.

 

[JimmiG]

This all got me thinking – What would others consider a “stable” build? What do you do to test your computer stability?

48 hours of Prime 95, 48 hours of OCCT, 48 hours of AMD Overdrive stress test (in case of AMD system). I make sure to start the next test before stopping the last so it doesn't get a chance to cool down.

Also I keep an eye out for general stability issues and investigate every crash to make sure it wasn't caused by the OC, especially during the first weeks.

 

[VirtualLarry]

Also I keep an eye out for general stability issues and investigate every crash to make sure it wasn't caused by the OC, especially during the first weeks.

See, to me, that doesn't make any sense. How can you rule out the OC being the problem, if you have an occasional crash?

 

[Maximilian]

I will not settle for anything less than 3 years prime stable!! Once it has passed this i then declare the CPU stable and throw it in the trash so i can upgrade to the latest model and start the process all over again.

Minesweeper bluescreened on me in the early 90's and wrecked my time so now I make sure my systems are 100% stable! :thumbsup:

 

[VirtualLarry]

I will not settle for anything less than 3 years prime stable!! Once it has passed this i then declare the CPU stable and throw it in the trash so i can upgrade to the latest model and start the process all over again.

Minesweeper bluescreened on me in the early 90's and wrecked my time so now I make sure my systems are 100% stable! :thumbsup:

There is some truth to that.

I started a thread some time ago in this forum, regarding my E2140 CPUs, that I had clocked up to 3.2. I was having instability issues with spontaneous reboots, at around the two week to a month mark, of running the rig 24/7. I finally decided to downclock that rig to 2.8, and the problems went away. It must have been the CPU chip itself that had slight issues with 400FSB, because those same motherboards ran a pair of Q9300 CPUs at 400FSB for a couple of years without issue. (Still installed and OCed.)

 

[max347]

windows logo

 

[Yuriman]

I run a game server in the background and thus can't afford for my computer to reboot or crash when I'm away. If I remember correctly, the last time I restarted my computer was in October when I was messing with voltages for IDC's delidding thread. Windows update claims 10/2/12 was the last time updates were installed.

When I was first stability testing my overclock I discovered quickly that a variety of programs needed to be used to ensure "stability". I found a voltage where I was able to run Prime for 48 hours and IBT for 12 with no issues, but opening up Dolphin, the gamecube emulator, resulted in an "application has stopped responding" within 5 minutes. Increasing my voltage by +15mv made this go away. After that I still had a few programs that would act up once in a while, such as windows gadgets crashing, so I gave it another +10mv and I haven't seen POST since.

 

[2is]

If Internet Explorer doesn't crash, I consider my system stable...

That may sound like a joke, and it KIND OF is, but my experience has been that I can pass all sorts of stress tests for hours at a time, but still get random "Internet Explorer has stopped responding" messages from time to time. Raising voltage or lowering OC makes them disappear. Odd I know, but true story.

First thing I do is test my temps. I find the highest OC I can (at whatever voltage is required to get it) while maintaining decent temps while running 10 passes of IBT. For me this is under 90C on Ivy Bridge CPU. Once I find this point, I know I will not be touching voltage (either up or down) at all, which gives me one less factor to worry about.

I then run the usual suspects of stress tests, usually overnight. If it fails, I lower my clock speed until I can pass all the tests. Once I pass all the tests I reduce clock speed an additional 100MHz for an added stability cushion.