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

[TylerS]

I’m by no means and expert in this area but I have been doing a lot of reading lately on stability testing in my efforts to squeeze every last bit of juice out on an aging system.

When I first built my computer (2008) I was pretty lax with my stress testing. I ran Prime95 Blend overnight and called the computer “stable.” In reading further, it seems I should have spent some time with other programs and at least run memtest86+. On the other hand the computer has run for 5 years without any day-to-day stability complaints. Sure it crashed occasionally, but what windows rig doesn’t? So really, who’s to say it wasn’t “stable” by some definition?

In my current OC work I am pushing my computer much further (thanks to a new Thermaltake Water 2.0 Performer with push/pull Corsair SP120 High Performance fans). I also intend to take a more thorough look at stability. Others even seem to be of the opinion that you should just play a few high-end games and see what the results are.

I am sure everyone has seen Idontcare’s “Gold Standard” for stability on the sticky above. This seems extermely comprehensive and given the source, likely the best way to ensure the most stable rig you can. I would love to achieve this, but my efforts to hit this so far have fallen flat at the speeds I want to hit. I may decide to pick something in the middle of this gold standard and my hap-harzard approach from before.

I have also read a lot of oppinions that Prime95 SmallFFT i(plus something to test RAM) and/or Blend to test the full system s "good enough." Not to mention the many other options out there.

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

 

[Idontcare]

In general, for stability testing I just follow the guidelines posted in the sticky: Overclocking CPU/GPU/Memory Stability Testing Guidelines

Those guidelines exist based on what I know to be the downsides of having a seemingly stable system that might actually be unstable. Data corruption sucks. So the guidelines were drafted with this in mind, to ensure the seemingly stable actually stands a chance at being stable.

However, if you don't care about data corruption then there is little reason to test for system stability with any app other than the ones you are actually going to use with your OC'ed computer.

If WoW is the most challenging thing you are going to throw at your rig, and you don't care about the potential for loss of data due to silent data corruption, then WoW is your stress tester of choice. Optimize your rig's OC such that WoW functions and be done with it.

edit: see my comment in this post for example.

 

[Puppies04]

If WoW is the most challenging thing you are going to throw at your rig, and you don't care about the potential for loss of data due to silent data corruption, then WoW is your stress tester of choice. Optimize your rig's OC such that WoW functions and be done with it.

But what if my machine crashes due to data corruption just as I am about to roll on a tier 27 (or whatever they are on now) chest piece and I end up losing it to that damn hunter that doesn't even need it?

Bad times, bad times...

 

[lehtv]

I consider a stable OC one that doesn't crash or produce errors in real-life long term usage scenarios. These may vary from user to user, obviously. For me, stress testing applications and standards are there only to ensure, with a short term experiment, that the overclock is long term viable.

Also, some overclocks that I don't consider stable can still be useful. My cooling setup can't handle my CPU at 4.7GHz in stress testing applications - temperatures reach near TJmax and produce errors in longer periods of stress testing. However, in Planetside 2 the overclock works fine and temperatures are in check, and compared to 4.2GHz, I get a 5 fps improvement during CPU heavy scenarios.

 

[TylerS]

edit: see my comment in this post for example.

Yeah, this post was one of the reasons I felt I should be at least a little more dillegent with stability testing this time around. My wife would kill me if I lost any of our (hundreds) of baby or wedding pictures just so I could play Skyrim.

 

[BrightCandle]

I go further than IDC, I do all those tests but as a starting point. Over a period of weeks I run games and other applications and if it ever crashes or shows odd behaviour at all I adjust the settings until I haven't had a problem for weeks to months. One crash a month and almost certainly due to software is my preferred approach which means dialling in a stable overclock will take me many months in the end. My overclocks are much more conservative than I see in the reviews and I suspect the high level of stability I am looking for is the reason why.

 

[Sable]

I consider a stable OC one that doesn't crash or produce errors in real-life long term usage scenarios.

Exactly this.

It's all very well being able to stress test with a benchmark for 24 hours or whatever but I've still found cases where a game which isn't taxing the CPU as greatly will cause a crash.

Same for GPU overclocking. I had MSI Kombuster running for 6 hours at 100% load and thought I had a stable OC. Fire up Far Cry 3 and play for a few hours and start getting artifacts.

 

[Idontcare]

Yeah, this post was one of the reasons I felt I should be at least a little more dillegent with stability testing this time around. My wife would kill me if I lost any of our (hundreds) of baby or wedding pictures just so I could play Skyrim.

That was me, lost all the photos from our honeymoon. Thankfully I had made one copy of them for a photobook but they were all inkjet printouts, not actual photos, so they are all faded and look like crap from moisture uptake.

The real answer to that concern though is to implement an appropriate method of back-up and archiving (with backup as well). Silent data corruption can occur even with stock clocks, avoiding OC'ing does not avoid silent data corruption entirely, just lessens the likelihood of it occurring.

Extra hard-drives and a USB/eSata external docking station are silly cheap nowadays I make backups of my backups

 

[TylerS]

I make backups of my backups

The real answer to that concern though is to implement an appropriate method of back-up and archiving (with backup as well).

At least I have that covered. I have offline "cloud" backup with versioning and local back-up on multiple hard drives. I think I'll also clean up an old Mybook for a thrid offline backup.

I guess if you are careful with that, stability can be what you want it to be!

 

[tweakboy]

When real world apps work and dont crash, I know I got a stable Rig.

 

[Hulk]

One that after initial setup you never worry about, never check temps, and never crashes. My current very conservative overclock has been like that for 18 months now. That's why I don't push the last few hundred MHz. I don't enjoy the fiddling now like I did 15 years ago. I mean I love the initial set up and tinkering but after a week of finding the limits I just back it down super conservative and that's it. I could probably run my current overclock at 4.4GHz instead of 4.2GHz at the same volts but I might get a crash very few weeks, or in the summer, or under some unusual load, or whatever. Not worth my time.

 

[Soulkeeper]

for me stable is zero errors/issues 24/7
ie: if I can compile code, game, browse the web, and have uptimes of many days/weeks without a single error.

what this means for me when comparing to reports of overclocks on the internet is usually day and night. ie: P4 ~3.2GHz was my stable, c2q 3.15GHz was my stable, llano 3.2GHz is my stable

With the llano case in mind I can post at 3.8GHz, maybe even get into the os and take a ss at 3.6GHz (if i'm lucky), I can pass a handfull of benchmarks at 3.4 and maybe 3.5GHz but it's guaranteed to fail at some point (compiling code being the quickest way to cause failure, where every bit counts).

 

[Puppies04]

Yeah, this post was one of the reasons I felt I should be at least a little more dillegent with stability testing this time around. My wife would kill me if I lost any of our (hundreds) of baby or wedding pictures just so I could play Skyrim.

I see from your later comment that you have it covered but even at stock clocks with a system from dell or hp etc if you have files which are important to you (family pictures, tax returns, business data etc) and they only exist in 1 place then you are just asking for trouble.

I know this has been said many many times around here but if you are reading this and you don't back up your data in at least 3 places (not all in the same house) then you are running the risk of losing it forever.

There was a very good thread about safeguarding your data around here somewhere I will see if I can dig it up for anyone interested.

 

[Nemesis 1]

Anything that keeps the cpu cooler than 70c the lower you keep the core temp increases the chance of good overclock . With my watercooling setup direct die using silver based water block at 5.2ghz I stay under 70c 5.3 and I go over and the system becomes unstable

 

[Nemesis 1]

I see from your later comment that you have it covered but even at stock clocks with a system from dell or hp etc if you have files which are important to you (family pictures, tax returns, business data etc) and they only exist in 1 place then you are just asking for trouble.

I know this has been said many many times around here but if you are reading this and you don't back up your data in at least 3 places (not all in the same house) then you are running the risk of losing it forever.

There was a very good thread about safeguarding your data around here somewhere I will see if I can dig it up for anyone interested.

If your an O/cer . One naturally protects the data does one not, There are 1T thumps out , That would do.

 

[Idontcare]

One that after initial setup you never worry about, never check temps, and never crashes. My current very conservative overclock has been like that for 18 months now. That's why I don't push the last few hundred MHz. I don't enjoy the fiddling now like I did 15 years ago. I mean I love the initial set up and tinkering but after a week of finding the limits I just back it down super conservative and that's it. I could probably run my current overclock at 4.4GHz instead of 4.2GHz at the same volts but I might get a crash very few weeks, or in the summer, or under some unusual load, or whatever. Not worth my time.

Same here. I go for the "set it and forget it" type OC's.

That is why I do what I wrote up in the sticky, which is a perpetual work in progress of course as we all work together to find newer and better ways to identify instablity. (we can never prove our rigs are stable, we can only determine that they are not expected to become unstable for a given period of time)

I've got 5 Q6600's in my basement that have been running at 3.3GHz OC's since 2007 and I have not had a single system hang, reboot, or crash in those 5 yrs, not even once. That is what I consider to be a stable OC.

 

[VirtualLarry]

There are 1T thumps out , That would do.

I believe he is referring to the new 1TB "thumb drive" that Kingston announced recently at CES 2013.

 

[ZipSpeed]

Stable overclock for my rigs means being able to run 24/7 distributed computing projects that don't produce any invalid results.

 

[Nemesis 1]

I believe he is referring to the new 1TB "thumb drive" that Kingston announced recently at CES 2013.

yep i should have said to be released

 

[Centauri]

My definition, of late, of a stable system is one that can run AMD's OverDrive stability test suite and Prime95 simultaneously for 1 hour.

 

[Diogenes2]

Stable overclock for my rigs means being able to run 24/7 distributed computing projects that don't produce any invalid results.

This for me too..

 

[TJCS]

0. Setup a event viewer notification for Kernel-WHEA events so you can stop any further stress testing when you know it's unstable.

1. Linx(10-pass) and Prime(2 or 3 hours) to test for basic stability and overclock errors.

Also ensure no sudden drops in GLFOPS Linx
​

2. If above passes I run the little tests(Cinebench, SuperPi, wPrime, etc) to quickly check if my scores are on par.

3. Then I use Warcraft3TFT > 1 vs 11CPU(Insane) with instant-build cheat code :sneaky:

I found this method quite effective for uncovering WHEA Errors on my 3570K. I think this method does a good job stressing the CPU at various loads due to the massive battles and units constantly being produced and killed.
​

4. Next, I use HandBrake to HQ encode(10K+VBR, 2-Pass) a file that can last at least a couple of hours.

I notice that sometimes if I play a flash video in a browser or some light-cpu usage tasks during a HandBrake encode it will help uncover WHEA or HandBrake errors.
​

5. Fire up my favorite games and play for a while.

6. Run Prime95 overnight (8~12hrs)

7. Be content, use the PC, and stay alert of unusual errors!

Stable overclock for my rigs means being able to run 24/7 distributed computing projects that don't produce any invalid results.

2nd this too!

 

[PrincessFrosty]

I don't settle for any errors, I don't think there's any such notion that overclocks can be stable for some apps and not others, either the CPU is faithfully performing calculations or it's not, that's a boolean for me, not a sliding scale. All you can really say is that some apps with ligher CPU usage might be less likely to generate errors, they're still going to eventually and that shouldn't be acceptable.

Just prime95 the crap out of the CPU for extended periods of time, at least overnight, longer if you can, get one instance on each core, completely max it all out, if it lives it's stable.

I'd love to run my 2600k @ 4.9Ghz all the time but it's just slightly too unstable, I can get several hours of intensive Planetside2 gaming out of the chip which legitimately helps because that game is CPU starved, but ultimately it's going to BSOD and I just can't accept that as stable, so alas it sits at 4.7Ghz right now.

 

[Idontcare]

...either the CPU is faithfully performing calculations or it's not, that's a boolean for me, not a sliding scale.

I get where you are going with that but unfortunately it is not true, it really is boolean because cpu errors are random. The same calculation ran over and over again (what you are doing by cycling prime or LinX) will return the correct result over and over again...until that one time it doesn't.

It is a probability, a distribution, and all we are doing with stability testing is sampling the distribution. We can never conclude the cpu is faithfully performing calculations, we can only speak to the probability that it is, because eventually there will be a time when it doesn't.

To see what I mean, take your OC rig and dial down the voltage until you are on the hairy edge of stability in LinX. Then run LinX repeatedly, taking note of when (in time) it stops for an error. Then run LinX again and again.

It won't always fail at the same cycle, the same point in test time. There will be a distribution to the time-to-fail.

Increase the voltage and you see the fails go away, but not really. All you really did was push that distribution way out to the right on the graph, further out in time. Let your LinX run longer - days and weeks - and you'd start to see the crashes again, same distribution just pushed out in time.

Even at stock there is a probability of fail in the calculations, unavoidable soft-error rate.

 

[TylerS]

we can only speak to the probability that it is, because eventually there will be a time when it doesn't.

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.

All you really did was push that distribution way out to the right on the graph, further out in time.

And so here too, it becomes a preference of the user as to how willing they are to accept the occasional crash on how often. Hopefully considering the risk and protecting against the potential of data loss.

I think I would be willing to accept a degree of instability with my current computer, simply because I need to push it to the "bleeding edge" in order to get six months to a year of use out of it. It simply stopped "keeping up" and was of little use to me at my previous level of performance.

Though as I write this I am also starting to wonder if I am already seeing the signs of hardware failure as my computer is doing some very mysterious things as of late. But that is the subject for another post.