PDA

View Full Version : What do you guys do for stability testing?


Zelda_fan
02-20-07, 02:54 PM
In my opinion SuperPI is not enough. I've ran into plenty of cases where I could run a 32M calculation of SuperPI, yet fail at other tests.

In my opinion Orthos (or multiple instances of Prime95 - 1 instance per core) is the best way of stability testing around. It creates a very stressful and intensive environment of 100% CPU and constant pagings to all of the RAM and if there is ANY instability in my overclocks it will bring it out.

In my opinion I don't consider a build stable until it can pass 12h of an Orthos Blend test. I personally won't run an overclock 24/7 unless it can pass 24h of Orthos.

My point is this. If nVNews is to start a serious overclocking community, we need to establish standards for stability testing.

J-Mag
02-20-07, 03:06 PM
Well if you tests for stability using, prime, orthos, superpi, or whateve ryou obviously aren't testing your GPU subsystem.

Personally I use whatever game I am into at the time for testing stability, because they will stress CPU, MOBO, MEM ,and GPU.

Xion X2
02-20-07, 03:23 PM
I would never use gaming software for stability testing, because if you have an unstable overclock it'll corrupt windows and any applications you're currently running on it at the time it crashes.

And you're right, Zelda. SuperPi does absolutely nothing for stability. It's also a single-core application. I was able to run SuperPi @ 3.85gHz @ 1.7v and afterward failed Orthos within seconds.

Dual-prime or Orthos small FFT's is the way to go. Can also use TAT. Measure temps w/ CoreTemp or TAT.

|MaguS|
02-20-07, 03:24 PM
I hip bump my PC, if it doesn't tip over its stable...

Zelda_fan
02-20-07, 03:27 PM
The problem with using games for stress testing is that they provide no feedback or checking on the calculations they preform.

For example. Say Orthos performs the calculation 1 / 2 which as we know equals 0.5. On an unstable system, the CPU might return a value such as 0.4 once in a while. Orthos tests all of it's calculations against the known correct values whereas a game does not and might keep right on running. You can read more about it here.

http://en.wikipedia.org/wiki/Prime95

I do see your point about the GPU benchmark, and I think the best solution is to run an instance of 3DMark and Orthos at the same time.

Xion X2
02-20-07, 03:28 PM
^^ lol

I play Vanguard. If it crashes to the desktop, hell, it's not stable! ;)

Zelda_fan
02-20-07, 03:28 PM
Dual-prime or Orthos small FFT's is the way to go. Can also use TAT. Measure temps w/ CoreTemp or TAT.

I prefer blend. Small FFTs don't stress the memory nearly enough.

Lfctony
02-20-07, 03:31 PM
I run Win Launcher XP. I assign the first core to 3DM06 and I use the second core on Sp2004. I also run Winamp in the backround and leave it on all night for testing. I then swap the 2 cores around. But now I have a problem in Vista since it likes to turn off the monitor despite the running on a 3d application. :( I'll probably end up disabling it just for testing and then enable the power saving option again.

Or I'll just stress test in WinXP. :)

Xion X2
02-20-07, 03:31 PM
I prefer blend. Small FFTs don't stress the memory nearly enough.

Neither does blend, really.

If you really want to test your RAM you can run Windows Memtest instead. This heats up your RAM more than anything else. Just keep an eye on how hot it gets, especially since you're going H20. I have to use some fans over my memory when I run this bench or it crashes.

Zelda_fan
02-20-07, 03:40 PM
Neither does blend, really.

If you really want to test your RAM you can run Windows Memtest instead. This heats up your RAM more than anything else. Just keep an eye on how hot it gets, especially since you're going H20. I have to use some fans over my memory when I run this bench or it crashes.

I disagree. I've encountered several situations where I'll pass memtest, but fail a blend test after 45 minutes. But I will take what you say into consideration and integrate memtest into my stability routine.

-EDIT

and in the above situations the only thing that I've changed is the memory speed/vMem/timings etc. so it has to be the memory not the CPU.

J-Mag
02-20-07, 03:40 PM
I would never use gaming software for stability testing, because if you have an unstable overclock it'll corrupt windows and any applications you're currently running on it at the time it crashes.


That's not necessarily true. A crash doesn't always corrupt your install (I would say it rarely corrupts an installation, unless you crash when files are being written). Many times people think an overclocked setting corrupts their install, but thats just because when they reboot the overclock is already unstable. Loading an OS when unstable can look like a corrupt install.

Xion X2
02-20-07, 03:44 PM
That's not necessarily true. A crash doesn't always corrupt your install (I would say it rarely corrupts an installation, unless you crash when files are being written). Many times people think an overclocked setting corrupts their install, but thats just because when they reboot the overclock is already unstable. Loading an OS when unstable can look like a corrupt install.

I never said it did it every time, but it does do it often. I had it happen very recently to me w/ several games I tried to play on an unstable overclock. One of them I even had to reinstall Windows over for because it corrupted my registry or some deep hidden file that I couldn't get to--even with a windows repair.

To each his own. I'm just telling guys what I've seen from my POV.

Xion X2
02-20-07, 03:48 PM
I disagree. I've encountered several situations where I'll pass memtest, but fail a blend test after 45 minutes. But I will take what you say into consideration and integrate memtest into my stability routine.

-EDIT

and in the above situations the only thing that I've changed is the memory speed/vMem/timings etc. so it has to be the memory not the CPU.

To each his own. I've had the exact opposite experience.

Run blend on your system and you'll notice you can still do plenty of web browsing, windows browsing, and anything else you like w/ ease. The reason is because it doesn't use the full capacity of your memory, whereas you can set it to maximum capacity on Memtest aside from what your OS is taking up.

If you don't believe me, run Memtest at your full memory capacity and try to do some web browsing or windows browsing. It moves at a snail's pace compared to blend.

[Edit] For the record, to test my 24/7 overclock I run two instances of blend for 6+ hours, two instances of small FFT's for 6+ hours, and two sessions of memtest for 3+ hours (all subsequently, obviously--not at the same time). I find this to give me the best all-around stability testing and haven't had any issues w/ my system since then.

J-Mag
02-20-07, 03:56 PM
I never said it did it every time, but it does do it often.


Well you insinuated every time because you said "if you have an unstable overclock it'll corrupt windows"

If states aren't quantum :p


I had it happen very recently to me w/ several games I tried to play on an unstable overclock. One of them I even had to reinstall Windows over for because it corrupted my registry or some deep hidden file that I couldn't get to--even with a windows repair.

To each his own. I'm just telling guys what I've seen from my POV.

Yeah I agree I have had it happen too, but just not that often. Admittedly I haven't been playing around with overclock settings in the recent months.

Xion X2
02-20-07, 04:06 PM
Come on, JMag, some things are self-explanatory. Do you honestly think that I believe for one second that an unstable overclock will corrupt a game every time it crashes? I'm not that gullible.

The game that I had this happen to me was Vanguard. And although that game is buggy as all hell right now, I did happen to have an unstable overclock at the time it happened. It corrupted the game so bad that even a Windows repair wouldn't fix it. I would log into my account and the screen would go black every time.

Since stabilizing my overclock, it has yet to happen again for over 50+ hours.

J-Mag
02-20-07, 04:18 PM
The game that I had this happen to me was Vanguard. And although that game is buggy as all hell right now, I did happen to have an unstable overclock at the time it happened. It corrupted the game so bad that even a Windows repair wouldn't fix it. I would log into my account and the screen would go black every time.


Some games are more prone to exhibit errors when overclocked...

When I first got BF2 I must have crashed 20-50 times in a few months dorking around with overclocks. I had to reinstall the game a couple times, but never the OS.

SlieTheSecond
02-20-07, 04:29 PM
Orthos with large ffts for cpu
Memtest86 for ram.
3D Mark and ingame benchmarks for gpu
Artifacts scanner for gpu/ram too.
Plus I'll do superpi and cpumark.

Stability is one thing. But you can increase your clock speeds and not see any performance gain too. Then you know somethings is funky or too hot.

I can't fathom how anyone could use SuperPI as a stability test.

Xion X2
02-20-07, 04:42 PM
Some games are more prone to exhibit errors when overclocked...

When I first got BF2 I must have crashed 20-50 times in a few months dorking around with overclocks. I had to reinstall the game a couple times, but never the OS.

Yeah, I'm sure that's the case. It's just a whole lot easier in the end to reinstall Prime95 or Orthos or Memtest if it crashes and gets corrupted in comparison to a game. You're talking a few seconds compared to several minutes, and in the case of Vanguard for me, hours since it's not only 16GB but has a slew of patches that have to be installed once you log into your online account.

In many cases, when reinstalling a game, you can also lose your save progress. I'd just rather do it the other way around.

Blacklash
02-20-07, 04:46 PM
I consider 8+ hours of Orthos blend stable.

J-Mag
02-20-07, 04:54 PM
Yeah, I'm sure that's the case. It's just a whole lot easier in the end to reinstall Prime95 or Orthos or Memtest if it crashes and gets corrupted in comparison to a game. You're talking a few seconds compared to several minutes, and in the case of Vanguard for me, hours since it's not only 16GB but has a slew of patches that have to be installed once you log into your online account.

You are especially biased against using games to test because of your current experience. I would be to if i had to download 16gb of crap everytime...


In many cases, when reinstalling a game, you can also lose your save progress. I'd just rather do it the other way around.

Well copy the save files before you re-install. Anyway, when I use an actual game to test overall stability it is usually timedemo loops or something similar and then you aren't really advancing in the game....

Xion X2
02-20-07, 04:57 PM
You are especially biased against using games to test because of your current experience. I would be to if i had to download 16gb of crap everytime...

Not just current. Repeated. It's happened on more than Vanguard and it's happened in the past, as well.

Well copy the save files before you re-install.

Easier said than done. They're often not easy to find or are embedded in other game files so that you can't see them.

J-Mag
02-20-07, 05:24 PM
Easier said than done. They're often not easy to find or are embedded in other game files so that you can't see them.

Maybe so, but I have never encountered any save games that didn't take more than a google search to find the location.

On a side note: Ever use 3dmark for stability testing? Does it tend to get corrupted easily? (I never bother because I don't have the full version and stability testing 3d under your normal res/settings is pointless IMO)

vandalous
02-20-07, 05:57 PM
For CPU (Kentsfield) stability, I run 4 instances of SuperPi @ 32M. If all 4 pass, I run 4 instances of Prime95 Torture Test for minimum 8 hours (preferably overnight). I also keep the Nvidia Monitor program up to watch core temps. If I can pass all of the above and keep my temp at or below 65 under load, I feel my CPU OC is stable.

For memory, I run memtest86+ with 5 passes on test 5 followed by 2 full tests. This program and BIOS updates are probably the only things I still use floppies for.

After I feel my cpu and memory are stable, I'll run 3DMark06 at a constant loop for 4-8 hours. The full version allows you to loop and specify certain tests and whatnot. If it passes that, I believe my rig is stable. If I've got a lot of time on my hands (which is rare now that I've got 5 projects at work that are battling for all my time), I'll run some looping timedemos in Q3/Q4 or run additional stability tests in SiSoft Sandra.