Linpack - Stability?

[deachus]

I have been stress testing my setup with Prime95x64, OCCT and SuperPI32M with various settings for the past 3 months trying to find my max stable 24/7 overclock with the first system in my sig below.

I started using Linpack yesterday since it seems to stress your system the most. I have passed Linpack with 20 runs @ 466FSB with all Residual (norm) values the same after upping the vcore a bit so that means my system is stable.

I then close out of Linpack and open SuperPI32M but it errors out right away. The same with Prime95.

Am I to assume that I must pass Prime95, SuperPI32M and Linpack to be considered 24/7 stable? This is driving me nuts because I change settings to be stable in one program but come to find out I am no longer stable in another program.

I guess what I am looking for is an explanation why one program is stable and one is not? Should I assume I am 24/7 stable if I can pass Linpack 20 runs with no errors?

[WaterFlex]

You must run this test to be sure your PC is full stable. Windows MemTest 3.7 is the most stress DRAM test.
I used to pass IntelBurnTest with no errors after 200 runs. The same situation was for OCCT, Prime95. But I had 3-4 errors with Windows Memtest after 100% coverage. After that, I had to encrease Vdimm +0,06 to make my Dimms fully stable.

[Jor3lBR]

I would say if your Pc passed 20 IBT + 3DMark06 + Memtest there's no way you get a crash or bsod under normal use. Don't focus too much on trying to find errors because it will happen. Even a 3.3v undercharged Cmos battery can cause your Pc to give Prime or Ibt errors...

[deachus]

I appreciate the input. I will try the Windows Memtest and see what results I get tonight after work.

I always run DOS memtest first before booting into windows with a new overclock and if I get 5 Passes no errors then I boot into windows and startin running Prime95 + SuperPI 32M + 3DMARK06 + IBT.

Is the Windows Memtest any different from the DOS MemTest?

[carepolice]

Is the Windows Memtest any different from the DOS MemTest?

Yes, more pressure for rams, usually DOS MemTest passes test far more easily than Windows based one.

And on topic, i like these linpack tests but usually i dont run full ram test just 200mb and 10x test just see that is my computer even a little bit stable.

Then Prime95 overnight + Fallout 3 to see really stability

[deachus]

I'm just not in agreement with the results Prime95 is putting out. I can play Far Cry 2 for 4-6 hours at my current overclock with no overheating (53c Load) crashes, freezes BSODS, etc. I can run 3DMARK06 with no crashes, freezes, etc. I can encode an .AVI file to DVD (encoding is the main use for my PC) with no problems and is pretty darn fast. I can run Linpack 20 runs using max memory with no errors.

But as soon as I run Prime95 I get a rounding error on 1 core after about 40-50 minutes.

Now, am I to assume my system is stable because I can run everything I use my computer for with no errors or crashes? Or do I still have tweaking to do since Prime95 is failing after 40-50 minutes?

I will run the Windows based Memtest program as suggested above and see what results I get.

[slim142]

This is what I do

Memtest86+ 2.01 Test#5 at least 10 times
Boot windows and run LinX (IBT from Dualist) full RAM 100 times
Memtest 3.8 at least 100% (run as many as you need depending on how many physical cores you have)
Then Prime95 for a couple of hours

If everything goes fine, you should be stable

[WaterFlex]

Absolutely right

[deachus]

This is what I do

Memtest86+ 2.01 Test#5 at least 10 times
Boot windows and run LinX (IBT from Dualist) full RAM 100 times
Memtest 3.8 at least 100% (run as many as you need depending on how many physical cores you have)
Then Prime95 for a couple of hours

If everything goes fine, you should be stable

I do everything you have stated except for the second to last step which is run Memtest 3.8 in Windows before running Prime95. I did not know that you needed to run a Windows based version of Memtest in Windows even after you have passed TEST#5 in Memtest86+ 10 times. I will add this to my stability process and report back in a few hours on my results.

I appreciate everyone's input! I am not knew to overclocking and have been doing it for many many years but I never seemed to have such a variance in stability from one program to another when stress testing. You learn something knew every day!

[slim142]

I do everything you have stated except for the second to last step which is run Memtest 3.8 in Windows before running Prime95. I did not know that you needed to run a Windows based version of Memtest in Windows even after you have passed TEST#5 in Memtest86+ 10 times. I will add this to my stability process and report back in a few hours on my results.

I appreciate everyone's input! I am not knew to overclocking and have been doing it for many many years but I never seemed to have such a variance in stability from one program to another when stress testing. You learn something knew every day!

Yeah, I understand what you say.
I used to run prime95 for 8 hours to say Im fully stable, but with this new utility (LinX 0.5.1) everything was simplier in just 3 hours. Generally, after LinX runs with no problems, is 99% guaranteed that Im stable, but I like to make sure

[CERO]

You must run this test to be sure your PC is full stable. Windows MemTest 3.7 is the most stress DRAM test.
I used to pass IntelBurnTest with no errors after 200 runs. The same situation was for OCCT, Prime95. But I had 3-4 errors with Windows Memtest after 100% coverage. After that, I had to encrease Vdimm +0,06 to make my Dimms fully stable.

QFT, same for me, HCI memtest is the best test and find errors way faster, usually within the first 0-200% coverage

[deachus]

I ran 1 hour of HCI Windows Memtest when I got home from work tonight at same settings that I passed 20 runs of Linpack last night with all Residual (Norm) numbers matching.

I then ran Prime95 and I'm getting a rounding error still on Core 3 after 40 min. Is Prime95 incorrectly reporting an error even though I passed 1 hour of HCI Windows Memtest and 20 runs of Linpack?

I will be running HCI Memtest overnight tonight when I go to bed and we will see what the results are in the morning. As far as I'm concerned I am stable but I could be wrong.

[SoulsCollective]

If you're using Vista, there's a known bug with SuperPI - http://www.xtremesystems.org/Forums/...41&postcount=7

[Dua|ist]

deachus, first of all, if you want to have your rig rock stable you'll have to pass all tests as there is no absolute and perfect stability checking program. From my experience successful (and long enough) testing with Prime/Orthos, Linpack and Memtest is usually enough to prove stability. Still, neither of these programs does test Northbridge well enough, so some failures in 3d (not necessarily 3dmark though) may occur due to NB instability even after successfull Prime-Linpack-Memtest runs.

As for Linpack, 20 tests just isn't enough in most cases. Why do people test their rigs with Prime for 5, 8 or even more hours but keep thinking that 20 Linpack runs mean stable system (when it takes less than half an hour)? Linpack is really great in determining CPU instability, but you should consider running it at least 50 times at maximum available memory if you want to call it stable.

What's more important, Linpack is sensitive to vcore only, wrong vtt/gtl settings and/or slightly unstable RAM won't cause errors in Linpack (at least during 20 runs). But will most likely cause them in Prime. If Memtest and Linpack pass ok, but Prime still fails, I'd try to tweak vtt & gtl settings.

[davidk21770]

I agree with Dualist.
All of the tests stress your computer in different ways. Unless you pass 'em all, you have some instability. Whether you agree with 'em or not -- it's there.
How much risk you live with is up to you, but you've described a situation that seems guaranteed to lead to having to reinstall windows and/or other programs when bad information is written to the disk. Been there, had to do that -- even with systems that were 4 hour prime stable. (Why I go for 12 hours now).

[deachus]

I ran LinX 50 runs last night and got an errors after 12minutes. I changed my Vtt and GTL settings and ran LinX again 50 runs and passed. I then ran HCI Memtest overnight (11 hours) and no errors when I woke up this morning.

I will run Prime95 again tonight and see if it is still giving errors in the 40-50 minute range.

I appreciate everyones input and help!

[Praz]

What's more important, Linpack is sensitive to vcore only, wrong vtt/gtl settings and/or slightly unstable RAM won't cause errors in Linpack (at least during 20 runs). But will most likely cause them in Prime. If Memtest and Linpack pass ok, but Prime still fails, I'd try to tweak vtt & gtl settings.

Partially true. Linpack works well for quick, course adjustments of VCORE, VTT and NB voltages. Really needs to be used with at least 4GB of memory and run as 64 Bit though. Nothing beats Prime for GTL adjustment as one can see the individual core that is failing. The single best program for stability is 3 - 4 loops of PCMark Vantage. The downside is looping it requires a Pro license I believe.

[deachus]

Well I have broken the 40min Prime95 wall with no errors. I have been running for 2 hours now and will let it keep running till tomorrow morning. Here are some shots of LinX 50 runs, HCI Memtest for 10 hours and Prime95 after 2 hours.

[mikeyakame]

You'll find prime95 errors after 15-20 mins are almost always gtl reference voltage related. Making fine adjustments to CPU GTLREF or NB GTLREF, and for more than minor adjustments vTT usually sorts it out. In my experience this has always been the case.

Personally If I can pass 2-3 hours of Prime95 blend without errors on any core I consider the system stable enough for my day to day use. If any instabilities come up I deal with them as they do.

My routine is Memtest86+ 1 full pass, and 10 passes of test 5 before I boot to OS, then I'll run a set of Linpack 64bit tests ( 6 x 4000x4008 size, 4 x 14200x14208 size ), if they pass then i'll run Prime95 for 2-3 hours, if that passes then I just hammer the system with 3DMark06/Vantage/Crysis/etc benchmarks that load the cpu / gpu simultaneously. After this I just keep an eye on Windows Event Viewer for a few days and look out for any random errors, warnings, counter corruption, etc that can only be explained by instability and data errors.

Edit: Fallout 3 is probably a bad game to use for testing cause it's buggy as all hell, It's not even stable on my stable system! It randomly crashes and freezes and I need to kill it with task manager.

[deachus]

Well I believe the minor Vtt change I made when running LinX 50 Monday night fixed all errors in every stress program I have thrown at my system. Per the screenshots I posted above I have passed 50 runs of LinX, 10 hours of HCI Memtest and 15hours of Prime95. I woke up this morning and Prime95 was still running error free.

I could not get my 4x1GB G.Skill D9GTS stable above 450FSB and I believe this is due to running 4 sticks instead of 2 sticks in the black slots. I ordered a Mushkin 2x2GB DDR3 1600Mhz kit that I have confirmed have Micron D9GTS ics. They should be here any day and will start the testing all over again once they arrive to try and get them stable at my current overclock.

[SoulsCollective]

I was in a similar situation to you with my P5E3 Premium, and I could not get four sticks working at 1,600 at all. Same as you, I switched to 2*2GB.

[Alatar1313]

What does it mean if I see excessively low gflops in ibt? With most of my stable runs, it's at around 21gflops. Sometimes, it's a lot lower at like 10-15. The residuals are still matching but aren't gflops a measure of cpu performance? Does this actually mean anything or should I just pay attention to the residuals?

[deachus]

Please provide your setup:

CPU
Motherboard
RAM
VGA
PSU
Cooling
Your Overclocking

Your numbers seem extremely low but you need to give more information. I am getting between 52-54 GFLOPS with my overclocked Q9450. I can say that when I get a bad overclock the GFLOPS drop dramatically then LinX reports an error.

[Alatar1313]

e5200
GA-EP35-PS3L
2x2GB G.Skill DDR2 800
8800GTS 512
Thermaltake 600W
Xigmatek HDT-S1283

I have it stable at 22GFlops at 3.75Ghz (313x12 or 358x10.5 both work) with 1.4375V. I'm pretty sure that the lower GFlops earlier just needed to run longer. In 100 tests, I got several failures. I'm pretty sure I'm stuck at 3.75Ghz without raising the voltage above 1.45V in the bios. Kinda makes me wish I didn't buy this hsf 'cause I had 3.5Ghz on the stock hsf. Is 22GFlops low for an e5200?

[deachus]

That sounds about right but I am not too familiar with E5200. I am running a QUAD @ 3.728Mhz and I get more than double that.