Guidelines for Thorough Stability Testing

[MrSeanKon]

Thank you again felinusz for your comments (OK my friend _dangtx_ helped not me).
justthinking OCCT monitor like any similar program is not accurate.
Anyway here at XtremeSystems is a thread for this topic.

To continue Hot CPU Tester Pro is ready for Intel HT and dual core CPUs but you can run two instances if you want.

 

[monkeee2002]

Awesome thread! There's some super info on these pages.
I'll be building my first OC~able PC right about Xmas, and I think I'll be studying this thread till then.
Kudos to all contributors!!

 

[MrSeanKon]

Continued...

Q. Do you trust MSDOS MemTest?
A. This is program is a good tool for determining whether your RAM is bad but does not guarantee 100% memory stability in Windows.
In my opinion a rough formula is Windows RAM frequency = MSDOS MemTest MSF - 5 MHz FSB for RAM stable in Windows.

This is taken from my O/c guide in Guidemania.
I mean that the maximum stable frequency for RAM running MSDOS MemTest is +5 MHz higher than running a Windows based stability program (SuperPi 32M, Prime95 Blends FFT) etc....
Don't forget that Intel CPUs have poor FPU performance --> SuperPi 32M is not sensitive for Intel CPUs.
But SuperPi 32M stretches CPU+RAM as we know and for A64s stretches the I.M.C. also.
Therefore if your A64 IMC is over (or near to its limits) then the above formula is false.
More details for this can be in advanced o/c guide of Guidemania.

 

[Hack30]

i would assume that 2 instances would be necessary for pentium D's as well not just HT procecssors

 

[Fuegodeth]

Great Thread - and a question

Hi all,
very interesting thread. I'm rather new to overclocking, so it's nice to know how to tell if what I've done is actually working properly.

I've begun overclocking my athlon xp 1900+. I seem to have reached the max my processor will run at, no matter what I do to the voltage. It is not unlocked, so it's just a FSB overclock. It's at 144, up from 133. Passed 25 hrs of prime95 on priority 10. Memtest has been running for 27hrs now. I got one error in test #4 on pass 43, at around the 19 hour mark. I'm not quite sure how to interpret that. I'm leaving it running to see how long it takes to get another one. I'm probably going to keep the settings as is because I don't have any really critical data on that machine. I am curious if this is normally the case with memtest. Do people get 1 error out of 24 hours or are there normaly a high number of them when something is wrong? Anyway, thanks for all the wonderful information.

On another note, I already know that 3Dmark does not like my system. I had a radeon8500 in there, and running 3dmark with everything stock resulted in huge triangles flickering on most of the tests. I tried alot of stuff, and eventually got a new video card. Because this isn't my best system, I put the new card in that one, and then moved the radeon9200 over to this one. 3Dmark ran fine on the 9200 on it's original system, but does the same thing as the 8500 did when in this mobo. Anybody got any ideas how to tackle that problem? Fastwrites is off.

I'm new to this forum, so I haven't done a sig yet. My system info is listed below.

Athlon XP 1900+
OCZ goliath All copper heat sink with 80MM fan mounted via adaptor.
Lian Li aluminum case with 4 fans
Enermax 450W PSU
Iwill XP333-R Rev 2 12/20/2002 Bios version
1x512MB Mushkin PC2700 overvolted to 2.7 - max by jumper
maxtor 80GB ATA133 HDD
CD Burner, NIC, floppy

I've got fastwrites off, and AGP driving strength at auto. AGP aperture is set at 128MB. Tried most variations on settings and still had the 3d mark issues on both vid cards on all settings. If anyone knows a cure for this let me know. By the way, it plays games fine, and does not artifact in any other program.

Thanks.

 

[felinusz]

I am curious if this is normally the case with memtest. Do people get 1 error out of 24 hours or are there normaly a high number of them when something is wrong? Anyway, thanks for all the wonderful information.

Typically, instability in memtest86 will manifest itself in the form of multiple errors. Usually in tests #5 or #8. A single error in Test #4 while peculiar, is not cause for serious alarm .

However, it does indicate that something went wrong... it could be the IMC of your processor (or in your AXP's case your motherboard northbridge), a temperature fluctuation... any number of varables. I would treat it by adding a fan over your memory - if you haven't already done so it should cool things down a bit and hopefully get rid of that (although I somehow doubt that that single error will be repeateably measureable - should you run memtest86 for another 24 hours I doubt you will see that error again ).




Your issue with 3DMark is indeed a peculiar one. Looking at your system, and knowing that your system plays games just fine, I would be inclined to say that your motherboard is likely the source of the problem.

It is an older board, and from a little-known manufacturer. Your motherboard is also based on the "rare" ALi Magik 1 revision CO chipset, not a common or popular chipset by any means (I didn't even know that it existed as a VIA alternative untill I did some quick background reading on your specific motherboard! )!

I think your issues with 3DMark are likely just poor software design. I suspect that 3DMark has not been properly "optimized" for the ALi Magik 1 CO revision chipset.

 

[Fuegodeth]

USB/PCI/AGP Stability

Hi all,
Had another question. Are there any tools for testing the stability of PCI/AGP/USB devices, and the HDD? I know that newer boards have ability to fix the PCI and AGP bus speeds, so that's not really necessary, but mine is an older board. I have an FSB overclock going, and PCI and AGP are running at higher speeds than stock. everything seems fine, but was curious if there is a way to test.

As a follow up to my post above, I ran memtest for an additional 24hrs with no more errors. I also did bit fade test, and no problems.

Now, if I can just get 3d mark to work propertly...

It actually seems to like the higher agp bus speed, as the triangles don't flicker any more. They are just there.

 

[soulrider4ever]

^^ Agree with what he says

"Definitely sticky material."

 

[Gnufsh]

I think SiSoft Sandra has a hdd performance test that you could loop for some stability testing. As for your error, there are sometimes random isolated errors caused by cosmic radiation or some such. It doesn't happen very often, but it does happen (one of the advantages of using ecc ram is that it should correct most of these, which is part of the reason you see it a lot in servers that have to be on all the time).

 

[splat]

i just downloaded version 2414 of prime95 and I can not get it to install a second instance. everytime i start the installer it only asks me if i want to unintstall. if i hit no, it closes...i never get to where i can select a new folder. how do I install a second instance?

** edit: sorry, i just found a post saying you install once, then copy the folder. then you set the affinity in both instances, one to 0 and one to 1, in case anyone else is clueless like me

 

[Gnufsh]

Or you can download the one without the installer, which just unzips...

 

[MrSeanKon]

System sometimes warns you....

Sporadically you understand that something is going wrong...
For example I encountered a strange phenomenon with Task Manager (Intel CeleronD based system).
It is described in Guidemania's stability part, therefore I don't want to repeat here the same stuff here.
Yesterday I was playing with RAM modules just to find some MSFs.
LOL before crashing ASUS PcProbe displayed an error message!
As you see in the photo LOL ASUS notify icon was visible!!
Similarly during programming (Microsoft Visual Studio) Windows Kernel shows (not always) some mystery messages (Java or C# handling errors)..
Nevertheless an unstable system won't warn you before final crash

 

[RollingThunder]

What 3D Mark is Now the Standard?

With all of the older versions available, it's not clear which one is the general standard to use. I see many sigs using different versions. I have a basic video card and am not a gamer so video testing is not a high priority.

Any suggestions?

 

[splat]

there is no "standard" 3dmark. 06 is the newest, i think it is simply a small revision over 05 to add some new dx9 features like HDR lighting. 05 is still dx9, and 04 is dx8, iirc. some people still drop back to 03 so they can post big numbers. personally, i like to use 04 and 05 so i can compare numbers to my old systems so I can see exactly how much each of my builds gains over my previous one.

 

[RollingThunder]

there is no "standard" 3dmark. 06 is the newest, i think it is simply a small revision over 05 to add some new dx9 features like HDR lighting. 05 is still dx9, and 04 is dx8, iirc. some people still drop back to 03 so they can post big numbers. personally, i like to use 04 and 05 so i can compare numbers to my old systems so I can see exactly how much each of my builds gains over my previous one.

Splat,

Thank you. Since I have no previous late model systems as a base line to compare, and don't need in-depth video testing, would PCMark 05 be preferred over 3DMark 05?

There are no stability issues that I'm checking for, just an additional program for another confirmation. I regularly use Memtest86, SuperPi and Orthos/Prime95 for overclocking checks. RT

 

[splat]

sure. 3dmark is basically a test to see how well your rig can display games. If games don't interest you, then you can use PCMark, SiSoft Sandra, or any other benchmark you'd like to see how well your system performs. Really, the only important benchmark is whatever program you use the most. Say you use Autocad the most. There might not be a specific benchmark for autocad, but you should be able to get a feel for the differences between 2 separate computers using the same version of autocad, and whatever one feels "better" to you is all that matters.

 

[MrSeanKon]

I am back folks.
Someone may ask How long should I run these programs??
PcPer poll and here is the OCAU poll
My surveys!

felinusz although in your first posts you mention links of some stability programs my TechPowerUp! thread here includes a larger list.
Note that exists variations of Prime95 (Orthos, StressPrime2004) and one version for quad core CPUs.
At this moment of speaking cos in computing anything changes rapidly.

 

[felinusz]

Note that exists variations of Prime95 (Orthos, StressPrime2004) and one version for quad core CPUs.
At this moment of speaking cos in computing anything changes rapidly.

Very true. I will update the first post with these variants.




EDIT: First post has been updated. Orthos in particular seems to be a pretty nice program.

It's really nice that someone polished up StressCPU and created a 'finished product' stress tester based on the Gromacs core... I imagine that this program is a highly effective stress tester. It is certainly user friendly anyways.

I also fixed all the broken links, and generally updated the guide. Removed (reworded) the focus on 3DMark2001, as this program is now obsolete, and totally inadequate for testing video cards based on newer versions of DirectX.

 

[Super Nade]

Ian, can you add CPUBurn to the list? Using it with error-checking turned off gets the CPU about 2-3 C hotter than with EC enabled. For dual cores, you couls set it to run on one core whilst having 3DM run on the other. That would stress both the GPU and CPU maximizing power draw.

Also, I'd like to emphasize the need for a good PSU before embarking on stress testing. Many a time a mediocre PSU's problems can be erronously attributed to a failing system.

 

[felinusz]

Hey Prashant, long time no see! How goes the green life ? I've unfortunately not been particularly active recently - overclocking and computers as a hobby have both become too expensive a venture for me, expensive in terms of both time and money (mostly time, school is my priority these days...) .



I updated the guide again. Ended up making more than a few changes... it seems that every time there is a suggestion and I come back to this, I add more details that were originally left out.

I added some information on PSU integrity, testing PSU voltage rails (didn't explain "how" to accomplish that, but I think that can be easily found elsewere ). Added some information on motherboard integrity, how add-ons can mess things up, why it is important to narrow down sources of error, removed (reworded) the blatent and (in my current opinion) erroneously justified preference for Prime95 3DMark and Memtest in lieu of stating that "one each of a CPU, GPU, and RAM specific testing application is neccessary". Also fixed some wording to make it clearer.




If anyone has any further suggestions on either content or format, please feel free to post them - I still check in frequently, and will happily make further alterations to these guidelines as is neccessary!