How stable is 3DMark?

[Syphon Filter]

Hi,

Quick question for you all. How stable is 3D Mark? Are issues experienced with it indicative of underlying system problems?

The reason I ask is that I've been trying to use it to test my cards/stress them. Generally speaking it runs fine but sometimes, usually on a looped run, it will crash to desktop. I think this happens more when using "custom" modes (mainly to crank all the settings to max).

Does this suggest that there's something wrong with one of my cards or that there's some other hardware related issue?

Any help would be appreciated.

 

[Nenu]

Yes, your system isnt stable.
Decrease clocks, increase voltage if temps are ok and increase cooling to improve stability.

 

[Syphon Filter]

Hmmm...

The system is at 100% stock settings, all my temps look fine and all GPU options are EVGA/nvidia default. I'm using the latest WHQL driver.

Faulty hardware then???

 

[Dreamerbydesign]

Hmmm...

The system is at 100% stock settings, all my temps look fine and all GPU options are EVGA/nvidia default. I'm using the latest WHQL driver.

Faulty hardware then???

Could be. The software is stable.

 

[Elf_Boy]

Each and every system is unique.

That said, I have found - each and everytime I've had troubles with 3d mark, an underlining hardware fault.

Ram, cpu and psu being the likely components and heat being a big possibility -- at least based on my history.

 

[Excessionoz]

Contrarywise, I've found different installations of 3DMark to be vunerable to quirky behaviour.

I was getting totally horrid figures from my SLI GTX570's, with the Physx score sitting on 1-2 frames per second (online figures were supposed to be in the 20+ per second range).

No amount of tweaking and carrying on, from my side, made any difference, the physx scores just plain sucked. I complained locally (Whirlpool forums) and was scoffed at '3DMark is the gold standard' they said.

So I uninstalled the STEAM version I'd bought, and downloaded it from Geek3D or some such, and it worked as advertised, the physx stuff ran butter smooth and my 'problem system' turned out to have been a problem with the STEAM version of 3DMark.

Just saying ... it depends which version of 3DMark you have AND where you got it from.

(Normally I'd trust Steam absolutely, but in this particular case, it was suck-ass).

 

[Syphon Filter]

This is somewhat concerning then.

From memory, I can ALWAYS complete a pass of the test run, be it at default, extreme or custom. It's the looping that causes a problem I think. My temps seem fine, I think last night the GPUs were in the 65*C region and my CPU was 50*C or thereabouts. I guess I need to do some more testing. Are there any BIOS settings I should check? I have my 16x slots running at 8x and set to Gen3.

So what are my first steps?

1. Run a memory test - what's the latest software for doing this?

2. Test of CPU stability - Again, what should I use? Aida64? I had an unspecified error after around 2hrs of this a few days ago but I Googled and it just seemed to suggest a "one of those things" type issue. The error was generic so no real indicator as to the root cause.

3. Test each GPU in isolation - is 3DMark the way forward for testing these?

4. How can I ensure the "whole" GPU is tested (i.e. all rendering paths/pipelines/shaders and all memory space)?

5. Is there anything else I need to test and if so, how should I go about it?

I'd really appreciate any help. I'm really hoping this isn't a hardware issue, RMA is such a ballache!

EDIT:

Excessionoz, now that is interesting. I am using the Steam version which I bought when it was on sale.

For what it's worth, I believe my system is "game stable". The only issue I have had is the odd BF4 crash to desktop but from what I've been reading this is quite widespread and the game is riddled with various bugs so I'm reserving judgement on this.

Other games seem to run just fine though. COD:Ghosts seems to sit quite happily at 90FPS (capped I believe). I have seen the odd flashing texture/shadow but I've seen quite a few reports of this. Metro 2033 seems fine though it does seem to tax the GPUs harder than other games. Darksiders is spot on (though it's not that demanding).

 

[Nenu]

Contrarywise, I've found different installations of 3DMark to be vunerable to quirky behaviour.

I was getting totally horrid figures from my SLI GTX570's, with the Physx score sitting on 1-2 frames per second (online figures were supposed to be in the 20+ per second range).

No amount of tweaking and carrying on, from my side, made any difference, the physx scores just plain sucked. I complained locally (Whirlpool forums) and was scoffed at '3DMark is the gold standard' they said.

So I uninstalled the STEAM version I'd bought, and downloaded it from Geek3D or some such, and it worked as advertised, the physx stuff ran butter smooth and my 'problem system' turned out to have been a problem with the STEAM version of 3DMark.

Just saying ... it depends which version of 3DMark you have AND where you got it from.

(Normally I'd trust Steam absolutely, but in this particular case, it was suck-ass).

This is performance related not stability.

 

[Syphon Filter]

I take it by purchasing via Steam I have to use the Steam version? Or does it grant me a license for the software and I can use any version (i.e. download from FutureMark)??

In any case, how should go about checking for hardware issues?

 

[Syphon Filter]

I'm having stability issues when running loop testing and get error about display resolution changing

While the loop tickbox in Custom tab does disable most of the anti-tampering checks, when loop testing for hours in full screen mode, the run can be interrupted if some background application draws a popup or task bar notification as it causes a switch back to the desktop view. 3DMark cannot recover from such display mode change.
For very long stability stress tests we recommend testing using the windowed mode. Windowed mode is not interrupted by task bar popups or other interrupts from background applications, only actual driver crashes or general system instability will stop the test. When looping all parts of a test, there are loading screens with light load between each part. For maximum load set the stress test loop to loop a single part of a test - Fire Strike Game Test 1, for example - and the loop will be continous without loading screens. Note that when looping, the run will never produce a score.

I've not had this resolution error but I wonder of it's something similar? I am going to have to do some testing and first kill off any random software in the background.

How long is long enough to determine stability?