GPU Stability testing

Considering my main focus is benching, and really mostly subzero benching, Im really ignorant with regards to stability testing for GPUs.

I dont like running a single benchmark like 3D06 (or whatever) and saying, "yup it passed it must be good!" since a single 10 minute benchmark hardly equates to GPU stability over 10 hours of gaming

Whats a good test of stability for a GPU?

Ive been looping 3D01 Nature indefinitely. I usually give up after 1 hour and say its stable to go for a real gaming session.

Is 3D01 Nature looping a good stability test? Is there something better?

TY!

I like this one. I run it in windowed mode and adjust my settings real time.
http://www.google.com/url?sa=t&ct=r...EAEgN9JonTXKidaNw&sig2=5ur8gu1ageUGSrv108reWg

I use ATITool for a quick check, but you have to be sitting there watching it for yellow pixels. After that for 5-10 minutes, I let it run and check for errors for another 20-30 minutes on its own. Ive never had a problem after that, but then I run 3Dmark at night anyways. Like I said though, once it passes ATI tool, its been good for me.

Edit: Furmark is pretty good too, but it tends to lag instead of give visible pixels with slight errors. Its more of a personal preference, but I like ATI tool better for testing because I can see the video card errors and know that they arent from another process temporarily locking my system up. Fur is still pretty good at pushing the temps, just run it at the highest res you can, and crank the AA up to 16.

Dawgdoc said:

Considering my main focus is benching, and really mostly subzero benching, Im really ignorant with regards to stability testing for GPUs.

I dont like running a single benchmark like 3D06 (or whatever) and saying, "yup it passed it must be good!" since a single 10 minute benchmark hardly equates to GPU stability over 10 hours of gaming

Whats a good test of stability for a GPU?

Ive been looping 3D01 Nature indefinitely. I usually give up after 1 hour and say its stable to go for a real gaming session.

Is 3D01 Nature looping a good stability test? Is there something better?

TY!

Click to expand...

3DM01 Nature only, set to repeat (not loop) 30 times with the Title Screens Disabled
(so the test reloads from memory every pass instead of disk) will put a lot of heat into
the core. If you can get through the Nature torture loop with locking you will be pretty close.

Viper

I'm finally going to OC my 88gtx. I don't really need to, but... its what we do. I get 11,800 in 3D06 currently. What freq's can I hope for? Link?

nightelph said:

I'm finally going to OC my 88gtx. I don't really need to, but... its what we do. I get 11,800 in 3D06 currently. What freq's can I hope for? Link?

Click to expand...

You should land between 620-660 Core and 1075-1100 memory.

Alrighty, thanks. The only game I play that might benefit (besides Crysis) is DiRT.

Dawgdoc said:

You should land between 620-660 Core and 1075-1100 memory.

Click to expand...

Actual? or just what you set Rivatuner to?

I only got 621/1512/1053 out of my 8800gtx, but that is the actual running speeds, obviously I can set Rivatuner higher on each of those before the next bump up, but that has no bearing on the speed.

Maverick0984 said:

Actual? or just what you set Rivatuner to?

I only got 621/1512/1053 out of my 8800gtx, but that is the actual running speeds, obviously I can set Rivatuner higher on each of those before the next bump up, but that has no bearing on the speed.

Click to expand...

My quoted #s were Rivatuner #s.

Most ppl dont know about the 27mhz stepping on these G80 cores, so I typically just quote Rivatuner numbers. You are correct though that you could go 26 more mhz from the last step up and still have the core speed be exactly the same until that 27th mhz

Mkay, thats good to know. Is there any preferred oder in OC'in the card? Core, Shader, Memory? If theres a specific guide on this previously posted, feel free to just point me at that.

I don't quite understand the question in 1st post...
Who says you should use some benchmark program to determine gaming stability?

I believe its been established that if it can pass a repeat 3D01 Nature test, it'll hold up under gaming. No?

largon said:

I don't quite understand the question in 1st post...
Who says you should use some benchmark program to determine gaming stability?

Click to expand...

Well, prior to doing a 10 hour gaming session there should be some other type of stability testing other than just a single 3D run correct?

So what should that stability testing be?

Or......simply because you passed 3D06, 3D01, whatever, do most ppl just jump right into a massive game session?

I would think that it would be advantageous to run SOMETHING,

Dawgdoc said:

My quoted #s were Rivatuner #s.

Most ppl dont know about the 27mhz stepping on these G80 cores, so I typically just quote Rivatuner numbers. You are correct though that you could go 26 more mhz from the last step up and still have the core speed be exactly the same until that 27th mhz

Click to expand...

The G80 cards ROP and memory clocks will always step on integer multiples of 27mhz. They will
also step on some 1/3 and 2/3 multiples as well (720 ROP is an example of 2/3 multiple).

The shader clock only steps in integer multiples of 54mhz.

Viper

I really see no reason to specifically to bench a video card in order to determine gaming stability. If it stutters/crashes/artifacts heavily while gaming then by definition it is not gaming stable. Atleast, that's how I do/see it.

ViperJohn said:

The G80 cards ROP and memory clocks will always step on integer multiples of 27mhz. They will
also step on some 1/3 and 2/3 multiples as well (720 ROP is an example of 2/3 multiple).

The shader clock only steps in integer multiples of 54mhz.

Viper

Click to expand...

Is this to say I should raise gcore freq from 575 right to 602 in RT?? This stepping in cards is quite new to me.

largon said:

I really see no reason to specifically to bench a video card in order to determine gaming stability. If it stutters/crashes/artifacts heavily while gaming then by definition it is not gaming stable. Atleast, that's how I do/see it.

Click to expand...

Granted the best gaming stability test is actually gaming, but short of actually successfully completing a 10 hour gaming session there has to be some intermediate way to see if there is an expectation of this occurring.

Heres the situation for more detail so that you understand WHY I ask this question.....

One of my GPUs (8800 GTS G92) is volt-modded, water cooled (MCW-60R/PA 120.3, D5 vario) at clocks 875/2200ish/1135 on 1.38Vgpu with Quimonda ram at stock Vmem of 2.01. This card was benched AGGRESSIVELY subzero for our ForumWars 2008 and I really beat the crap out of it LOL...... (see below)

I want to isolate this card IN THIS RIG and see if I can prove it to be unstable as I am playing a brand new just released game (Age of Conan Hyborian Adventures) and I want to investigate further. I occasionally hang, or sometimes even massively artiface with colored blocks across the entire screen and then freeze requiring a reboot. Sounds like GPU issue and likely a memory issue to me but I want to test to see if I can prove if it is:

1. Server end problems
2. GPU problems
3. EVGA 750i FTW video corruption mobo issues. This is a known problems on the 750i and 780i, but I honestly have not researched this part yet to see if its relevant.

By testing this system away from the game, I can more or less remove #1 from the equation and then go into more detail testing this GPU in another rig or pursuing a motherboard issue.

CPU was holding me back on these runs....a shame too since Im not sure I can reproduce those GPU clocks again...at least not easily

I'm pretty sure it's your Qimonda GDDR3...
It's typical for them to "fake-OC" and seem like they're stable but then some time later they collapse and artifact/crash completely.

Got my card bench stable at 629, 1374, 940. Another 600pts in 3D06. Meh, gotta check out Crysis.

largon said:

I'm pretty sure it's your Qimonda GDDR3...
It's typical for them to "fake-OC" and seem like they're stable but then some time later they collapse and artifact/crash completely.

Click to expand...

Largon, when this happens with the Quimonda ram and it artifacts/crashes completely do you mean that in these situations the ram is no longer stable at the previous OC that you were using, or just crashing and no longer stable as in dying?