GPU Stability stress testing

[Rubnik]

Hello everyone,
I've been trying to OC my GTX 770 (MSI N770 TF 2GD5\OC), i believe I hit a stable OC at +80 (1245 boost clock) + 240 (3744 memory clock) at 1.2V.

The past few days I've been trying to push it a little further, for example +100 at 1.212V but I'm a bit confused with stress testing.
For example, with the +100MHz settings, it ran Furmark for 2 hours stable, temp about 72 Celcius, but when I used Unigine Valley, the program stopped working after a bit, and then displayed the same error again on every OC setting until I restart the system, so I considered it to be a software error.

So my questions are :

• What program would you recommend to test my GPU stability? So far I've been using Kombustor 3.3, Unigine Valley, Furmark and 3D Mark (mainly for the benchmark run though).
• How much time does the stress test need to run,for the OC to be considered stable?
• Do different settings on the stress test (resolution, AA etc) have different impact, or does GPU core just needs to be at 100% load?
• To test my memory clock stability, does the VRAM need to be at maximum load?

 

[BiggieShady]

Two benchmarks can both have 100% GPU load, and still not stress the GPU the same way.
Your GPU overclock may be stable in 3D mark and fail in Valley.
Also all your games may work fine with that overclock, and still GPU may fail running burn in benchmarks because games rarely push the GPU that much as those benchmarks.
GTX 770 is based on GK104 GPU so it's already overclocked GTX 680, and often there is less overclocking headroom especially considering memory is already at effective 7 GHz.

What program would you recommend to test my GPU stability? So far I've been using Kombustor 3.3, Unigine Valley, Furmark and 3D Mark (mainly for the benchmark run though).

I would stay away from Kombustor and Furmark because as so called power viruses they never test real life scenarios.

How much time does the stress test need to run,for the OC to be considered stable?

Watch the temps and if they stabilize, keep it there few hours (as average gaming session would take)

Do different settings on the stress test (resolution, AA etc) have different impact, or does GPU core just needs to be at 100% load?

Resolution has an impact on vram usage and complexity of calculations in post processing full screen shaders (SSAO or HBAO, DOF, SMAA, FXAA etc.) ... with bigger resolution come lower framerates and although gpu usage is always at 100%, OC can still fail because of different complexity of post processing.

To test my memory clock stability, does the VRAM need to be at maximum load?

In 3D mode memory always runs at max clock, no matter how filled it is. But to test all memory chips read/writes it is best if video memory is full and not only that but also constantly used whole for rendering the benchmark. I believe Unigine benches use plenty of vram.

 

[Rubnik]

What about Unigine crashing though? I read other people have similar problems with Unigine, some times even on stock settings.
If the program\game stops working it's not necessarily because of the OC right? Unstable OC= driver error message artifacts and screen freeze?
Should i just leave 3DMark running in loop mode then?

 

[RealNeil]

How does the system run Catzilla? Will it crash with it?

 

[Rubnik]

Never tried it. Downloading it now and I'll give it a try

 

[RealNeil]

Make sure that your GPU drivers are up to date. It will not run if they're not.

 

[Rubnik]

Checked both settings I mentioned above at 720p, didn't crash, but still it's just 1 run each.

 

[RealNeil]

I think you can get the advanced version on Steam on sale for $14.99 right now.

 

[the54thvoid]

You won't get a benchmark to give you predictable overclocks. My clock runs BF4 24/7 but will give me artifacts in Unigine Valley bench. Benchmarks tend to be for max clocks, not game stability. Find a bench stable clock and try gaming on it. Some games don't like overclocks either. It's a lottery but in real life, the same max overclock will not work across all games.

 

[newconroer]

You won't get a benchmark to give you predictable overclocks. My clock runs BF4 24/7 but will give me artifacts in Unigine Valley bench. Benchmarks tend to be for max clocks, not game stability. Find a bench stable clock and try gaming on it. Some games don't like overclocks either. It's a lottery but in real life, the same max overclock will not work across all games.

This is the unfortunate reality. Point and case for me is that I can get a stable Unigine and Furmark over clock, and then play Far Cry 3 for where upon it will crash randomly. If I reduce/remove the overclock, it doesn't happen. Meanwhile, that little bit of overclock makes or breaks for me whether I hover around 55fps or I cap at 60.
Using these programs to get you started is ok, but your best bet is to just play games for several days and see what happens.

What would be helpful for people to understand is whether the core clock or the memory is the culprit in most failed clocking scenarios.

 

[sneekypeet]

What would be helpful for people to understand is whether the core clock or the memory is the culprit in most failed clocking scenarios.

This is why you clock one at a time That way you know if its core or VRAM.

 

[newconroer]

This is why you clock one at a time That way you know if its core or VRAM.

I meant as a general rule of thumb. In my experience it's been the core clocks. The VRAM seems to handle itself (unless you go extreme), until you get into elongated and miscellaneous testing(a.k.a. gaming), where upon it starts to sh*t on itself.