What is the best way to check for stability on a GPU OC?

[chimaxi83]

My 7950 will run anything I throw at it, clocked at 1255/1725 with core voltage of 1.3V and RAM voltage of 1.675V. 3dmark loops, Heaven loops, hours of BF3, etc. It'll even do Furmark in its normal state, which causes GPU to throttle. Its water cooled on the core and doesn't go any higher than 41C, RAM and VRMs are heat sink cooled and VRM stabilizes around 60-62C during all usage.

Once I did the workaround to allow Furmark to suck up all it wanted, VRMs shot up to 95C before I exited the program. That, to me, is useless. What I wanted to see was how cool my setup could keep the core under an unlocked Furmark watt sucking load, but what I got was an entirely unrealistic power draw which any "normal" maximum GPU load would never come close to touching.

Furmark and any other "power virus" is designed for maximum current draw, nothing more. They might be good to test your cooling solution, but not a true method of game stability testing. Hours of stable 3D gameplay are a tried and true method, and that's really all there is to it.

 

[fstime]

Just ran Furmark with the ASUS GPU tweak monitoring log enabled with stock clock speeds.

Runs for 1-2 minutes before crashing and causing a blank screen.

Here are the log results right before crashing:

GPU Temp: 70, Memory Temp: 72, Power Temp: 72, GPU usage: 97, GPU Voltage: 1005, Memory Voltage: 1601, PCIE12V: 12031, Fan Duty: 32

Fan doesn't even kick in, only 32/100.

Something seems strange, memory is not at the 100C I was expecting to be causing the crash.

Like I said before, I can run games for hours but this won't run for 2 minutes.

 

[vollman1]

Looks to me like practical testing is the best route to checking the stability of your OC.

Any other games beside Crysis that are good for stability testing?

 

[bullbearish]

BF3 is good too.

I tested my OC with Furmark, but then had to back off a bit because I ran into some stability problems during gameplay.

 

[superjim]

Honestly gpu intensive games like Crysis2 at heavy detail settings seem to be the most reliable stress test for me. I've been able to loop benchmarks (Heaven, 3dMark11) and gpu burning utilities for hours and have no issues, then play 15 minutes of Crysis at the same settings and crash or see artifacts because of the OC being unstable.

^this

Same experience. Stress test for 24+ hours, everything passed. Start up a game and BOOM! I usually always have a recent game with a built-in benchmarking tool that I run a few times to get initial stability and then I play the game normally (fully expecting a crash, so no PvP and I save a lot). This generally gives me the best results. No stress testing tool will ever beat "playing the damn games" for hours on end.

 

[xxslay3rxx]

Heavens benchmark

 

[Mopetar]

Edit: Imagine if bridges were designed, only to handle medium or light traffic, and not a full highway full of cars. Would that be acceptable to you? Like I said, GPUs are under-engineered.

A bridge is probably designed to handle being packed with vehicles, but if you were to stack those vehicles in such a way that all of their weight were placed on one side of the bridge (think car hauling semis) or stacked so that their weight was all distributed to a minimal point of the bridge, it may no longer be safe.

Engineers could certainly design a bridge that would safely bear such loads, but it would be impractical because such scenarios won't occur. Similarly, a GPU could be made to better handle things like FurMark, but what they do serves little practical purpose so there's no point in building a GPU capable of doing such.