OCCT's GPU test, error checking, and overclocking

[tviceman]

How important is it to playing games (in relation to GPU stability) that OCCT's "error check" be at zero over a given length of time (say 20-30 minutes)? Is there a certain threshold that is considered acceptable?

The reason I ask is because I'm willing to bet many people's "stable" overclocks would in fact induce errors in OCCT's gpu test with error check enabled. I for one thought I had a 100% stable overclock with my GPU. Everything plays great, no artifacts, no lockups... but OCCT will slowly trickle out errors after a minute or two of running the GPU test with what I thought was a normal, stable overclock.

 

[MrK6]

How important is it to playing games (in relation to GPU stability) that OCCT's "error check" be at zero over a given length of time (say 20-30 minutes)? Is there a certain threshold that is considered acceptable?

The reason I ask is because I'm willing to bet many people's "stable" overclocks would in fact induce errors in OCCT's gpu test with error check enabled. I for one thought I had a 100% stable overclock with my GPU. Everything plays great, no artifacts, no lockups... but OCCT will slowly trickle out errors after a minute or two of running the GPU test with what I thought was a normal, stable overclock.

That's why stability is a relative term . It's really what you "feel" is stable, but I always add the caveat that a GPU on the edge of stability may degrade the gaming experience in ways that one might not generally attribute to overclocking (many times stuttering or patchy framerate rather than artifacting). That said, I use a combination of OCCT and Left 4 Dead (oddly enough, it's the best game I've found to detect errors) to test for clock stability.

 

[Flipped Gazelle]

How important is it to playing games (in relation to GPU stability) that OCCT's "error check" be at zero over a given length of time (say 20-30 minutes)? Is there a certain threshold that is considered acceptable?

The reason I ask is because I'm willing to bet many people's "stable" overclocks would in fact induce errors in OCCT's gpu test with error check enabled. I for one thought I had a 100% stable overclock with my GPU. Everything plays great, no artifacts, no lockups... but OCCT will slowly trickle out errors after a minute or two of running the GPU test with what I thought was a normal, stable overclock.

I can't answer your question directly, but I'll give my own anecdote here.

My most recent build included am MSI Cyclone GTX 460 1gb. It clocked to 900mhz core, passed an hour of Furmark and 3dmark Vantage. It subsequently failed my "Crysis: Warhead deathsuite" - 99 consecutive runs of the Framebuffer Crysis Warhead Benchmark. I had to dial back the core to 860mhz to get it stable.

Similarly, on my own PC with the Radeon 5770, I was able to benchmark OK at 980mhz, but had to dial that back to 960mhz for Metro 2033.

I guess the moral of my story is, ya gotta use multiple tools.

 

[tviceman]

Agreed with both posts. It still makes me go hmmmm to see that a particular clock speed starts producing errors within a minute or two of OCCT, yet can run furmark and benchmark through other games at a half hour no problem whatsoever.

 

[MyLeftNut]

From my experience with the OCCT gpu stress test, if I can pass it with low amounts of artifacts, it won't ever crash in games and can fold an entire day without errors. Furmark, 3dmark Vantage, EVGA OC scanner, and MSI Kombustor doesn't test stability very well as I've had overclocks that were stable or artifact free in those, but would artifact or crash in game and prove to be unstable when folding.

 

[bunnyfubbles]

from my experience, if I'm getting errors with a tool like OCCT or Furmark, it might be stable enough for games in the short term, but chances are the card will degrade much faster to where it will become flakey/unstable, if not die outright, within the worthwhile life of the card (ie before I plan to upgrade)

The meager few extra MHz or so are only good for epeen anyways, not worth the real world results, so I just back it down.

 

[Bman123]

I've always used furmark to test stability, if it artifacts in furmark i turn it down until it doesn't error then I get my game on