I Can't See Artifacts

Aeradom

Distinguished
Jul 31, 2011
100
0
18,680
To start, I'm new to overclocking GPU. And as such, I'm not entirely sure what I'm susppose to be looking for. Following the TomsHardware guide on the subject, I'm using Unigen Valley and using the program to gradually increase the memory, pausing to look around for specks, stars, or spots of solid color. But I couldn't find any. I knew something wasn't right when I pushed it to +900 and still wasn't seeing anything. It probably was a dumb idea I know that. The only reason I know for sure that it was too high was because my driver crashed.

I guess I'm asking is there another benchmarking software, or some sort other way to tell when I hit that threshold and need to pull back? Also, why is it important to raise the voltage because everywhere else says not to do that.
 
Solution
Artifacts are one possible clue that your card is being pushed harder than it can handle. They happen when tiny compute errors in the card occur because some component of the GPU isn't return the correct values - usually because the frequency is too much for the card to handle.

Another very common clue that an OC is too much is a driver crash. So whether you "crash" or see artifacts is kind of irrelevant. In both cases, it's a clear sign that you either need more voltage or less frequency for stability.
You're meant to leave the voltage at base, and gradually increase both the memory frequency AND the core clock. Keep doing this until you encounter what are called artifacts (looks like static fuzz) while benchmarking, increase the voltage and try again.
Rinse and repeat this process until you hit a point where no matter how much you increase the voltage, you can't get it to stabilize. Revert to the last stable clock and move your speeds back a bit and you have your final speed.
 
Artifacts are one possible clue that your card is being pushed harder than it can handle. They happen when tiny compute errors in the card occur because some component of the GPU isn't return the correct values - usually because the frequency is too much for the card to handle.

Another very common clue that an OC is too much is a driver crash. So whether you "crash" or see artifacts is kind of irrelevant. In both cases, it's a clear sign that you either need more voltage or less frequency for stability.
 
Solution


I would personally strongly advise that you OC your memory and core clock separately. Otherwise it's difficult to know which one failed. Tackle one first - I'd suggest core clock, because for most cards it has a much greater impact on performance - and once you have settled on a nice stable OC for that, start the process again for the other.
 

Aeradom

Distinguished
Jul 31, 2011
100
0
18,680


How long should I wait in between raising the clock speed another increment? Like how long does it take for something to happened?
 


I'd go the Heaven benchmark over Valley. I've found it exposes issues with OC earlier and others seem to say the same: http://forums.evga.com/Unigine-Heaven-40-vs-Valley-10-which-one-is-more-stressful-m1922332.aspx

Early on, I've found that most errors will crop up within one or two passes of Unigen Heaven. Once you've settled on a final OC though, it's worth leaving it looping for a good few hours or even overnight just to be sure things are stable.

If you're not sure what impact voltage has on OCing, you might be interested in a response I gave to a thread a while back. I'm sure there are more comprehensive guides out there from people who understand these things much better than I do, but I was struggling to find something that described it in an accessible way, so had a crack in that thread myself. If you're interested read my ("rhysiam") first response to this thread: http://www.tomshardware.com/answers/id-2693821/explain-overclocking.html

If you want the TL DR version, upping the frequency is what makes things faster, but at some point (usually fairly quickly) it'll crash or artifact. Raising the voltage means you can hit higher frequencies without crashes/artifiacts, but it raises power draw and heat and, critically, higher voltage means shorter life spans and in extreme cases sudden and catastrophic hardware damage... So while raising voltages is really the key to higher overclocks, you need to understand what you're doing or closely follow guides written by someone else who understands what they're doing before playing around with voltage sliders.
 

Aeradom

Distinguished
Jul 31, 2011
100
0
18,680


You know, I've seen Heaven referenced before so I think might just download that one instead and run it because the one thread I found that was talking about this here said they did see red spots so I'll give that a whirl. As to the voltage, the more I've done research, it seems like everyone just prefers to keep it at the base voltage and not mess wif it, so on that one I think I'd just prefer to play it safe. It's weird then that the two guides I've read, advised maxing out voltage when testing.

Speaking over that guide, can you make heads or tails of this or give a layman's view on this?

The website that has already overclocked the MSI GTX 970 Gaming has said that it was able to achieve 1300 Mhz GPU Clock speed.

[Maximum Overclock Achieved] - [GPU Clock Speed] = [Additional Clock Speed Needed]

Applying To Example: 1300 Mhz - 1102 Mhz = +198 Mhz

We want to give ourselves some headroom so increase the core voltage to about 3/4 of the "Additional Clock Speed Needed To Achieve Maximum Community Tested Clock Speed."

198 Mhz X 3/4 = +150 Mhz

In MSI Afterburner
Increase the "Core Clock (Mhz)" +150 Mhz
Press "Apply"

 
You should let the benchmark run for about 3-5 mins every time, closely watching for artifacts or harsh spikes in frame rate. Unigine Heaven is much better than valley for its consistency and better stressing of cards.
Typically I do the memory overclocking after i've found a stable core clock, but its all up to personal preference.