Unstable/Hot GTX 670s, help?

Avetiso

Honorable
Sep 18, 2012
38
0
10,530
Okay, so I've spent the last few days overclocking (trying) my GTX 670's. I maxed at out 1228mhz for the stable GPU clock, and a max of +588 on the memory, which is somewhere around 3590mhz.

However, I'm way over TDP. WAY over. GPU-Z is showing max TDPs of %137, with spikes to %387 right before a crash (I assume this is just an anomaly related to crashing, and meaningless). I get throttled around 129%.

So I figured, hey, not the best overclockers. Whatever, right? Well, at those clocks, I was getting throttled double. First, because I couldn't keep my temps under 70C, even at max fan, MX-4 TIM, and very good airflow through my HAF 922. Ambient temps was 72F. So I lowered them. And lowered. And lowered. Down at 1202mhz on the GPU and +400 on the memory, I'm still getting throttled, still breaking to 74C, and still hitting 132%+ on TDP.

So I revered to stock. Still can't stay under 70C; I'm slightly eeking to 71C-72C. Tried with side panel on and off. At +122% power target, I'm still seeing 125% TDPs.

Is there any way to get my temps down besides alternate cooling? Other people are getting 65C on reference cards at overclock, and I can't stay under 70C at stock. Bummed.

Is there any way to lower my TDP, or is this just the silicon lottery?

This cards have been a nightmare to OC. Nothing seems to make sense or work according to the guide. I might get 5 Heaven runs at a certain GPU OC; then I lower it by 3mhz, and I crash instantly. I did a few stable runs at +595mhz on the memory, then I crashed. Kept crashing all the way down to +588mhz.

I've looked around and have seen a few people with VRAM issues. They sound very close to home, because they all have: low, low OCs, high TDPs, high temps, and slight artifacting, even at stock clocks.

I seem to be getting all of these. When running Heaven at stock clocks, stock power, stock volts, etc, I still get slight artifacting. Textures will appear/disappear quickly, the clouds don't change color or form smoothly, rather, they do so in jagged lines. Some textures fail to load fully, like that scene where you are hovering over the house with the chimney. The detail on that chimney is always low res, as if it never loaded correctly.

Should I just be looking to RMA? How likely would it be that eVGA would accept both cards on RMA?

Or is there anything else that I have not considered?

Or maybe I'm a bloody idiot that has no idea what I'm doing. Feels like it.
 

Avetiso

Honorable
Sep 18, 2012
38
0
10,530


When experimenting with memory clocks, it's usually a mixture of exe crash/driver crash. When experimenting with GPU clocks, it's only exe crash.

I'm looking into aftermarket coolers for reference cards atm.
 


The crashing doesn't seem like an overheating issue. If you wanted something that looks cool though I suggest something from the arctic accelero lineup. Back to the issue though, what driver version are you using? What does Event viewer say about the crashes?
 

Avetiso

Honorable
Sep 18, 2012
38
0
10,530

I'm not crashing any more. I'm using driver 320.18.

I found my max stable clocks for card 1, haven't worked on card 2 yet. The thing is, I can't seem to stay under 70C. So even though these clocks are "stable", I can't seem to actually test them. So even if I lower my overclock by 50%, I still hit TDP throttle, as well at temp throttle.

Things look slightly better at stock clocks, but I still get temp throttled.

Still have some memory-related weirdness, IMO. Even at stock, on Heaven, some textures appear/reappear, random jagged lines, clouds don't change colors smoothly, etc.

I haven't been looking at event viewer when crashing, but I'm not crashing any more, so I'm sure it's relevant. Is it?
 
Hmmm, What about in furmark? Thats probably better at stressing your cards than heaven, its designed to do that, heaven was just meant as a show piece. Are textures loading properly during games/etc?

I personally would leave it at stock settings and let GPU boost take care of it until it hits its thermal target. I'm not really sure what the texture issues are about though. I might try seeing if it only does it with one card or the other.
 

Avetiso

Honorable
Sep 18, 2012
38
0
10,530


I'll try FurMark and 3DMark11.
 

Avetiso

Honorable
Sep 18, 2012
38
0
10,530
So I plugged in my second card. I ran card 1 overclocked, number 2 stock. Number 1 is now staying under TDP! For the most part, at least. It also only got throttle once, due to temps! What gives? Does having SLI increase the efficiency of the cards?
 
When you ran them was it in SLi? If so then the cards downclock them selves to the speed of the lower clocked card. And what were the temps where they throttled them selves. Most cards can handle (albeit its not good for them) up to 100c. (yup, not a typo, Celsius)
 

Avetiso

Honorable
Sep 18, 2012
38
0
10,530


I was running them one card at a time.

The OCed card does not downclock to the other card. This has been true for GTX 5/6 series. Card one ran my OC at 1228mhz and 5980 mhz memory while the other card ran at stock. GPU-Z, afterburner, and furmark confirm.

Which is what is weird. With the second card added in, the OCed uses less energy to run the same clocks. I don't get why, but I'm certainly not complaining.
 


Interesting. Something I did not know, good info.

That is weird. Hmm, you do happen to have a rather beefy psu. Gold rated also, it might be that you are hitting its efficiency rating since you are using more power and as such are getting slightly better/more stable voltages? Dunno that was honestly just a shot in the dark :)

I take it, it doesn't crash during furmark/3dmark/etc?
 

Avetiso

Honorable
Sep 18, 2012
38
0
10,530

No crashing. I guess putting them in SLI works wonders for overclocks. Whatever, even though I wish I understood why this is the way it is, I guess I'll be content with where I am.