[SOLVED] Unstable ASUS ROG Strix Geforce GTX 1080 (Advanced Edition)

[Magic-I-an]

I've had the setup in my signature for about a year now. I bought it used from a guy who originally wanted to be a game streamer, but ended up suddenly leaving the country without his PC. All of the components were disassembled and stored in their original boxes when I bought it. As far as I can tell, he never overclocked anything and don't believe he used it for crypto mining. When I first started using it for gaming, I didn't overclock anything, since it ran most games fairly well at stock speeds. Eventually, I decided to give it a go and push it to its limits. The GPU overclock was disappointing to say the least. An increase of only about 20-30 MHz caused it to become unstable in Unigine benchmarks. Assuming I lost the silicon lottery, I was content with stock speeds.

Over time, more and more games would crash. Oddly enough, running games above 60 Hz refresh rate was the biggest culprit, so I settled for 60 Hz. Seemed like a shame to have a 144Hz monitor, so I tried underclocking my GPU by about -50 MHz, which seemed to do the trick. Eventually, my GPU became so unstable that games were crashing regardless of the settings I used. Even Unigine benchmarks could only run for a short period of time before crashing, where originally they easily ran the whole length at stock speeds.

I've never seen a GPU degrade so drastically over a short period of time, so I tried doing some basic diagnostics. I did a clean driver installation using DDU, I updated my BIOS, cleaned out the GPU with compressed air, re-seated it in the motherboard, tried a different PSU, but nothing helped. I even plugged in an old GTX 760, which had no stability issues with the same system, even at a 100 MHz overclock. Temperatures seemed normal as well. Idle was about 30-40 C and under load was about 60-70 C, with a max temp of about 74 C. Nothing outrageous, but it could be better.

Assuming temperature was the issue, I decided to buy a couple more 120 mm fans for my case and re-apply some new thermal paste. The factory thermal paste looked terrible. Only a thin dried out layer on the top of the chip and most of it was squeezed to the sides. I applied some Arctic MX-4 and plugged it back in. Temperatures were drastically better. Idle was now 27 C and full load was max 59 C. By using a more aggressive fan curve, I was able to drop that to 25 C and 55 C.

Things were back to being stable, so I tried overclocking again. Same problem. Even before reaching the max temperature, a modest overclock of 30 MHz would cause the Unigine Superposition benchmark to crash.

Any thoughts on how can get the most out this GPU?

 

[Phaaze88]

A card like that was already overclocked from the factory for you. There isn't much headroom left - the factory took most of that fun from you.
You may get more out of pushing the Vram - I don't think the factory touches that one.

 

[Magic-I-an]

A card like that was already overclocked from the factory for you. There isn't much headroom left - the factory took most of that fun from you.
You may get more out of pushing the Vram - I don't think the factory touches that one.

That's disappointing to hear. The stock boost sits comfortably around 1949 MHz, but I've read that many people are getting clock speeds from 2000-2100 MHz on their GTX 1080. I don't imagine I'd see much of a performance increase by pushing the memory.

 

[Phaaze88]

I don't imagine I'd see much of a performance increase by pushing the memory.

No, you won't, but it does see larger values, like 300+. You can't do that with core clock... at least, not without LN2 and some bios hacking + shunt resistor modding.

Overclocking as a whole doesn't yield much performance anymore. It's not as productive compared to what could be done on older hardware, and it's more or less the same with today's cpus.
Gpu Boost: Started with the 10 series. AMD has it on their cards too, but I'm not as familiar with them.
The gpu dynamically 'overclocks' itself based on power, thermal, and voltage. Out of those 3 parameters, the 2nd one(thermal) is the one we users have the most control over.
The cooler the card runs, the higher the SUSTAINED boost clocks will be - bar power limits, anyway.

Overclocking leads to more heat being produced by the card, and if that's not accounted for, an OC can end up doing much of nothing, or in some cases, it's worse because the sustained boost clocks are lower due to Gpu Boost noticing the gpu running hotter than it'd like, and thus lowering clocks a little more. Gpu Boost becomes a bit of a roadblock here. The older cards didn't have that feature; you could OC them to like 5C away from throttle temp, and they wouldn't stop you.

I'm saying all this, but it's not THAT big a deal. Like every 5C nets you 15mhz up/down, or something like that.

I've read that many people are getting clock speeds from 2000-2100 MHz on their GTX 1080.

It's easy to get, but is it actually sustainable everywhere?
I could push 2012mhz on my 1080Ti if I wanted, but in Time Spy/Firestrike, Unigine Superposition, or certain games, it drops into the 1900s rather quickly... and that's with a Celsius S36 on it - I blame the 300w power limit.
It never sees over 40C on the core, but 1949mhz is much easier for it to hold these days.