GTX 1080: PWR PerfCap, Weird Behavior, Memory, and Temperature problems.

[parufy]

Hello! I came here to ask some questions about my GTX 1080 and some possible problems I am having with it.

So first up, I recently installed the Kraken G12 and an EVGA CLC 360 on my 1080. I know the AIO is overkill, but it was on sale for lower than a Kraken x53, so why not? Anyways, I installed the cooler and became immediately obsessed with overclocking my card. I used the Unigine Superposition benchmark on 1080p extreme because I thought that would be an ideal workload for my card. After a few hours late in the night, I got the card to clocks and voltages that were rock-solid in the benchmark (2164mhz, 1050mv) and with the PerfCap reason "none" in GPU-Z, so I fired up 3Dmark. The card immediately started drawing ~150-170w (is this okay?) off of the lone 8-pin on the card and slammed me with the PWR PerfCap reason, along with destabilizing the clock speed, so I stopped the benchmark and went to bed.

I am here to ask about that result now. It seems every other benchmark I've run since then has had the same power/instability problem. I even ran the SAME Superposition benchmark, but in 4K, then on 1080p high, and experienced identical issues to 3Dmark and every other benchmark and game I tried. BUT NOT 1080P EXTREME!! That benchmark still runs perfectly fine. No problems, stable, no PerfCap reason... I don't get it. I think I'm just salty for the lost sleep spent overclocking for nothing, but I digress. I guess I should just stay away from the 1080p extreme benchmark then? What would explain this, and what would you guys recommend I use instead?

Next up, temps. I have already seen my card hit 48C running stock in a mere 10 minutes (minecraft path-traced shaders lol). That temperature is awfully high for a GTX 1080 running on a 360mm AIO right? I have Corsair SP120 RGB pros running ~1100 RPM and I feel quite a bit of airflow. I also used the pre-applied thermal paste that came with the cooler. I'm fairly certain re-seating the cooler and using some fresh MX-4 is the play here. Thoughts?

Following up, VRAM. I tried overclocking the memory and was receiving good performance gains, but after adding around 150mhz consecutive runs resulted in lowered scores and eventually crashes. My card does not have memory temp sensors, so I cannot answer reliably on temperature issues. Previously, with the card's air cooler, I got the memory to around +350mhz. Does this result sound more like instability, temps, or something else? With the G12 installed there's no direct cooling for the little guys. Are they okay running with good chassis airflow? They are Micron brand chips. I've read that the G12 cools them just fine, and the complete opposite; that I need to install heatsinks for them. For some context on my system cooling, I am using the O11 Dynamic with all fan slots filled and I swapped the stock NXZT fan for an NF-A9. I would think the air coming from the bottom of the chassis combined with the air from the NF-A9 would be sufficient for the memory, but maybe not? I could try strapping the unused NZXT fan to the left side of the card to give the memory better cooling?

All in all, I am excited to have finally been able to mess around with some PC stuff again! I would likely be doing this exact mod to an RTX 3080 right now... but that's not the current reality. 2150mhz+ ain't so bad for a Gigabyte Windforce card. This will probably be the last thing I change in my PC till I get a new card in a year or two so I want it to be running properly! Any help is greatly appreciated. Thank you all.

 

[Vya Domus]

Following up, VRAM. I tried overclocking the memory and was receiving good performance gains, but after adding around 150mhz consecutive runs resulted in lowered scores and eventually crashes. My card does not have memory temp sensors, so I cannot answer reliably on temperature issues. Previously, with the card's air cooler, I got the memory to around +350mhz. Does this result sound more like instability, temps, or something else?

The more you overclock the VRAM the more it eats away at the power budget until eventually you start to lose clock speed and performance, it always works like this with Nvidia cards.

The card immediately started drawing ~150-170w (is this okay?) off of the lone 8-pin on the card and slammed me with the PWR PerfCap reason, along with destabilizing the clock speed, so I stopped the benchmark and went to bed.

Sounds like you left the default power target, so yes, 180W is the stock power limit. You're not going to achieve much if you don't set the power limit all the way up. Even then it's not going to run at 2164 Mhz all the time, it will go down.

Next up, temps. I have already seen my card hit 48C running stock in a mere 10 minutes (minecraft path-traced shaders lol). That temperature is awfully high for a GTX 1080 running on a 360mm AIO right? I have Corsair SP120 RGB pros running ~1100 RPM and I feel quite a bit of airflow. I also used the pre-applied thermal paste that came with the cooler. I'm fairly certain re-seating the cooler and using some fresh MX-4 is the play here. Thoughts?

I also have a 1080 with a G12 and an 120mm AIO and temps are about 48C, 50C at the very most. 48C does seem unusual for a radiator that big but it's not terrible. Maybe the pump is too slow ? How do you control it ?

 

[parufy]

The more you overclock the VRAM the more it eats away at the power budget until eventually you start to lose clock speed and performance, it always works like this with Nvidia cards.

That makes sense, but wouldn't it just crash immediately if the memory was taking too much power? I could get like 2-3 runs in without increasing clocks and would have lower scores every time before crashing.

Sounds like you left the default power target, so yes, 180W is the stock power limit. You're not going to achieve much if you don't set the power limit all the way up. Even then it's not going to run at 2164 Mhz all the time, it will go down.

Sorry, I meant 150-170 off of just the 8-pin. Card power draw was around 210w. I've read that an 8-pin PCIe is meant for 150w max, hence my worrying.

I also have a 1080 with a G12 and an 120mm AIO and temps are about 48C, 50C at the very most. 48C does seem unusual for a radiator that big but it's not terrible. Maybe the pump is too slow ? How do you control it ?

Before installing I did some research on what temperature ballpark I should be in. I saw a video of someone running an RTX 2080 on a 240mm with the card running around 42C max. I think my temps should be much, much lower. About the pump speed, it should be on max. I have the pump plugged into my motherboard's AIO header, and both my motherboard and Hwinfo show the pump running at ~1425 RPM. This is likely the wrong reading as stated in this forum: https://forums.evga.com/CLC-360-Pump-Speed-Question-m3065388.aspx Sadly, I cannot use the EVGA software because the USB header on the waterblock is obstructed by the thumbscrews used for installation.

 

[basco]

you could try to use the oc scanner tool in msi afterburner or evga precision and work from there or compare what you got.

and what Vya said open up that power target to max .

 

[Vayra86]

That makes sense, but wouldn't it just crash immediately if the memory was taking too much power? I could get like 2-3 runs in without increasing clocks and would have lower scores every time before crashing.

No. Nvidia practically perfected its GPU Boost with Pascal.

GPU Boost – Nvidia’s Self Boosting Algorithm Explained

Graphics Card technologies have progressed by leaps and bounds over the last few generations with each generation bringing a substantial improvement not

appuals.com

The TL DR:
- GPU has a Base clock and won't go under it
- GPU has a Boost clock and will try to hit that even at stock, as long as there is voltage/thermal headroom
- Any OC uses an Offset clock to add or deduct from the Boost algorithm

So you're basically stuck with Boost any time, and the 2164mhz you are seeing is your 'peak boost' when temps are below 50C AND there is voltage headroom to keep stability at that frequency. If any of these factors are not met, you will find the GPU dropping what is called 'boost bins' - increments of 13mhz each.

The reason your 1080p Extreme run doesn't drop clocks similarly is probably because it is less demanding, the most plausible reason being less pressure on frame buffer (VRAM) which means more power budget for the core. Temps are not the culprit, well maybe for your first loss of 13-26mhz, but you've got them nice and low already.

The reason your subsequent runs of benches are lower in points, is because the GPU is heating up faster as its already warm as it came in. The best Pascal bench is always the first run Losing the last few extra degrees Celsius on a hot GPU takes a LOT longer than losing the peak temp.

Note that 2164mhz is a very high boost clock for Pascal GPUs, especially a GP104. Count your blessings The bestest samples reach 2200mhz and only in marketing.
The highest clock I could pull out on air was 2134mhz, but it would generally drop to under 2100mhz in-game. And stability wasn't guaranteed - you can still pull the offset too far out and get crashes.

When you want to take your OC in another direction, I'd suggest looking for the 'highest stable sustained clock'.
On air, this gets me to a 95% power target (!!!) with a hefty offset on core and memory. No extra voltage, because temps would limit me. In your case, you'd likely not change too much, apart from playing around with the offsets. Peaking not quite as high could net you more budget for longer/higher sustained clock. But its worth looking at the effects of a slight voltage reduction too - you might get to a similar place because now the boost bin is not dropped because you can stay under 50 C.

Sorry, I meant 150-170 off of just the 8-pin. Card power draw was around 210w. I've read that an 8-pin PCIe is meant for 150w max, hence my worrying.

You get 75W from the PCIe slot too, so a single 8 pin 1080 can draw up to 225W. An 8-pin is specced for 150W but it can really do quite a lot more if you'd mod it.

 

[parufy]

Sorry for going silent for a while; I got caught up in school lol

the 2164mhz you are seeing is your 'peak boost' when temps are below 50C AND there is voltage headroom to keep stability at that frequency.

I have a voltage curve set, yet it doesn't maintain voltages and frequencies even though the temperatures are low. So that means other components on the card are trying to draw more power than the core? If that's right, that makes sense.

The reason your 1080p Extreme run doesn't drop clocks similarly is probably because it is less demanding, the most plausible reason being less pressure on frame buffer (VRAM) which means more power budget for the core.

That makes a lot of sense. I was sort of thinking that, but VRAM usage between 1080p and 4k was only around 700MB which made me think otherwise. Maybe it's because 4k is more memory clock intensive while not using much more VRAM in this case?

The reason your subsequent runs of benches are lower in points, is because the GPU is heating up faster as its already warm as it came in. The best Pascal bench is always the first run

This is understandable, but I just tried retesting memory. I set everything to stock to get a baseline score, then set the VRAM to +300 (after heating the card without OC). The first run with the OC was ~250pts more, however, after 3 runs the score was lower than stock. When running the test without memory overclocked, scores remain unchanged

Count your blessings

Word.

95% power target (!!!)

So you're running less than a 100% power target? Would that really apply in my case because temperatures are so much lower? (read on)

Speaking of temperatures... screw the stock thermal paste EVGA uses! I just disassembled the cooler and reapplied with some MX-4 along with tightening the screws a little further. Temps went from around 48C in 10 minutes to 36C MAX while overclocked!?! wtf EVGA? I'm gonna have to do some retesting. again.

 

[Mussels]

I also have a GTX 1080 with a 280mm AIO with the G12
Never passes 45C, usually under 40 if the ambients aren't high (Au... so we can hit 40C ambient here)
I used Conductonaut because i have balls of steel.

How your rad and fans are setup in the case will determine your temps here, which you havent shared (pics help. people do dumb shit with water and AIO setups all the time without realising)
If the load temps build up over time, then you need to change the setup (fan speeds at least, or how its all assembled at worst)


You will always be limited by something.

 

[R-T-B]

Speaking of temperatures... screw the stock thermal paste EVGA uses!

EVGA uses Shin Etsu, which is actually better than many stock pastes. I've found they often botch the application though.

 

[parufy]

EVGA uses Shin Etsu, which is actually better than many stock pastes. I've found they often botch the application though.

I feel like it was a mix of my mounting pressure and the paste. When I took off the cooler I could see a lot of shiny untouched die.

I used Conductonaut because i have balls of steel.

I'm chill with 8g of MX-4 for $7.99 lol. I might have too much, actually. Does this stuff expire or does it just separate over time?

How your rad and fans are setup in the case will determine your temps here, which you havent shared (pics help. people do dumb shit with water and AIO setups all the time without realising)

Images: Imgur. I really wanted to mount the GPU's rad on the bottom, but I know that's basically non-functional and very bad lol

 

[Mussels]

Wow yeah, i was worried for a sec that was bottom mounted, you'd get air in the pump FAST that way and have serious issues
The fans on that rad would be better as push and not pull, imo - but i'm not that familiar with your case

 

[parufy]

Wow yeah, i was worried for a sec that was bottom mounted, you'd get air in the pump FAST that way and have serious issues
The fans on that rad would be better as push and not pull, imo - but i'm not that familiar with your case

Do you mean pushing as an intake behind the rad or pushing as exhaust in the current position? Also, I thought push/pull had negligible differences in performance?

 

[Vayra86]

Sorry for going silent for a while; I got caught up in school lol


I have a voltage curve set, yet it doesn't maintain voltages and frequencies even though the temperatures are low. So that means other components on the card are trying to draw more power than the core? If that's right, that makes sense.


That makes a lot of sense. I was sort of thinking that, but VRAM usage between 1080p and 4k was only around 700MB which made me think otherwise. Maybe it's because 4k is more memory clock intensive while not using much more VRAM in this case?


This is understandable, but I just tried retesting memory. I set everything to stock to get a baseline score, then set the VRAM to +300 (after heating the card without OC). The first run with the OC was ~250pts more, however, after 3 runs the score was lower than stock. When running the test without memory overclocked, scores remain unchanged


Word.


So you're running less than a 100% power target? Would that really apply in my case because temperatures are so much lower? (read on)

Speaking of temperatures... screw the stock thermal paste EVGA uses! I just disassembled the cooler and reapplied with some MX-4 along with tightening the screws a little further. Temps went from around 48C in 10 minutes to 36C MAX while overclocked!?! wtf EVGA? I'm gonna have to do some retesting. again.

No, you want to push voltage as high as you can dissipate the heat really. Just illustrating there is a point of negative or diminishing returns, but on water you prob wont see it.

 

[Mussels]

Do you mean pushing as an intake behind the rad or pushing as exhaust in the current position? Also, I thought push/pull had negligible differences in performance?

I don't know your case, but ideally you'd want the heat exhausting the system, not being pulled into it