Question GPU Power Limit Constantly hit and Core Clock Slowly Decreases.

krytex

Prominent
Feb 2, 2019
9
0
510
I have a Zotac RTX 2070 Super AMP and when benchmarking on userbenchmark it says my GPU is underperforming, which I think it is in GTA V. People say userbenchmark isn't totally reliable so I benchmarked using Superposition Benchmark on High and compared to others and my score is quite lower. I set the power limit to 111% (highest). Afterburner says that the power limit is constantly on 1 (well 1 then 0 then 1 then 0 etc. switching every second or less). When benchmarking I saw on the OSD that the power limit hits 112% then tries to go back up but hits 112% then goes back down again. I get that the limit is 111% but my GPU seems to really want to go higher and that may be the problem.
On the other hand when I use the benchmark the GPU boost clocks to 1950Mhz (no oc) as it should but then the core clock slowly goes and the temp slowly goes up.

EDIT: I have sorta fixed the problem. I undervolted the card and the power limit is barely hit and the boost clock lowers much slower and much less this time. As I lowers much slower, sadly it still does and it's because of the temperature.
Turns out the Turing (RTX) cards lower the boost clock as the card goes above 65 degrees celcius, even though 65 is still stable.I took the side panel off and put the fans to max while using Superposition benchmark and the clock stayed at 2010Mhz until it hit 66 degrees then decreased to 1995Mhz. The temp maxed at 68 degrees so the clock didn't lower after that but with the fan speed not maxed as the temp goes the clock goes down (I assume due to increase in resistance like a thermistor) and caps at 75 degrees on superposition where the boost clock lowers to 1920Mhz from 2010Mhz.
Thanks for the help but I guess I'm going to either water cool the card and hope it stays below 65 degrees or (more likely) wait till RTX 3000 series is released, sell this card and purchase one of those and hopefully they will have fixed this issue.
 
Last edited:

Phaaze88

Titan
Ambassador
It's behaving normally.
It won't boost as high when it hits the power limit; the warmer it runs, the less aggressive the boost clocks.

https://www.techpowerup.com/vgabios/212242/zotac-rtx2070super-8192-190623
The vbios info for your gpu. A default power limit of 215w, 240w through software adjustments.
Gpus of different models will have different power limits. Here's the Zotac Twin Fan model: https://www.techpowerup.com/vgabios/213443/zotac-rtx2070super-8192-190619
It's capped to 215w.

You've already adjusted the power limit. Nothing else to do besides keeping it as cool as you can...
Oh, and overclocking obviously makes it use more power(more heat), so it'll hit those limits more frequently.

What kind of temps does it see normally?
 

krytex

Prominent
Feb 2, 2019
9
0
510
(all in degrees celcius) Overwatch: 63 in the training
Ranbow Six Siege: It starts at around 62 degrees when entering a round then slowly moves up, after a full round it moved to 70 degrees but seemed like it could just keep going
GTA V: When doing the benchmark it did the same as Siege, it started at 53 then slowly went up. By the plane part it was at 60 and by the end it was at 64. The wierd and annoying thing is that the core clock slowly decreased by 15mhz every so often, started at 2025 then went down to 2010 just before the plane part then 1995 by the end AND the GPU power limit was only at 65% to 90% the whole time so I have no clue why the core clock keeps decreasing.
 

Phaaze88

Titan
Ambassador
None of that is wierd. That's completely normal.
It's just how Nvidia's Gpu Boost algorithm works. You won't see fixed core frequencies on that gpu like you do on a manually set Intel cpu.

How high the gpu will boost depends on the available power and thermal headroom available.
You can download and run Gpu-Z: https://www.techpowerup.com/gpuz/
Click on the Sensors tab, and find Perfcap Reason. It'll tell you what's holding the gpu back at the time, so test it out. It'll give you one of the following reasons:
Idle = Self explanatory.
vRel = Reliability. Indicating performance is limited by voltage reliability.
VOp = Operating. Indicating performance is limited by max operating voltage(Hardware Limit).
Pwr = Power. Indicating performance is limited by total power limit.
Thrm = Thermal. Indicating performance is limited by temperature limit.
Util = Utilization. Indicating performance is limited by GPU utilization.
 

krytex

Prominent
Feb 2, 2019
9
0
510
When benchmarking on Superposition 1080p Extreme is constantly on Pwr so that's definetely the issue. I have the power limit on 111% and cannot go any further if I wanted on Afterburner. Would getting a higher Watt PSU help or should I try and enter the VBIOS and increase it or leave it would you recommend as you seem quite knowledgable P.S I like the title "Herald", reminds me of Galactus/Silver Surfer
 

Phaaze88

Titan
Ambassador
When benchmarking on Superposition 1080p Extreme is constantly on Pwr so that's definetely the issue.
This really means your gpu is working normally. It's going to hit one limit or other, and Power is probably the most favorable one; the gpu is going all-out.

I have the power limit on 111% and cannot go any further if I wanted on Afterburner.
Yeah, 240w limit according to the vbios file. The card will back down some when it hits that, and then bounce back up, over and over and over...

Would getting a higher Watt PSU help
Nope. Psus don't magically raise the vbios power limit.

should I try and enter the VBIOS and increase it or leave it
That would involve a hacking. That's not something that can be accessed regularly.

Nvidia was conservative on the limits of these cards to cover their asses - or assets, depending on how you look at it, and to keep the average Joe from blowing up, or melting, their products.