This is why you are getting less than 2000Mhz on your new 10 Series GPU.

[QwerkyPengwen]

If this has already been covered in another thread forgive me and feel free to merge this if you like or just leave it be as I may end up going further into detail about some things that another thread probably didn't or you can end up closing it and deleting it if you must.

Ok. So I've got quite a bit of information for you showing and explaining why you are getting sub-par performance with your new 10 series GPU from NVIDIA.

What I'm about to go into ONLY applies to the new 10 series GPU's and will only partially apply to previous generation cards.

So I spent quite a lot of time testing and preparing all of the information for this so I hope you enjoy and find it informative.

As a base line to keep in mind when I did this testing It was kind of late at night and I live by the beach so my ambient temps in my room were about 45f or aprox. 7c.

The initial testing was done with an open case and the final result was done with a closed up one.

So without further ado, let's get started.

So basically NVIDIA has this custom curve for their GPU's when it comes to the 10 series cards that it turns out is quite aggressive as you will see in this post.
This "Curve" is in conjunction with GPU Boost 3.0 and governs your clock speeds and this "Curve" is governed mostly by your temps and voltage.

Now this curve basically ends up either raising or dropping your clock speeds by 13Mhz every time you go above or below a certain threshold mainly with your temps.

The GPU I used in this test was an:
ASUS Strix GTX 1070 Gaming 8G

The software used to overclock and monitor everything was:
MSI Afterburner v4.4.2 with RivaTuner Statistics v7.0.2.15196

The game used for pushing the card in order to monitor this was:
Tom Clancy's Ghost Recon Wildlands at 1080p with Vsync off and graphics on the Very High preset.

NVIDIA Game Ready Drivers v390.77

Testing was performed between the hours of 12am and 5am PST on Monday February 19th 2018

So starting things off, I opened up my case by taking off the side panel to allow for the best possible temps when doing these tests. As you will see in this picture below, I started with my baseline for temps at idle on the desktop and set my overclock settings to their maximum stable level that I found out through the standard methods before doing this test.

As you can see, I have all of my bases covered for reaching maximum stable overclock here and my base idle temps are sitting at 17c on the desktop. In the following screenshots you will see how my temps rised and where I balanced off and where my clocks started to be affected. From my initial starting temps my clock speed went up to, and remained at, 2012Mhz with 1.093v all the way up to 36c (it actually hit 2025 in the beginning but for a split second before it was adjusted by the curve)

But after a minute or two at these levels, NVIDIA's crapware decided that these temps were going too high too fast so in it's attempt to maintain a "balance" of sorts, it dropped my voltage to 1.081v which caused a drop of 12Mhz on the core

So now I'm at 2000Mhz with 1.081v, great. Now let's start up the game shall we?
I load into the game and watch as my temps rise 1 degree at a time and take my screenshots with every degree risen along with screenshots of any other changes. So my temps rise up and then I hit 42c where I get the first change by way of voltage jumping back up to 1.093v.

But wait? If my voltage went back up shouldn't that mean that I get back that 12/13Mhz?
The short answer? No. Why? Because at 42c the stupid curve decided yet again that it needed to drop my core clock by another 13Mhz and it decided that was too soon so it though that it would be OK for temps if it let the voltage rise back up to counteract it in order to maintain a so called "balanced" that this curve tries to keep. And so after this my temps balanced out at 45c and 2000Mhz with 1.093v.

So yeah, this is great. I've got 2000Mhz on the core in game, woo-hoo! But unfortunately, the fan speeds are maxed out at 100% and that is not going to fly with me since it is extremely loud and annoying. So let's try dropping our fan speeds and see what changes and how much will change shall we?

Ok, 90%. No change in temps and a tiny bit quieter, but obviously not quiet enough. So let's keep going down in increments of 10% shall we? Next stop: 80%

And what do you know? The moment I go to 46c that curve strikes again! It apparently thinks that at 46c with 80% fans is a little too dangerous so it drops my voltage back down to 1.081v which causes me to lose yet another 13Mhz. Now we are down to 1987Mhz which is below our 2Ghz desire for this card. Yay! So much fun!

So I continue with the dropping of fan speeds and waiting on each drop for the temps to max out and being on the look out for further changes. I end up getting to 50% fans, and at 52c my voltage goes back up again which tells me that I got another drop of speed by 13Mhz that the curve countered with a rise in voltage again back up to 1.093v

From here my temps maxed out at 54c with no further changes.
So I continued with dropping the fan speeds yet again down to 40% and when my temps hit 56c my clocks went down by another 13Mhz putting me at 1974Mhz with 1.093v

And my temps maxed out at 58c.
Going down again to 30% my temps rose and when I reached 62c I got yet another drop in voltage putting me at 1.081v which brought my clock down to 1962Mhz and my temps maxed out at 65c with no further changes.

So now the only thing left in this part of the testing is to drop the fan speeds to their lowest setting which is 25%. And not this meager 5% decrease in fan speeds results in my temps going up to 70c at which point the game froze up. Apparently my +130 offset wasn't stable. But I saw no change when the temps went up to 70c from 65c. So I dropped my offset by 10 points which gave me +120 offset and I continued to game on it along with other benchmarks for over 2 hours with no issues. So now that I fixed that issue I could continue. But a setback has occurred with lowering the offset by 10 points. Doing this results in a loss of 13Mhz on the core in general.

So now I proceeded to find my preferred noise to cooling ratio and found that I like the sound of 40% but am willing to deal with up to 60% fan speeds.

Closing my case up resulted in a rise of nearly 6c in general compared to open case and this resulted in me having even lower clock speeds.

At the end of it all I get 1949Mhz at 40% fans and 1962Mhz at 50% fans. Going up to about 65% I can get another bump taking me up to 1975Mhz but that's a little on the noisier side than I'd like it to be.

So as we can see above and to compare. with an open case (and adjusting for the 13Mhz offset from my overclock)
At:
40% fans = 58c and 1962Mhz
50% fans = 54c and 1974Mhz
with closed case
40% fans = 63c and 1949Mhz
50% fans = 59c and 1962Mhz

That is a 5c difference with having the side panel on my PC vs off of it which caused a 13Mhz drop for both speeds on it's own.

I hope you found this post informative and possibly helpful.
Please share the love and let people know about this and try to get enough people to throw up a stink about it that NVIDIA will fix this super aggressive curve they've placed on our GPU's as of late and give us back the performance we not only deserve but that we were promised in the beginning and additionally paid for. In my case it was $700.

With all of that being said and with these findings I am now 100% certain of myself investing into a custom loop for just the GPU alone in order to get my temps way down low at acceptable noise levels so that I can reap the benefits of this card's true potential. I recommend you all consider doing the same. Peace.

 

[Kr00n]

Hi mate maybe im wrong but i dont think this is any maliciius programmed curve. If you read about electricity and hot temperatures you will find out the next; hot metals lose a percentage of conductivity, in other wordes
the resistivity of most materials increases linearly with temperature, that is why the drop in mhz.
As you say is not point to set a curve to protect the gpu when this didnt reach dangerous temperatures, is just physic.

 

[QwerkyPengwen]

Hi mate maybe im wrong but i dont think this is any maliciius programmed curve. If you read about electricity and hot temperatures you will find out the next; hot metals lose a percentage of conductivity, in other wordes
the resistivity of most materials increases linearly with temperature, that is why the drop in mhz.
As you say is not point to set a curve to protect the gpu when this didnt reach dangerous temperatures, is just physic.

I understand that but that's not the case with GPU Boost 3.0
It actually backs off the clocks based on the temps in the drivers.
My point is that this was not the case when pascal first launched, it wouldn't start backing off your clocks until a much higher threshold and it would back off the clocks at larger intervals than it does now.

nvidia has made GPU Boost 3.0 more aggressive as of late. If you are right and this is all due to electricity and temps and what not and not GPU-B3 doing this then explain why these cards would stay at higher clocks with the same overclock settings at these temps in the beginning?