I actually already discovered the solution for this, but it's still very odd. If you change the refresh rate to 144Hz using the Windows app, the GPU never goes into true idle. But if you use Precision X to do it, it will go true idle.

Eternifity
Hell, you can set it to 120Hz and it will idle, but not 144hz. I don't know it never goes into idle in 144hz, but it totally should. If any from NVIDIA or EVGA sees this, please fix it.

If you think about it, running 1080p at 144Hz is the effectively the same as running 2 1080p monitors at 60Hz plus another at 24Hz (or UHD at 36Hz). So really you're pushing your card into the higher resolution mode which is why it is idling so high as the card is expecting to be pushing a lot more pixels once you're past that threshold and that is much more demanding for the card. 

stalinx20
That is correct. It is a weird "feature" for 144hz. However, can you tell the difference from 144hz to 120hz? I can't.
 
Don't get me wrong, I have 144hz monitor too, but I still don't see the differences. it's just numbers.

+1
 
Your eyes cannot see the difference, they can't work that fast

I'm  seeing all these comments about how 144hz is such a strain on the GPU and how idling is hard. Except this isn't true. And the idea that "it's only 24Hz difference" is stupid. If it's only 24Hz, why does my GPU flip out about it. And I forgot who got it right, but you have to set the refresh rate to 144Hz through anything other than the Windows Settings.

Again, refresh rate has nothing to do with resolution or load on the GPU. The GPU clock down issue at 144Hz refresh rate was a bug in the Nvidia driver and is fixed in the last couple of driver revisions. Install the latest Nvidia driver and the problem should be corrected.

Edit: Nevermind.  Found my problem.

Edit: Phew! I can now post replies longer than a couple sentences. What happened to eVGA's server?

I wasn't talking about how much load on the GPU, just load. Just because you set the refresh rate to 144Hz does not mean the GPU is under load all the time just because the refresh rate is 144Hz.

Yes, it is under load all the time. Just rendering the same thing over and over again is load. Idling at the desktop is load. Providing an image to your monitor 144 times a second is load. Providing an image to your monitor, even if it is an unmoving desktop image is 33.36 Gb/s to your monitor at 2160p @ 144 Hz. Every image is compiled individually (it can never be assumed that the new image will be the same as the last -- you can only determine that after doing all the work), stored in a frame buffer (memory) 33.36Gb/s, timed with the monitor's refresh rate, read from memory at 33.36 Gb/s, and sent to the monitor when the timing is correct. This means that the GPU is computing and transfering to and from its memory at least 33.36 Gb/s of data while "just idle".

It IS LOAD. Look at your GPU usage. Look at your memory controller load. Idle is load. Everything is load. Everything faster and bigger is more load. 2D is a load.

The "feature" of it not clocking down at 144Hz refresh just does not make sense as only 144Hz refresh was affected. Not 120 or 100Hz. Why? If what you say is true about Nvidia purposefully doing that, then why not do the same thing at 120Hz or 100Hz?

NVIDIA used to prevent downclocking at 120 Hz and above for years. Then NVIDIA started making 144 Hz the threshold for not downclocking. GPUs change and NVIDIA's research findings and opinions change. It was something NVIDIA did on purpose and it wasn't a bug. I don't know the exact combination of frequencies or number of monitors which prevents downclocking; and it keeps changing. I do know that for years, 2 monitors at 60 Hz or 1 monitor at 120 Hz or higher caused the card not to downclock. Recently, it was 144 Hz (at what resolution? I don't know) or multiple monitors (how many?) which caused the GPU not to downclock. It was based on NVIDIA's research.

The big part of the downclocking issue is whether the video card's memory can process enough Gb/s in downclocked state. When the GPU is downclocked, the memory and memory controller (GPU core) reduce frequency drastically. The memory bandwidth capability in the downclocked state can obviously not be less than the monitor's Gb/s requirement. Unfortunately, if a card needs to be clocked up, it is going to use a considerable amount of power even if barely utilized. Unfortunately, if the downclocked state has too much of a performance impact, it will need to not downclock and instead run at high clocks even if the load is only a couple percentage of its normal full clock capabilities. Since the GPU has to read and write 33.36 Gb/s to/from memory in the example above, the above example would require a minimum of 66.72 Gb/s or memory bandwidth (since read and write may be syncronous). If the card is only capable of 60 Gb/s (for example) of memory bandwidth when downclocked, too bad, but the card will have to upclock. I know it seems wrong to see the card running at high clocks when only having a 2% load, but unfortunately that is just the way it has to be (in this example). That's the price of having a card which is much more powerful than you always need it to be.

I am almost positive that part of the reason why NVIDIA has been able to downclock more and more at higher and higher resolutions and refresh rates is because they discontinued the unified driver approach. They used to have to provide universal support for older products and we know that aging products wouldn't be able to perform anywhere close to the same level downclocked as current products. Since NVIDIA dropped backwards compatibility, their ability to make more specific changes has increased.