When multiple people tell you that you misunderstood and explain how and why, then post the evidence to back it up say that they're wrong because they posted a link to the definitive source of information on that topic?
No man, that's all you. You do not understand this topic or how a conversation works. Make a claim, and back it up with proof - or don't make that claim.
GPU bottlenecks result in uneven frametimes and render latency, as the CPU renders ahead while it's waiting. Frametimes and FPS are the same thing expressed in a different way. 1000Hz in a second, divided by the FPS... thats a frametime. This is not the same as render time.
CPU bottlenecks result in microstutters at worst, or a lower ceiling cap to max FPS. That said, you're going to have lower input latency in this situation.
Variable refresh rates allow the system to enter new information in a shorter period - If you had a 240Hz display with a VRR range of 1Hz to 240Hz, a 1FPS signal is still updated in the speed the 240Hz is - meaning that a new frame can be displayed at any of those 240 incremental steps, as soon as it's ready. This allows a much faster recovery from any stutters, and offers the reduced input latency of the maximum refresh rate even if the frame rate is lower.
The only reason people use uncapped framerates is because some older game engines like CS gave a competitive advantage to high frame rates as an engine bug, combined with reducing the input latency of a maxed out GPU. VRR provides that reduced latency with Vsync on, and an FPS cap gives you all the benefits without needing anything.
You'd know all this - If you weren't too lazy and arrogant to click a damned link.
It takes 10 seconds to fire up nvidias monitoring stats and see this in real time, it's not rocket science.
These are render latency values. These are from nvidia at the GPU stage of the pipeline only, before the monitor is involved.
Every ms of render latency is a delay in you seeing the image, which means you are responding to an older image. It's not input latency from your input device, but it IS an added delay to you reacting to what is
actually happening.
60Hz, Vsync on. Oooh yay i love 31ms of input lag!
View attachment 291088
Exactly the same but with a 60FPS cap that prevents the CPU from rendering ahead of the GPU. (2 frames in DX12, so 60 +2 rendered ahead by the CPU, just like if the GPU was maxed out)
5.6ms.
Let's not use an FPS cap and enjoy 5.6x more render latency
(There is some beauty in that it's 5.6ms, and that vsync was 5.6 times higher)
View attachment 291089
"BUT I WANT VSYNC OFF AND MAXING THE GPU IS BETTER BECAUSE THATS HOW I BENCHMARK"
View attachment 291091
Sure, we go from 59FPS to 151 on the those 99% lows but... oh wait the input latency doubled.
letting my GPU max out its clocks to get that maximum performance managed to... still be worse.
View attachment 291093
Freesync/Vsync is a whole nother bag of fun on top of this, because freesync behaves differently between AMD and Nvidia, and Gsync is nvidia exclusive and different again.
The main key is when they work properly (Vsync off and maxed out GPU's is not properly) they can update at divisions of the full refresh rate.
Using Samsungs 240Hz displays as an example here, they have a minimum of 80Hz because it's 1/3 the max refresh rate.
At 1/2 or 1/3 of a supported refresh rate, the highest rate is used - and the signal sent for a new frame is sent at the start of the final repetition, not at the start. So 80Hz is 12.5ms, 160Hz is 6.25ms and 240Hz is 4.166ms
80FPS at 240Hz VRR would give you better input latency and response times because it asks for the new frame at the start of the final duplicate - 4.166/4.166/
4.166
In a GPU limited situation you lose that benefit with frames rendered ahead end up right back at the higher latency values anyway.
Running 235Hz and FPS would give you 4.255ms input latency. If your GPU was maxed out and the CPU had to render ahead to cover up the lack, you'll end up at 8.51ms or 12.765ms, despite being at the higher framerate
Unlimited framerates are not a positive. Maxing your GPU is not a positive. They need spare performance ready to render the next frame when it's needed, because if they're even 0.000001 milliseconds late, you're getting doubled or tripled input latency and that's the most common microstutter there is, as input latency values go crazy.
Doing stupid things like running the GPU at 100% all the time forces you into that higher latency state all the time and people mistake that consistent latency for being the best they can get.
