Emm, you do realise that there is no input possible for generated frames so if real frame generation is 30FPS but with DLSS3 it's 60FPS, input is still only 30FPS or 33.3ms.
Nvidia Reflex or not, input is capped at actual real framerate, so yeah lag!
GPU bound 30-35 FPS will have an enormous latency around 120-130ms by default
reflex by itself cuts its to 40-60 ms range. and DLSS3 add some on top of it, and you still end up somewhere between the middle.
GPU bound 60 FPS will also have more latency than
reflex-ed 30-35 FPS, in most cases
you're simply disregarding GPU bound input latency which adds immense amount of additional input lag, which Reflex solves. Reflex is not magically reducing input lag out of thin air. its main target is the GPU bound pressure latency that existed for a long time. most people ,AMD and NVIDIA users included, played with this heavy GPU bound pressure latency for years without any problems. the people who did have some problem with latency often used frame caps to reduce it.. you can cut down this GPU bound input lag by introducing a framecap that pushes GPU utilization below %90 but there will be scenes that will demand more GPU power and end up causing input lag again. naturally, it wasn't an elegant solution and it only worked if your GPU was super powerful for a game so you could always know that your GPU will always have plenty headroom for your chosen framecap
here's an actual case analysis
witcher 3 at 66 FPS with GPU bound settings have 70-75 ms of input latency
and GPU bound 34 FPS has an enormous input latency around 130 ms (as I predicted)
Reflex massively cuts its down to 50 ms ranges
So in the end, Reflex'ed 34 FPS has less input latency than non-Reflex'ed GPU bound 66 FPS.
You can always have even lower latency by using Reflex on top of GPU bound 66 FPS, that's true. But that wouldn't be an option if NVIDIA never introduced Reflex to begin with. If you were fine with GPU bound 70+ ms of latency at 60-70 FS, you should technically still be fine at 40 FPS with Reflex that is being interpolated to the same 60-70 FPS. because they both still have literally the nearly same input latency. if you however feel entitled to the new low input latency you can achieve at actual 60-70 FPS, then it is just that, entitlement. but here's what you get if you Reflex the GPU bound 65-70 FPS. you get the peak lowest latency experience possible.
NVIDIA,
if they wanted, could've opted out to give Reflex to everyone and only makes it so that it can be enabled alongside with DLSS3. if they did that, DLSS3 would always come out with positive input latency gain and no one would understand how it happens and none of this discourse would take place.
So by default, people played at 50-80 FPS for years that had actually higher input latency than a 30 FPS cap would. as I said, you don't need Reflex. here's the same with 30 FPS cap without reflex;
30 FPS cap, no reflex (even lower latency than reflex-ed 34 FPS)
and Reflex does nothing to the input latency, as much of the GPU bound pressure has been already beaten with the frame cap;
So in short;
Input latency chart (highest to lowest)
Low framerate no reflex > High framerate no reflex (default experience) > Low framerate + DLSS3 + reflex > Low framerate reflex > > High framerate reflex (20 ms)
I'm not bringing DLSS2 to any of this equation latency reduction coming from DLSS2 is simply the magic of having higher framerate.