I would never be playing games without Reflex or a frame rate limiter configured to minimize latency - so the comparison is not a valid one, in my opinion.
Since games will queue up to three frames in advance, by default, it's easy to make it look like DLSS3's latency is negligible in that scenario.
I'll reuse some images from a post of mine a month or so ago.
I was using
Gone Home in this example, but the game is irrelevant.
PresentMon latency stats are at the top-center of the images.
Now, latency is a complex thing and I'm only looking at PresentMon stats here, not end-to-end latency - so this doesn't tell the whole story.
But whatever the game's latency is, you're going to have a couple of frames latency added on top of that by DLSS3 (minimum 17ms at 120Hz).
If we take the starting point of 40ms latency, use Reflex to bring that down to 8ms, and then add two frames for interpolation (17ms) - sure, we end up with lower latency than we started with, at ~25ms.
But my starting point was using Latent Sync for <1ms latency, and now this will be adding at least 17ms.
That's like downgrading my current TV's latency to that of one 10 years old.
Even if you were to start with just NVIDIA's Low Latency mode enabled, DLSS3 would be doubling the latency.
