I see this has been modded to work, but I've been reading feedback and experiences about performance and the numbers don't seem to add up for me.

For example:

\*Works great on stalker 2 on 4090

No fg = 170 fps

4x= 290 fps

6x=370 fps

Feels fine looks fine with dlss quality and dlaa\*

This is not working well at all. The performance cost here is absolutely insane and not \working great\. This received number of up votes as well.

Do 40 series owners just think this is normal because of no prior experience? Are latency values actually in check?

I see stuff like this (and other posts) and I get the sense it has a rather prohibitive performance cost that is actually nothing like Blackwell.

Or has it been fixed? Does anyone have a comparison between say a 4080 and 5070Ti to show performance regression and latency?

Here are some tests. I'm not from Poland and can't read Polish, but the numbers are still easy to understand. https://ithardware.pl/testyirecenzje/dlss\_mfg\_na\_geforce\_rtx\_40\_test\_recenzja\_opinia-53764.html

Are you sure 170 isn‘t with 2x? Stalker 2 shouldn’t run that fast without FG It would mean:

85 base 2x
72,5 base 4x
61,6 base 6x

Math would check out.

(M)FG isn't as effective on higher FPS, as the frame generation adds a mostly fixed processing time to the render time, so this wouldn't be out of the realm of possibility for cards that aren't designed for MFG.

For example in the MW4 Beta it didnt matter if I went from 160 non-FG fps or 210 non-FG fps, both ended at roughly 500fps with 6x fg

Looka like the 4080 is doing better at lower res and falls in line with the 5070ti at higher resolutions judging by the cyberpunk graphs.

Are you sure it’s 170 FPS without FG, or is it without the mod, which would be 2× FG on a 4090 which would then sort of line up very neatly here, or at least neatly enough to say that the rest of it is computational cost of doing more frame generation.

Im playing stalker 2 yesterday with mfg 4x. works great, but reflex dont cap my frames anymore.

rtx 4090

Can't get this to load.

It works fine if you already have a high base frame rate. Latency is fine but you can visually feel its not as what the fps counter says

Nah it's copy and pasted from the reddit post.

I got it working in Cyberpunk 2077 with DLSS 5 and path tracing on my 4070. Base FPS was in the 30s which is atrocious I know, but with frame gen 2x it actually wasn’t bad and felt somewhat stable. I wouldn’t play it like this, but for image quality comparisons and running around the city with a controller it’s fine.

Turing on MFG didn’t really increase fluidity. Higher framerate but there is something going on with the frame delivery that doesn’t feel as smooth. I’m still in the 30s for my base FPS so it’s not a huge drop from 2x but something is definitely off.

Honestly though I need to test it in a game with a much higher base FPS. The shaky frame times might also be from taxing memory bandwidth with all of the advanced features, so that probably comes into play.

Overall though with the way it felt it’s not something I’d actually use.

4090, tested on stalker two in 4x with DLAA.. 4x is great IF you can maintain 60-80fps native. The input lag is tolerable. I run ultra plus at the highest settings, so there are areas where the real FPS plummets. The latency skyrockets and the input lag is obnoxious. I just left it at 2x. I've used it on slightly less demanding games and it does work well.

3x is also very usable. You don't need to use 4x. I rarely do.

That said, I run Stalker with 2x because frametimes aren't amazing and it helps mitigate OLED VRR flicker.