More DLSS...

[LukeTbk]

statement like those will be game and target resolution dependent (lot of those people could be testing at 4k, where performance is still a 1080p native res), if you test on a 1440p you would start at 720p

 

[XoR_]

I keep seeing this claim over and over saying that DLSS 4 can run at performance and look just as good as it did on DLSS 3 at quality. Maybe that's true in most of the games they are testing but I tried it in the Last of Us part 1 and even balanced looked like pure shit in movement. Like the area I was testing in I could actually see shimmering and a line being drawn anytime I used below quality DLSS 4. Going back up to quality completely cleared up that line that was being drawn in front of me walking and removed 90% of the shimmering.

Are you sure you used proper dll and dlss profile?
Or the game has patch?

 

[III_Slyflyer_III]

I keep seeing this claim over and over saying that DLSS 4 can run at performance and look just as good as it did on DLSS 3 at quality. Maybe that's true in most of the games they are testing but I tried it in the Last of Us part 1 and even balanced looked like pure shit in movement. Like the area I was testing in I could actually see shimmering and a line being drawn anytime I used below quality DLSS 4. Going back up to quality completely cleared up that line that was being drawn in front of me walking and removed 90% of the shimmering.

Out of curiosity, was it using preset "J"? I believe it is supposed to with the new DLSS automatically, but I also read not all games will and you have to set it with nvidiaprofileinspector. Make sure you copied ALL DLSS .dll files properly. You need 2 or 3 of them depending on the game and what it supports.

 

[MistaSparkul]

Out of curiosity, was it using preset "J"? I believe it is supposed to with the new DLSS automatically, but I also read not all games will and you have to set it with nvidiaprofileinspector. Make sure you copied ALL DLSS .dll files properly. You need 2 or 3 of them depending on the game and what it supports.

The new driver is supposed to drop tomorrow so you'll be able to just select Transformer model directly from there.

 

[jobert]

Out of curiosity, was it using preset "J"? I believe it is supposed to with the new DLSS automatically, but I also read not all games will and you have to set it with nvidiaprofileinspector. Make sure you copied ALL DLSS .dll files properly. You need 2 or 3 of them depending on the game and what it supports.

The Last of Us part 1 only uses DLSS super resolution so there's only the one dll file needed from what I understand. That game does not have frame generation or ray reconstruction.

 

[KickAssCop]

Btw I wanted to make a long ass post about DLSS and FG etc. etc. as I had been advocating using them all the time at 4K. Well guess what. I started looking for artifacts and latency especially in the games I am playing a lot these days (Stalker 2 and BO6). And holy shit balls. The omgrafix leave a lot to be desired. I remember last time finding this crap with Horizon Forbidden West and Alan Wake where turning on (DLSS) such features led to graphics degradation. Avatar also had some issues. I had already found DLSS artifacts in Stalker and reverted back to TAA (there is no other worthwhile option) but BO6 is a big difference. I am now using what they call nvidia image scaling in BO6 which looks much better and runs slightly slower. My PCL (as per frame pacing app) is down to 15-20 MS vs. 30-40 MS using FG and DLSS Quality in BO6.

So thanks guys for ruining it for me (again). I will now revert to using native or some other hacks which look better than exclusively DLSS. FG is also a bit off the table. What does this all mean for me? I probably need a 5090 pronto.

 

[III_Slyflyer_III]

Btw I wanted to make a long ass post about DLSS and FG etc. etc. as I had been advocating using them all the time at 4K. Well guess what. I started looking for artifacts and latency especially in the games I am playing a lot these days (Stalker 2 and BO6). And holy shit balls. The omgrafix leave a lot to be desired. I remember last time finding this crap with Horizon Forbidden West and Alan Wake where turning on (DLSS) such features led to graphics degradation. Avatar also had some issues. I had already found DLSS artifacts in Stalker and reverted back to TAA (there is no other worthwhile option) but BO6 is a big difference. I am now using what they call nvidia image scaling in BO6 which looks much better and runs slightly slower. My PCL (as per frame pacing app) is down to 15-20 MS vs. 30-40 MS using FG and DLSS Quality in BO6.

So thanks guys for ruining it for me (again). I will now revert to using native or some other hacks which look better than exclusively DLSS. FG is also a bit off the table. What does this all mean for me? I probably need a 5090 pronto.

Have you tried using DLSS 4 .dll in Stalker 2? It looked WAY better to me when I tried it, using DLAA and FG. As for BO6, I just use DLSS Quality at 4K for frames, if I play SP or Zombies I'll go DLAA and FG. I don't notice these artifacts I guess with such fast-paced games.

 

[KickAssCop]

Have you tried using DLSS 4 .dll in Stalker 2? It looked WAY better to me when I tried it, using DLAA and FG. As for BO6, I just use DLSS Quality at 4K for frames, if I play SP or Zombies I'll go DLAA and FG. I don't notice these artifacts I guess with such fast-paced games.

Nope. I will use it when driver is out.

 

[III_Slyflyer_III]

Anyone having luck with the new driver and nvidia app combo? All my games that SHOULD be supported according to nvidia still say unsupported, even after following the instructions.

Guess I'll have to keep doing it manually for now? Ugh...

 

[kalston]

Yea the nv app override is a bust and barely works, on only limited titles. Still have to rely on third party tools.

 

[deceptXS]

Preset K that is in the new driver looks pretty good. It's less over-sharpened than Preset J. Overall it's an improvement, I had switch between J and K a few times to really start noticing the differences, but after looking at finer details, I found myself preferring Preset K . It also improved the faces of distant NPCs (e.g., party members in FF7 Rebirth), they have slightly less distortion. It's not a massive difference like going from CNN to Transformer, but it is an improvement.

 

[KickAssCop]

What is the preset J and K that you speak of. I really need to play some games.

 

[deceptXS]

What is the preset J and K that you speak of. I really need to play some games.

DLSS-SR presets.

 

[Captain Newmackwa]

Anyone having luck with the new driver and nvidia app combo? All my games that SHOULD be supported according to nvidia still say unsupported, even after following the instructions.

Guess I'll have to keep doing it manually for now? Ugh...

There's a whitelist for the games that are currently supported:

https://www.nvidia.com/en-us/geforce/news/nvidia-rtx-games-engines-apps/

Nvidia claims that they will be adding support as more games have been validated by them.

That said, doing it manually without the NV app is still an option for those who don't want to wait for support or those who don't want to install NV app altogether.

 

[XoR_]

There's a whitelist for the games that are currently supported:
https://www.nvidia.com/en-us/geforce/news/nvidia-rtx-games-engines-apps/
Nvidia claims that they will be adding support as more games have been validated by them.
That said, doing it manually without the NV app is still an option for those who don't want to wait for support or those who don't want to install NV app altogether.

There should be soon option for nvProfileInspector to enable switch in driver for given game.
I would not touch this for online games with anti-cheat like e.g. Fortnite because of possible ban. Single player games are free for grabs using any method - which there a few.

Myself tested Control - had no luck with anything related to swapping DLSS versions and here it just worked. At least for 3 minute test I think it worked and I have less artifacts at Performance preset than before Quality.
Wanted to check Fortnite and game is unfortunately unsupported. I hope people will push Epic to update DLSS.

Also nice is that now we can force DLAA without needing to deal with any external tools.
Control does not have this option but I will be trying it when nvVidiaProfile injector has option to override in profiles what is supported functionality.

 

[Captain Newmackwa]

There should be soon option for nvProfileInspector to enable switch in driver for given game.
I would not touch this for online games with anti-cheat like e.g. Fortnite because of possible ban. Single player games are free for grabs using any method - which there a few.

Myself tested Control - had no luck with anything related to swapping DLSS versions and here it just worked. At least for 3 minute test I think it worked and I have less artifacts at Performance preset than before Quality.
Wanted to check Fortnite and game is unfortunately unsupported. I hope people will push Epic to update DLSS.

Also nice is that now we can force DLAA without needing to deal with any external tools.
Control does not have this option but I will be trying it when nvVidiaProfile injector has option to override in profiles what is supported functionality.

I just tested the NV App DLSS SR override in Battlefield 2042 and it seems to work without triggering EA Anti-Cheat.

Disclaimer: I tried this at my own risk. YMMV.

 

[KickAssCop]

I tried the new drivers (assuming the DLSS is updated). No real changes in BO6 or Stalker 2.
Btw if you want to see what DLSS or FG does to image quality just try it on BO6 menus. You will see that the edges of many things are so blurred that you can't even see them. Maybe I can take screenshots and post here for reference. I did not know that this was happening as I have been using DLSS for almost 2-3 years when it came out. But generally the IQ is not comparable in many games and the blurring is quite annoying. Even with sharpening it does not work correctly. Another thing I noticed especially in Stalker 2 are the UI artifacts. They are so annoying and appear as moving white spots around the UI.

When it works it is great and some games you can't play without FG (e.g., Cyberpunk) but you need to be able to make the IQ compromises. Transformer model is a step in the right direction for the most part.

 

[elvn]

You may get somewhat different results depending on things like

native resolution and panel type (e.g. 1440p vs 4k native, DLLS performance 4k vs DLSS quality 4k, VA vs OLED)
and
what your average native fpsHz is (how much has changed over how large a period of time between frames).

if you want to see what DLSS or FG does to image quality just try it on BO6 menus.

From what I read in the nvidia pdf's that help developers intergrate dlss into their games, Nvidia outlines how devs are supposed to turn DLSS off when in menus, at least Frame Gen, (DLSS-G)

When to disable DLSS-G

• Temporary Events (may retain resources, see below):
  • A fullscreen game menu is entered
  • A translucent UI element is overlaid over the majority of the screen (ex: game leaderboard)
• Persistent Events (must not retain resources):
  • A user has turned off DLSS-G via a settings menu
  • A console command has been used to turn off DLSS-G

. . . . . .


Nvidia devs have said they are working on improving DLSS so that it will be better able to identify what *not* to apply itself to, e.g. animated textures (like an animated gif) on simulated video screens in games.

. . .


Catanzaro said in that DigitalFoundry interview about how far DLSS has come with things like skin on people, etc :

"Some of it's just polish, we've had another year to iterate on it and we're always increasing the quality of our data sets, we're analyzing failure cases, we're adding those failure cases to our training sets and our evaluation methodology".

"Part of it is algorithmic development but the new model is much bigger and being much more compute in it ... it has much more capacity to learn, it's more intelligent"

"A lot of times when we have a failure in one of these DLSS models, that it looks in game like shimmering, ghosting, blurring - right? we consider those model failure. The model's just making a poor choice, right? - It needs to be smarter"

.
On animated textures (emphasis mine) :

"I noticed on the B-Roll, Rich was looking at animated textures, which have always really bothered me. You know, it's a really tricky thing for DLSS super resolution or ray reconstruction to deal with animated textures because motion vectors from the game that are describing how things are moving around don't go along with the texture, right? It's just a television, it's just sitting there, you know, and yet you don't want the screen on the TVs everywhere to just be blurring as stuff moves around so that requires the model to kind of ignore the motion vectors that are coming from the game. The model has to basically analyze the scene and recognize that "Oh! This area is actually a television with an animated texture on it. I'm going to make sure to not blur that". It was really hard to teach the prior CNN models about that. We did our best, and, we did make a lot of progress, but I feel like this new Transformer model opens up a new space for us to solve these problems".


. . . .

Personally, I think there might be potential there for game engines+devs to tell DLSS what to ignore in the future, and inform it better overall.

. . . .

I think it will get a lot better as time goes on. It may also have differing results on different games, by different game devs, though those may also improve over time like patching games tends to do (at least for devs that support their games well over time). Hopefully it will start to become more standardized and more integrated in future game development.

 

[KickAssCop]

It is already better than before but I am noticing some bugs/issues in some of the newer games.

 

[XoR_]

When it works it is great and some games you can't play without FG (e.g., Cyberpunk) but you need to be able to make the IQ compromises.

FG does nothing to fix biggest issue with low frame rate: responsiveness.
It only makes the issue worse because we don't have perfect scaling and we stall displaying last generated frame <- which means that by the moment GPU starts displaying that last real frame game already generated next frame unlike how it works without FG where GPU will display next frame as soon as it was generated.
In result you always get much laggier experience.

The question thus is: how high should be your FPS to for accounting for scaling and display lag enabling FG doesn't make game visibly more laggy.
Answer I derived from my testing is: way too high fps for FG to be useful on my 'just' 360Hz monitor.

For 2x scale I can at most get 360fps so 180fps real frame rate.
In this case FG might be okay to use - lag after FG isn't big so game is still playable and visibility of artifacts is small enough to be too hard to notice.
The issue again is: if I can push 180fps real frames with FG I can push 240+ frames without it. In this case using FG is pointless because smoothness difference between 240 and 360 is not that noticeable. Lag-wise it is very different experience between even native 180fps and FG'd 360fps let alone if we take scaling in to consideration so more like native 240fps (note: may be more or may be less but should definitely be more than 200fps).
Also we don't talk about playability but merely experience. Game can be playable at given settings but still could be improved in such a way that it plays better. Getting 360fps through FG is in this case worse setting than getting 240+ native. 360fps though FG would look slightly more smooth though.

And even if we had +100% fps scaling we still talk 180fps - it is silky smooth already. More than 180fps if we can use VRR. And even at 180fps we still get less lag without FG than with it.
The only exception would be if I had to drop refresh rate to 180Hz and did it naively (using default CVT-RB without correcting for pixel clock) then... actually it would still slightly beat 360Hz with FG.

If I lower my target frame rate it doesn't really help. There is no range where FG makes sense on my monitor.
Maybe if I had 480Hz monitor... nope, if I push 240fps (so 300+fps without incurring FG processing tax) I care for smoothness even less.
All that better monitor does is that lag is less of an issue while visual benefits from FG are less and less.

tl;dr
I totally don't agree it can ever happen that FG can help game moving from 'unplayable' to 'playable' category.
It is not real frames and doesn't fix real issue with low frame rates: high latency.
All that FG does is change how game motion look from "cinematic" style to "soap opera" style.

Also BTW if you make such claim please provide what fps you have with it and what without.

p.s. I haven't even mention in this post yet that without FG you have frame to frame consistent reaction of game to mouse input compared to with FG game reacting differently to frames depending on if its real or fake frame.
It might not be too relevant at high frame rates as when I boost use FG on ~30fps but these things really scale with frame times. This effect is there no matter if I boost frame rate from 30fps to 60fps for 60Hz monitor as when I boost it from 180fps to 360fps on 360Hz monitor.
And complaining about lag or latency inconsistency on ~180fps might feel like bonkers but in real life when playing multi-player games I still notice big difference between 180fps and e.g. 240fps and between 240fps and 360fps.

p.s.2. Of course you might be bothered more by "cinematic" style of graphics (with cinematic here meaning how motion looks at lower frame rate) than increased input lag and it would still be an improvement for you to use FG.
I am just talking about your claim: that FG can help make game playable.
Unless maybe your claim is that visually having more cinematic motion is making the game unplayable faster for you than game having slow and inconsistent response to inputs. In my nervous system this is inconceivable but I can try to imagine it (via not imagining it and instead just using pure textual thought processing ) can be different for different people and you might be totally fine with input lag of 120fps on 120Hz monitor but somehow are bothered by 60fps motion (or however GPU can push without having to generate frames) or whatever fps you boost to whatever fps.

 

[KickAssCop]

Dude I can’t read long posts.

 

[XoR_]

Dude I can’t read long posts.

Yu have very short context length in your mind and no ability to increase it.
Noted.

 

[pendragon1]

not everyone can be bothered to read a wall of text and your tldr is just a long as the "i aint reading all that" section, in response to one line. tighten it up a bit, eh.

 

[elvn]

I read it. It took me about 1 minute to read lol.

I appreciate your perspective XoR, I just don't agree with it overall.

In today's crowd, I think a lot of people are reading on phones too, where you get like 3 words per line, so each simple paragraph becomes a scroll to the floor. Twits and BlipTocks ruined a lot of people.

. . .

I think FG is still young and that it will get better. I think 10ms input lag or less is decent, and that competitive online gaming servers aren't even showing you where things are to aim at temporally as far as the server is concerned so advantages are mostly marketing in very high fps hz as far as online gaming advantages go. The lowest peeker's advantage ~ rubberband you can get on a 128tick server is 72ms, if you are at 128fpsHz minimum (not average), where a 60fpsHz player would get 150ms. So there is some advantage to that point. However, as most realize, there are a ton of online games with woefully low tick rates besides.

I can still appreciate the desire for the extremely low input lag feel though, even aesthetically and sort of "ergonomically" in a way, just like I appreciate very high fpsHz's benefits of much lower sample-and-hold-blur of the entire viewport, approaching a non-blurred look at 480fpsHz to 1000fpsHz depending how fast the viewport is moving. That and the greatly increased motion definition, motion articulation/pathing, and even increased animation cycle definition which probably doesn't hit diminishing returns until at least 480fpsHz.

I don't want to keep repeating, but in my opinion, this is the future and eventually when boosting say 100fps minimum (120fps to 130fpsHz average depending on the game perhaps), FG +3 and even higher when it gets more intelligent, will be able to exceed a peak Hz of a very high Hz screen. With 100fps minimum graph x4 via FG, that might mean capping a 480fpsHz screen to 400Hz. The system may eventually cap the Hz automatically even, potentially. That might still apply at 100fps minimum vs 1000Hz screens someday, to avoid overages. Anyway when FG gets that good and it's making multiple frames like that on 480Hz to 1000Hz screens, it will be exceeding the cap Hz of the screen so VRR won't be needed anymore. As far as the monitor is concerned, the frame rate is never changing. That while aesthetically the FoV movement is massively more crisp vs blur and the motion defintion/pathing and animation cycles are are much higher. A 120fpsHz graph might be something like 100fps <<< 120fps >>> 140 fps, so the frame durations would be 10ms <<< 8.3ms >>> 7.14 ms.

If using VRR at lower frame rates, your input lag is going to be tied to your frame durations, and even if using low latency tech (like that available for consoles on TVs), you aren't actually seeing any new information from a game to react to. Low latency tech typically allows you to react sooner, twice as fast locally but you aren't seeing anything twice as soon to react to. Same goes for frame gen too as far as online gaming servers are concerned, the local frames aren't on the same clock or under the same judgement until you get the next tick. In fact, even without FG, your local machine is predicting frames while it waits on the server's next tick. They're just on your native rendering pipeline rather than AI generated.

. . . .

In order to get the lowest peeker's advantage~ rubberband on a 128tick online gaming server, you need 128fpsHz as your minimum.

In order to get better performance from frame gen, and going forward into more multiples of frame get (+3 to +9, x4 to x10), it might turn out that you'd need to be at something like 100fps minimum / 120fpsHz average natively.

In order to get reasonable input lag, some might consider that at around 100fps minimum for 7.14 to 10ms frame durations.

Despite the marketing of frame gen and high hz screens, it might end up that you can't get blood from a stone.

They are working on better input lag tech for DLSS though. I think there will be some growing pains but that it is the way forward as we get very high Hz OLEDs in the future.

 

[KickAssCop]

Www.chatgpt.com works wonders to cut words.

This person is discussing the future of gaming performance, focusing on input lag, frame generation (FG), and high refresh rate screens. They argue that while high FPS and low input lag are important, the actual competitive advantage of high FPS may be overstated due to server tick rates and the nature of online gaming. They believe that FG and higher Hz screens will improve gaming aesthetics and performance, but there's a limit to the benefits, and better input lag technology is needed for further advancements.

So basically sky is blue?

 

[elvn]

The devil is in the details. Besides stripping my reply of the personal interaction and throwback to positions XoR holds - that AI compressor omitted actual rates suggested/required mentioned my reply, as pertains to input lag, online gaming, and frame gen - which are very pertinent to the dialogue. In fact, after the four dots, it was already made pretty concise for the more A.D.D. addled readers, and the AI response relayed none of that information.

Beep Boop.

 

[LukeTbk]

I think there is a bit of a semantic trick going on, about the FG at 30 fps is useless.

The question could be asked (and in blind test answered has yes) can the subjective experience of some player at 30issh something -> 60 fps via frame generation better than keeping it at 30iss something, and maybe what people fear about the tech, console game dev starting to use it to be able to say 60 or 120 mode on the game box.

With a reflex 2 type of tech the added latency could become reduced latency for your camera and the low FPS when using framegeneration will be a double case of the visual gap between 2 frames is so large that it become really hard to generate that information, with Transformer DLSS type doing it on 7000 series that will have the 7060 featuring 2,800 tops of 1.5 bits it could quickly become better than playing at 35 fps in all regard for many.

 

[Sycraft]

tl;dr
I totally don't agree it can ever happen that FG can help game moving from 'unplayable' to 'playable' category.

Well... there is one possibility that might make it happen and that is the whole reflex 2/frame warping thing. I'm not counting on it or saying it happen but if they can get the warp to happen fast for each generated frame, it could make it feel much more responsive by lowering the latency between what you are doing and what you see. The LTT guys did a bit on it a couple years ago with a game engine/demo that could do asynchronous reprojection (not precisely the same thing but similar idea) and the results were pretty good. So I'm not going to write it off as impossible, they may find a way to quickly warp the frames enough that they can cut the perceived input latency, and if the results are good enough, then cool.

Still, I feel like overall it is a tech more suitable/targeted for high refresh rate monitors. With the new scalers coming out I imagine we'll see 4k 480Hz monitors soon, and it'll likely just keep going from there. Well, there's no way we are going to have games that have shiny high end graphics and also manage to crank out that framerates at 4k. We are going to need FG to take advantage of the smoothness a display like that can offer.

 

[noko]

If I take a shot on a small fast moving target, using reproduction or Reflex 2 on a generated frame, the game engine only responds on actual generated frame states where the target would be off from my aim. I do not see how that would reduce meaningful latency. It would seem to introduce errors in the mechanics of the game. I see it maybe reducing motion latency but not input shot point, unless games actually interprets in between positions from mouse input.

If reacting to what is on the screen, with 4x FG, roughly 75% of the time one would be responding to a generated frame state. Plus would not small variations in actual frame rate cause up to a 4x variation in the displayed frame rate? This too can cause inconsistent feel for the timing on interactions in a game.

I just have to experience it to see subjectively the experience. Plus a 120hz TV has made FG DOA for most games and pointless from my experience. FG on AMD with 120hz, lol is not a good experience.

 

[Sycraft]

If I take a shot on a small fast moving target, using reproduction or Reflex 2 on a generated frame, the game engine only responds on actual generated frame states where the target would be off from my aim. I do not see how that would reduce meaningful latency. It would seem to introduce errors in the mechanics of the game. I see it maybe reducing motion latency but not input shot point, unless games actually interprets in between positions from mouse input.

The idea with reflex 2 is it would respond to where your cursor was when you triggered the shot, not what was on the generated frame, hence reducing perceived latency.

Also something to remember when talking about this stuff: Not everyone plays high speed shooters, in fact I'd go as so far to say as most people don't play them most of the time. This tech isn't necessarily targeting them. There's nothing stopping a company from making a game with fairly simple graphics so it can render at high frame rates natively because people want to play a twitch shooter. However for other games higher fidelity graphics can be nice and this kind of thing can help FPS.

The thing to keep in mind is that it is not a choice of do this or make a card that is faster. They are pushing cards as hard as they can. The 5090 draws 575 watts for fuck sake. We can't just say "We want 4x the FPS, just make a card 4x as powerful!" If it was easy, and nVidia refused to do it, someone else would. The reason this is being done is we are hitting limits and we need clever ways around them.

 

[noko]

The idea with reflex 2 is it would respond to where your cursor was when you triggered the shot, not what was on the generated frame, hence reducing perceived latency.

Also something to remember when talking about this stuff: Not everyone plays high speed shooters, in fact I'd go as so far to say as most people don't play them most of the time. This tech isn't necessarily targeting them. There's nothing stopping a company from making a game with fairly simple graphics so it can render at high frame rates natively because people want to play a twitch shooter. However for other games higher fidelity graphics can be nice and this kind of thing can help FPS.

The thing to keep in mind is that it is not a choice of do this or make a card that is faster. They are pushing cards as hard as they can. The 5090 draws 575 watts for fuck sake. We can't just say "We want 4x the FPS, just make a card 4x as powerful!" If it was easy, and nVidia refused to do it, someone else would. The reason this is being done is we are hitting limits and we need clever ways around them.

How would the game know where the cursor was? FG is basically working in previous 2 rendered frames, meaning anything you are reacting to is a past position while your input is current unseen state. Game records and acts on your input in state you are not observing. Anyways the game could use a larger hit box where you are not really on the target but registers as a hit.

 

[LukeTbk]

FG is basically working in previous 2 rendered frames,

Say you realFrame 1 happening at 50ms, realframe 2 happening at 100ms, trying to show the generated frame 1-2-3 at 62-74 and 88 ms

If the mouse moved at 55 ms the generated frame generated to be showed at 62ms will take that movement into account, it is explained here:
https://www.nvidia.com/en-us/geforce/news/reflex-2-even-lower-latency-gameplay-with-frame-warp/

It will not be a full game lower latency, just things that affect the frame rendered via a mouse camera position change lower latency

 

[Sycraft]

How would the game know where the cursor was? FG is basically working in previous 2 rendered frames, meaning anything you are reacting to is a past position while your input is current unseen state. Game records and acts on your input in state you are not observing. Anyways the game could use a larger hit box where you are not really on the target but registers as a hit.

Same way it always does: By where the mouse input said it was going. The idea with Reflex 2 is you sample the mouse input at a much higher rate than you render (this is normally done anyhow) and you take where the input is as close to that moment as possible, warp the display to match, fill in the blanks, and show that to the person. Well given the game engine is sampling the cursor position it can also sample button presses. So it can know when you fired and where you were aiming.

We'll see how well it works in reality, but it is a solid idea in theory. The simulation and the output frame rate don't need to run at the same rate (and often don't in many engines). You can track position of camera and objects far more often than you can render an entire scene.

 

[noko]

Say you realFrame 1 happening at 50ms, realframe 2 happening at 100ms, trying to show the generated frame 1-2-3 at 62-74 and 88 ms

If the mouse moved at 55 ms the generated frame generated to be showed at 62ms will take that movement into account, it is explained here:
https://www.nvidia.com/en-us/geforce/news/reflex-2-even-lower-latency-gameplay-with-frame-warp/

It will not be a full game lower latency, just things that affect the frame rendered via a mouse camera position change lower latency

Very cool, will have to see how well it actually works. The SDK plugs into the game/engine for reflex 2 so input appears to indeed be associated with what you aimed at in the frame you clicked on.

 

[XoR_]

Does anyone know if Reflex2 is already designed to generate more frames or just shift single last generated frame?
Generating more is such a no-brainer that it is how frames should be generated.
In this case we talk seeing 100% generated frames and perhaps more susceptibility for artifacts - but still hundreds times better way to do frame generation.

Not sure how such FG would handle 30fps artifacts-wise and with issues like generated frames not matching game state but still would be much more usable than having input lag of 30fps multiplied. Especially with multi-player games 120+fps is usually enough as even sweats don't change directions that often. It is just much easier to aim when game reacts instantaneously to your mouse movements.

If I take a shot on a small fast moving target, using reproduction or Reflex 2 on a generated frame, the game engine only responds on actual generated frame states where the target would be off from my aim. I do not see how that would reduce meaningful latency. It would seem to introduce errors in the mechanics of the game. I see it maybe reducing motion latency but not input shot point, unless games actually interprets in between positions from mouse input.

I think that even if no considerations were taken to this aspect of aiming at game's side it would still be helpful for aiming for online games.
In single player games it should be non-issue unless perhaps running very low frame rate like 30 and below.

And we should not make games which can only run 30fps even with upscaling.
Biggest issue with current FG is that when you move mouse it all feels floaty and unpleasant. If you only play with this lag and with FG you get used to this lag and this response of mouse and your brain will compensate but if you play at good frame rate and then move to FG it feels even more unplayable than without FG. At least I didn't feel it improved anything at any frame rate except made game look more smooth - which is an effect I don't particularly care about. Give me 60fps with less lag over 120fps with lag.

 

[noko]

Does anyone know if Reflex2 is already designed to generate more frames or just shift single last generated frame?
Generating more is such a no-brainer that it is how frames should be generated.
In this case we talk seeing 100% generated frames and perhaps more susceptibility for artifacts - but still hundreds times better way to do frame generation.

Not sure how such FG would handle 30fps artifacts-wise and with issues like generated frames not matching game state but still would be much more usable than having input lag of 30fps multiplied. Especially with multi-player games 120+fps is usually enough as even sweats don't change directions that often. It is just much easier to aim when game reacts instantaneously to your mouse movements.


I think that even if no considerations were taken to this aspect of aiming at game's side it would still be helpful for aiming for online games.
In single player games it should be non-issue unless perhaps running very low frame rate like 30 and below.

And we should not make games which can only run 30fps even with upscaling.
Biggest issue with current FG is that when you move mouse it all feels floaty and unpleasant. If you only play with this lag and with FG you get used to this lag and this response of mouse and your brain will compensate but if you play at good frame rate and then move to FG it feels even more unplayable than without FG. At least I didn't feel it improved anything at any frame rate except made game look more smooth - which is an effect I don't particularly care about. Give me 60fps with less lag over 120fps with lag.

Looks like reflex 2 will warp the coordinates of the mouse or click to the game engine that it adequately allows non rendered frames to accurately represent game status. In other words you shoot at a target in a generated frame, Reflex 2 on the next rendered frame transposes the shot to where it would be on the rendered frame. Have to see how it works out.

 

[Blackstone]

How do I know what preset to use for DLSS override? If I want DLSS Tramsformer Model in Battlefield 2042 for example, is it “Latest” or Preset J or K or what? Where are the presets defined?

 

[pendragon1]

"As of January 30th, 2025, the “Latest” model for DLSS Super Resolution has been updated to Transformer "Preset K", a minor refinement to Transformer Preset J, which showcases improved temporal stability, reduced ghosting, and enhanced detail in motion."

you have to choose which you prefer.

 

[Blackstone]

View attachment 708776
"As of January 30th, 2025, the “Latest” model for DLSS Super Resolution has been updated to Transformer "Preset K", a minor refinement to Transformer Preset J, which showcases improved temporal stability, reduced ghosting, and enhanced detail in motion."

you have to choose which you prefer.

So if I do this override, does it matter whether in game DLSS is set to Quality or Balanced or does this override that?

 

[pendragon1]

So if I do this override, does it matter whether in game DLSS is set to Quality or Balanced or does this override that?

https://nvidia.custhelp.com/app/answers/detail/a_id/5620/~/enabling-dlss-4-overrides-in-nvidia-app
this explains it. you have settings on in game, then make the adjustments in the app.