• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Fake frames: An uncomfortable truth about DLSS 5

Fake frames and upscaling aren’t bad technologies; they have their place.
But they’re necessary because RT is “heavy,” and RT is heavy because it sells very well.
Here’s an example:


View: https://www.youtube.com/watch?v=gsZiJeaMO48

But with monitors operating at 500–1000 Hz, fake frames are necessary. Yes, when we click, we need a real frame for the system to respond, but first we need to visualize when to click, and this is exactly where a higher FPS helps—regardless of whether the frames are real or fake.
 
kram182 said:
If gamers don't bring in the majority of money/profits anymore - and AI does - then gamers are literally not the bread and butter by definition. But regardless - Nvidia still brought in $3.6 billion from gaming in their last quarterly report (vs AMD bringing in just $843 million from gaming in comparison - Intel doesn't even disclose it's that low - thrown in with 'Client Computing' in their reports)

If anyone thinks Nvidia - or Jensen - is going to willingly give up that money on the table - or cede this/these crowns to anyone else (as much as you wish he would):

View attachment 791743

View attachment 791741

then I also have a bridge in Brooklyn to sell you and you should know the people you're hating better.



In your buddy's own words:



To which:
Click to expand...
If you’re going to take on everyone here then figure out who said what. I don’t know who the person is you’re referring to as my buddy, more over, I do not agree with that analysis . Looks like you confused yourself.
 
Hecate91 said:
Forcing upscaling, and now AI hallucinated lighting instead of pre-baked in lighting are all economic decisions that have already been explained in this thread, it's about cutting costs on the gaming division while Nvidia is raking record profits on AI.
It's clear they're caring less and less about their bread & butter market that has built up Nvidia to where they are now.

I don't like that AMD is copying Nvidia, they're damned it they do, and if they don't. I guess we'll see if they go along with the AI image filter effect.
Click to expand...
Follow the money flows and you'll figure out where the problem lies.
 
  • Like
Reactions: erek
like this
MaZa said:
That is a strawman and you know it. Frame gen is basically no different from you turning on motion smoothing on your TV, you have the real actual frames and then there are the "fake" frames which the motion smoothing algorithm tries to guess what they look like based on the real frames on hand. Sometimes it succeeds without artifacts, other times it creates a mess, but the end result is still slower real performance and increased input lag because it has to buffer frames for the generation part, that thing takes time and it cannot guess the future, it has to work on old frames and hold them back before showing them. Nvidia is simply peddling an ancient solution as a miracle performance enhancer (which it is not) and that rubs people, who actually know how things work, the wrong way.
Click to expand...
It's not. It's genuinely bizarre to denigrate a feature as "fake" in a space where quite literally every aspect of computer imaging for the last fifty years is "fake" in one way or another. This just happens to be yet another in a very long line of tricks designed to fool the human visual system into believing that something fake they're seeing on screen looks and feels real. You know, like 24-bit color (fake/not real), AA (fake/not real), rasterization (fake/not real) and everything else. People should chill out - none of this stuff is "real" or supposed to be real.
 
sphinx99 said:
It's not. It's genuinely bizarre to denigrate a feature as "fake" in a space where quite literally every aspect of computer imaging for the last fifty years is "fake" in one way or another. This just happens to be yet another in a very long line of tricks designed to fool the human visual system into believing that something fake they're seeing on screen looks and feels real. You know, like 24-bit color (fake/not real), AA (fake/not real), rasterization (fake/not real) and everything else. People should chill out - none of this stuff is "real" or supposed to be real.
Click to expand...
Also if you don't like it, don't use it. If you insist on only natural, organically computed pixels nobody is making you use these technologies... but you are going to have to turn the graphics down or suffer low FPS. You can disable RT, turn down detail, etc and basically run games with more console like graphics just fine. But if you want the shiny graphics and want them at high FPS, well then "fake frames" and "fake pixels" it is. Too many people seem to think games could have amazing graphics and run at 4k240 if only developers just wanted hard enough. Nope.
 
sphinx99 said:
It's not. It's genuinely bizarre to denigrate a feature as "fake" in a space where quite literally every aspect of computer imaging for the last fifty years is "fake" in one way or another. This just happens to be yet another in a very long line of tricks designed to fool the human visual system into believing that something fake they're seeing on screen looks and feels real. You know, like 24-bit color (fake/not real), AA (fake/not real), rasterization (fake/not real) and everything else. People should chill out - none of this stuff is "real" or supposed to be real
Click to expand...

Yes it is because you are latching on the one word of what is being said instead of the argument itself. All of those examples you mentioned are done in real time, even the much maligned TAA that uses temporal data from the previous frames, the game is running and reacts to your inputs in real time, or as fast as the real refresh rate the game is running at. Frame generation, or motion interpolarion, inherently is not, it cannot be because it cannot predict the future. It has to buffer the real frames, increasing lag noticeably. The lag can be minimized by making sure the game is running as fast as possible, increase the real refresh rate, to make it impossible for human to see or feel that buffering but at that point why are you using motion interpolation in the first place as the game is now running fine on pure raster.

This is the crux of the issue. It is not increasing performance. Where it is most needed is when your real time rendering performance is low but it is also when it fails most spectacularily, lag that not only is felt but also seen, and artifacts all over because the jumps between frames are too big for it to comprehend fully what is going on.

It is nothing more than a trick to artificially make FPS numbers look bigger in marketing. This is why it is called fake. In reality, at high real FPS you do not really need it, at low real FPS where a boost would be needed it is useless and actively hinders your gaming experience.
 
MaZa said:
Yes it is because you are latching on the one word of what is being said instead of the argument itself. All of those examples you mentioned are done in real time, even the much maligned TAA that uses temporal data from the previous frames, the game is running and reacts to your inputs in real time, or as fast as the real refresh rate the game is running at. Frame generation, or motion interpolarion, inherently is not, it cannot be because it cannot predict the future. It has to buffer the real frames, increasing lag noticeably. The lag can be minimized by making sure the game is running as fast as possible, increase the real refresh rate, to make it impossible for human to see or feel that buffering but at that point why are you using motion interpolation in the first place as the game is now running fine on pure raster.

This is the crux of the issue. It is not increasing performance. Where it is most needed is when your real time rendering performance is low but it is also when it fails most spectacularily, lag that not only is felt but also seen, and artifacts all over because the jumps between frames are too big for it to comprehend fully what is going on.

It is nothing more than a trick to artificially make FPS numbers look bigger in marketing. This is why it is called fake. In reality, at high real FPS you do not really need it, at low real FPS where a boost would be needed it is useless and actively hinders your gaming experience.
Click to expand...

Because the game still looks WAY smoother at 210fps than it does 70. And at that point, the trade offs are increasingly minimal. That's it, that's the reason it exists.

I have a 240hz OLED. It is night and day difference in how fluid the presentation is.
 
socK said:
Because the game still looks WAY smoother at 210fps than it does 70. And at that point, the trade offs are increasingly minimal. That's it, that's the reason it exists.

I have a 240hz OLED. It is night and day difference in how fluid the presentation is.
Click to expand...

Sure, I can understand that as long as you also understand the trade offs and are willing to take them. As long as you realise you are not making the game run better, it just visually looks smoother. If the game you are playing is not latency sensitive, the trade of may be acceptable because user input plays a lesser role in your gameplay. But if you play any fast paced games where quick reactions are necessary, you are handicapping yourself by introducing several frames of lag.

My pet peeve has never been that motion interpolation exists in the first place, but the fact that it was marketed as performance improvement which it is not. You can have as big multiplier as you like in your interpolation algorithm (IE lossless scaling 20x frame gen) but it still doesn't solve the problem of gamers computers running out of power to run their games fast enough. It never can. Unless it can start somehow predicting the future and we move from interpolation to extrapolation, all-in on AI hallucinations! Obviously not gonna happen.
 
MaZa said:
Obviously not gonna happen.
Click to expand...

Never say never, just need to know the location of everything in the game engine, possibilities of from current position/situation from there on out, probability of those possibilities over others - and the possibilities are not limitless - everything including user input/options are limited by the probabilistic framework/coding of the game/engine/etc - to which it would all have access to/knowledge of. Current framegen is already doing a lot of this. Just takes number crunching and reinforcing.
 
kram182 said:
Never say never, just need to know the location of everything in the game engine, possibilities of from current position/situation from there on out, probability of those possibilities over others - and the possibilities are not limitless - everything including user input/options are limited by the probabilistic framework/coding of the game/engine/etc - to which it would all have access to/knowledge of. Current framegen is already doing a lot of this. Just takes number crunching and reinforcing.
Click to expand...

It would be pretty damn incredible if that happen. Because games are very chaotic, a lot of thing can change wether you keep moving, change directions etc... and if interpolation artifacts are already bad when it has limited information, think of how bad the artifacts are when extrapolation guesses wrong. It thinks you keep moving when infact you have stopped, you have few frames of the wrong guesses before yoir computer goes "oh shit!" and the image suddenly jumps to the actual stationary image. 😁
 
MaZa said:
damn incredible if that happen. Because games are very chaotic, a lot of thing can change wether you keep moving, change directions etc... and if interpolation artifacts are already bad when it has limited information, think of how bad the artifacts are when extrapolation guesses wrong. It thinks you keep moving when infact you have stopped, you have few frames of the wrong guesses before yoir computer goes "oh shit!" and the ima
Click to expand...

Hence the 'reinforcement' until it gets it right/right enough + keep improving/reinforcing from there on out - the amount of number crunching and resources required for is the biggest hurdle. But look how far we've come from Pong, and what that ran on - never say never :)
 
erek said:
opinion?
Click to expand...
No just if that’s what DLSS5 can do, could developers build the game around low polygon models with low resolution textures to save money and memory. Then fill it with markers for the AI to upscale with the desired effect?

So DLSS in this scenario would essentially be dressing mannequins rather than touching up something that was attempted to be a high quality render.

Because if DLSS5 is going to do all that work and if it can do it reliably and repeatedly, why bother with all the other work if it’s just going to be replaced.
 
Lakados said:
No just if that’s what DLSS5 can do, could developers build the game around low polygon models with low resolution textures to save money and memory. Then fill it with markers for the AI to upscale with the desired effect?

So DLSS in this scenario would essentially be dressing mannequins rather than touching up something that was attempted to be a high quality render.

Because if DLSS5 is going to do all that work and if it can do it reliably and repeatedly, why bother with all the other work if it’s just going to be replaced.
Click to expand...

the more input/assets you train it with/it has to work with - the better the result

garbage in/garbage out

but yes, how far down you can whittle/how little assets/quality/detail of you need to get the then desired output, especially as it matures (and less it needs to process), is a/the question

kram182 said:
sounds like it would increase that performance issue everyone's supposedly complaining about
Click to expand...
 
Lakados said:
No just if that’s what DLSS5 can do, could developers build the game around low polygon models with low resolution textures to save money and memory. Then fill it with markers for the AI to upscale with the desired effect?
Click to expand...
Which is great for Nvidia users. Sucks for AMD and Intel users.
Because if DLSS5 is going to do all that work and if it can do it reliably and repeatedly, why bother with all the other work if it’s just going to be replaced.
Click to expand...
Because it's very likely that this will result in even worse image quality for gamers. If you think upscalers ghosting and artifacting looks bad now, just wait and see what an image filter is going to do. And yes, this is an image filter. We also don't know if this will have any other adverse effects like higher latency in games? There's a reason why this is shown with not much motion going on in the games.

I like how we have all this AI tools that can help gamer developers make games easier and quicker, but we gravitate towards things like Ray-Tracing and now image filters as a quick and dirty way to making games. Expedition 33 using some AI made textures and gets it's title removed, but AI image filter is good now?
 
DukenukemX said:
Which is great for Nvidia users. Sucks for AMD and Intel users.

Because it's very likely that this will result in even worse image quality for gamers. If you think upscalers ghosting and artifacting looks bad now, just wait and see what an image filter is going to do. And yes, this is an image filter. We also don't know if this will have any other adverse effects like higher latency in games? There's a reason why this is shown with not much motion going on in the games.

I like how we have all this AI tools that can help gamer developers make games easier and quicker, but we gravitate towards things like Ray-Tracing and now image filters as a quick and dirty way to making games. Expedition 33 using some AI made textures and gets it's title removed, but AI image filter is good now?
Click to expand...

Don't worry, AMD is working on their own version as well (DLSS will prob be the superior IQ option one though as with all the other things they had to play catch up with) with Microsoft/Sony supporting such things as well (Intel working on it too you can find references to)

1773771602891.png

1773771615045.png

1773771803466.png
 
DukenukemX said:
Which is great for Nvidia users. Sucks for AMD and Intel users
Click to expand...
Not so much, that functionality is a core part of Neural Rendering and a bunch of other aspects of the upcoming DX12 update.

I think this is more like the earliest versions of Nvidia’s Ray Tracing for DX12 before it was officially rolled into the platform.
 
GoldenTiger said:
It's not an image filter. Watch the videos.
Click to expand...
And let's say it is? Does it improve things? Then maybe I want it. Again I'll point out the elephant in the room RESHADE. People love that shit and most of what people run on it is... IMAGE FILTERS! While there are plugins that get more deep into the engine and re-wire things (like RenoDX) most of them are post-processing image filters. Why is this a bad thing?

Heck if you want to use the "image filter" argument let's go back to FXAA, SMAA, and the like. Post processing anti-aliasing. Largely displaced by TAA these days for various reasons, but still good tech.

So even if this tech is processing just based on the final image, no data from the engine (I'm not clear if that's the case or not) how does that make it an invalid way of doing things? Post processed AA, contrast adaptive sharpening, tone mapping, color grading, these are all done in the same way and are already a part of games and/or added on by people when they want them.

Personally I am real interested to try this and see how it works. I'll see if I like it in game, if so what games, and so on.

Even previous DLSS technology is like that, I like it but it isn't useful all the time. Like DLSS 4.5 I have yet to find a game where I'm using it. Games that use Lumen for lighting, the issues it causes with that make it not worth the clarity increases. Like Satisfactory. Despite the "meh" Lumen implementation in that game, I still like how it looks with GI on rather than off so I use that. However I find DLSS 4 looks better than 4.5 with that. 4.5 does clear up some ghosting and details, but increases the already existing issues with light and shadow flicker to levels I just don't like. With Tempest Rising, which doesn't use Lumen, DLSS 4.5 looks fine... but after playing around I settled on no DLSS because I just can't really tell any difference given the nature of the graphics (it's an isometric RTS) and it already runs stupidly fast. Same kind of deal with frame generation. In Satisfactory it is a no-go, causes massive flickering, and when I tried Hogwarts Legacy with it there was way too many artifacts (supposedly that is fixed now). However in Cyberpunk 2077 it was fantastic, and it works well in Hitman 3 with only the occasional very minor glitch making it worth it.

I'll treat DLSS 5 the same way: If a game supports it, or if there's an override, I'll give it a shot. If I use it, or not, will depend on the experience. Do I like how it looks? What's the FPS cost? If it makes some things worse, do the things it makes better make up for it? Probably will vary on a game-to-game basis.


Also, given that lighting is what it is supposed to be interacting with, it is the kind of thing that needs to be watched in a longer form video (not these stupid sizzle wipes nVidia does) and really experienced in person to decide how it looks. Reason is Path Tracing is the same. Like if you look at an A/B screenshot of Path Tracing vs rasterization it does look better, but often not transformative. Particularly depending on the how the rasterization is done. Even if it does look transformative, it doesn't always look better, just different. Then you see a video, where things are moving and the lights are interacting with the environment and now it starts to click just how much better it is. Things look more... "right" for lack of a better term. Then you actually play a game and it really clicks. The lighting and shadows of the world just interact in a way that feels better that they just don't no matter how good the rasterization is.

So I'll want to see this IRL before making any real verdict, or at least a long-form video where you get some good comparisons of how the world looks moving through it, not just a focus on a character's face.
 
Sycraft said:
Again I'll point out the elephant in the room RESHADE. People love that shit and most of what people run on it is... IMAGE FILTERS!
Click to expand...
Yep. They tend to be the most downloaded mods, too! Dlss5 is not just an image filter, but that kind of debunks the "gamers only want the artists' intent!" (which will take into account Dlss5 anyway) thing.

P.S. Read the rest of your post too, just on my phone :).
 
kram182 said:
View attachment 792027
Click to expand...
Pretty much :D it amazes me how much of game discussion online comes down to "stop liking something I don't like!" You see it with games too, a dev will make decisions that someone doesn't like and rather than it being a case of "Ya not for me," the response is "This is bad and everyone should hate this. If you DON'T hate this you are bad or brainwashed or stupid!"
 
Comparing anti aliasing to this is retarded. Aliasing is a byproduct of how pixels draw an image. It isn’t developers intent to have jaggies or shimmering. If you like it that’s fine, but trying to prove a point by comparing it to AA sounds like you have no idea what AA is. DLSS5 is changing art work intentionally written to the game by developers.

I personally don’t care as long as I’m not forced to use it. Like frame gen. I don’t like it and don’t use it, but I have no qualms that the option exists.

I am a little annoyed by calling it DLSS when there’s nothing SS about it and the fact they’re showcasing a tech which needs two 5090s that now cost 2x msrp.
 
WilyKit said:
Comparing anti aliasing to this is retarded. Aliasing is a byproduct of how pixels draw an image. It isn’t developers intent to have jaggies or shimmering. If you like it that’s fine, but trying to prove a point by comparing it to AA sounds like you have no idea what AA is. DLSS5 is changing art work intentionally written to the game by developers.
Click to expand...
Yet FXAA and SMAA aren't properly antialiasing. The correct way to do it is SSAA. You oversample each pixel and render enough sub pixels to get the proper color for it. 16 per pixel works pretty well. However, that costs a fuckton of computation, so we don't do that realtime. FXAA and SMAA (and others like them) are post-processing filters. They take the final image, aliasing and all, and programmatically try to say "What looks like aliasing and how can I blend it to remove that?" They are taking the art the game has rendered and modifying it, guessing what it should look like.

How about denoising for ray tracing? Same shit there. The correct way to eliminate noise is to shoot more rays, but it can take MILLIONS per pixel to completely eliminate it so basically nobody does, even Hollywood. You shoot however many rays you want, the more the better but usually just 1 per pixel in realtime rendering and get an image. It is noisy as fuck. You then process that image with something that understands the kind of noise that RT generates and is able to remove it. Again, looking at the final image and guessing what it should look like.

Same shit with DLSS, but more intense. Here you are taking data from multiple frames, combined with motion vectors and geometry information and running a neural network to guess what a higher resolution image should look like from a lower resolution render.

This is the same idea, but different in the details. It is still about saying "What do we think the lighting should look like, given the information we have?" I'm unclear if that information is purely based on the rendered image or if it gets other things like geometry but either way we are talking about the same kind of thing we have being doing to try and fix up images for a long time, just a new facet of it. Will it be worth it? We'll have to see.

In part it'll depend on what it requires. If it does end up needing 2x 5090s then I imagine very few people will be interested. I won't because, money aside, my system doesn't have the room and my computer room doesn't have the power to run that. nVidia implies that they are working on getting it running on a single card. They may be quantizing the model down to FP8 and FP4 to make it take less memory and run faster, in which case a single 5090 or 5080 could probably do it. Then it will be of interest to more people. In part it'll depend on how it looks in actual gameplay. Really for me, I'm more interested about how the environments look than faces. Turns out in most games I don't spend a lot of time starting at the faces of characters, it is usually the environment. Even when I do, they often look kinda unrealistic since, well, we have a lot of limitations in terms of rendering and animation. Even if faces end up a bit "Tiktokified" I'll probably use it if it makes the lighting in the environment better in most games.

We'll see though. I'm quite interested, but I'm not saying "This is the future of gaming!!!111one" I'm saying people should chill the fuck out, stop knee jerking about AI, and let's see what it is when we get it.

As I've said with "fake pixels" and "fake frames" and plenty of other shit gamers like to whine about: It doesn't matter how, what matters is if the game looks good. Is the end result good? Then I'm happy. There's no point in being a "purist" because not only is it self defeating but this shit has NEVER been "pure" We've never had games that properly solved the rendering equation for their output. It has always been hacks and shortcuts and tricks to make it work. I don't like or dislike ones that use neural networks and more than ones that use regular code.
 
Trolling, sniping, personal attacks, and general bickering is not allowed. Please read the rules, please post on topic, and if you are addressing the poster instead of the content of the post, expect to be excluded from this thread. Remember, there is an IGNORE SYSTEM. I would suggest you use it if you want to avoid someone that is causing you so much grief.
 
Lot of people on this subject were saying not even a year ago it was impossible for dlss to look better than native, there is a bit of a lack on what is going on about signal and what is possible in general (or just about what is going on in general) in the pundit discourse.

DukenukemX said:
And yes, this is an image filter.
Click to expand...
games use a lot of image filter (bloom, color grading to film grains or anti aliasing technics), the semantic aware generative aspect of this make it a bit different, it take so much more input than just the image, it is knowledge adding signal information.
 
LukeTbk said:
Lot of people on this subject were saying not even a year ago it was impossible for dlss to look better than native, there is a bit of a lack on what is going on about signal and what is possible in general (or just about what is going on in general) in the pundit discourse.


games use a lot of image filter (bloom, color grading to film grains or anti aliasing technics), the semantic aware generative aspect of this make it a bit different, it take so much more input than just the image, it is knowledge adding signal information.
Click to expand...
Better than native, or is It changing native? I would say the latter. Up until nVidia's latest development the changes were minimal but not so in your face that most people were OK with it most of the time. Now? It's just creating new textures and lighting that aren't even there.
 
kac77 said:
Better than native, or is It changing native? I would say the latter. Up until nVidia's latest development the changes were minimal but not so in your face that most people were OK with it most of the time. Now? It's just creating new textures and lighting that aren't even there.
Click to expand...
become subjective it was said in the sense that the native 4k was some upper limit of details for it ( a render that never existed in the pipeline and did not made much sense to be seen as a theorical limit).

yes that the first goal to add photo realistic information "missing" from the game because of general game restriction. If people do not like it, it will simply flop, games will not use it and go in a long list of death tech like gpu physix, apex
 
Sycraft said:
Yet FXAA and SMAA aren't properly antialiasing. The correct way to do it is SSAA. You oversample each pixel and render enough sub pixels to get the proper color for it. 16 per pixel works pretty well. However, that costs a fuckton of computation, so we don't do that realtime. FXAA and SMAA (and others like them) are post-processing filters. They take the final image, aliasing and all, and programmatically try to say "What looks like aliasing and how can I blend it to remove that?" They are taking the art the game has rendered and modifying it, guessing what it should look like.

How about denoising for ray tracing? Same shit there. The correct way to eliminate noise is to shoot more rays, but it can take MILLIONS per pixel to completely eliminate it so basically nobody does, even Hollywood. You shoot however many rays you want, the more the better but usually just 1 per pixel in realtime rendering and get an image. It is noisy as fuck. You then process that image with something that understands the kind of noise that RT generates and is able to remove it. Again, looking at the final image and guessing what it should look like.

Same shit with DLSS, but more intense. Here you are taking data from multiple frames, combined with motion vectors and geometry information and running a neural network to guess what a higher resolution image should look like from a lower resolution render.

This is the same idea, but different in the details. It is still about saying "What do we think the lighting should look like, given the information we have?" I'm unclear if that information is purely based on the rendered image or if it gets other things like geometry but either way we are talking about the same kind of thing we have being doing to try and fix up images for a long time, just a new facet of it. Will it be worth it? We'll have to see.

In part it'll depend on what it requires. If it does end up needing 2x 5090s then I imagine very few people will be interested. I won't because, money aside, my system doesn't have the room and my computer room doesn't have the power to run that. nVidia implies that they are working on getting it running on a single card. They may be quantizing the model down to FP8 and FP4 to make it take less memory and run faster, in which case a single 5090 or 5080 could probably do it. Then it will be of interest to more people. In part it'll depend on how it looks in actual gameplay. Really for me, I'm more interested about how the environments look than faces. Turns out in most games I don't spend a lot of time starting at the faces of characters, it is usually the environment. Even when I do, they often look kinda unrealistic since, well, we have a lot of limitations in terms of rendering and animation. Even if faces end up a bit "Tiktokified" I'll probably use it if it makes the lighting in the environment better in most games.

We'll see though. I'm quite interested, but I'm not saying "This is the future of gaming!!!111one" I'm saying people should chill the fuck out, stop knee jerking about AI, and let's see what it is when we get it.

As I've said with "fake pixels" and "fake frames" and plenty of other shit gamers like to whine about: It doesn't matter how, what matters is if the game looks good. Is the end result good? Then I'm happy. There's no point in being a "purist" because not only is it self defeating but this shit has NEVER been "pure" We've never had games that properly solved the rendering equation for their output. It has always been hacks and shortcuts and tricks to make it work. I don't like or dislike ones that use neural networks and more than ones that use regular code.
Click to expand...
100% agree with everything you said.

I think this is one of those cyclical things. Heck, back when HW T&L first started showing up, there were people upset about it, claiming that HW T&L is fake and proper ray tracing is the only way to go. These new neural models in DLSS5 are simply another rung on a very tall ladder full of incremental baby steps toward what ultimately are fake images that look real--which is all a computer game actually is.
 
As we’ve seen, the “nice” filter in DLSS 5 doesn’t care about the light source, so whether it’s better than the original is also a matter of subjective opinion.
Furthermore, the main idea behind DLSS is to achieve a higher FPS, not to lower it.
Nvidia simply made a wrong decision to include this filter in DLSS instead of just creating another software filter.
 
AVATARAT said:
As we’ve seen, the “nice” filter in DLSS 5 doesn’t care about the light source, so whether it’s better than the original is also a matter of subjective opinion.
Furthermore, the main idea behind DLSS is to achieve a higher FPS, not to lower it.
Nvidia simply made a wrong decision to include this filter in DLSS instead of just creating another software filter.
Click to expand...

Somehow I didn't even think of this

Yeah, I think DLSS is like... pretty beyond it's actual vision and purpose at this point. This should have just been its own thing.

Also where the fuck is Reflex 2
 
AVATARAT said:
As we’ve seen, the “nice” filter in DLSS 5 doesn’t care about the light source, so whether it’s better than the original is also a matter of subjective opinion.
Furthermore, the main idea behind DLSS is to achieve a higher FPS, not to lower it.
Nvidia simply made a wrong decision to include this filter in DLSS instead of just creating another software filter.
Click to expand...
That is perhaps the thing that bothers me most. The lighting.

Nvidia just spent 8 years. Telling everyone light maps were inferior. That everyone should be tracing rays. Creating the most realistic lighting possible.

Now they seem to be saying no no. Lets let this model weigh the scene and judge what light map would be the most pleasing to look at. Never mind real physics or where the lights are coming from. Lets just put a diffused ring light in front of every NPC. Shadows? Eh the AI can figure out where those will go. (the shadows are actually wrong in more then one of their demo scenes)

I thought RT was not worth the computational cost... but I never argued that it wasn't actually superior from a technical point of view. I have always believed a properly made light map is superior even to RT for video games. (I mean not counting the flashy things like reflections in mirrors) In general game play scenarios imo a well created light map with artistic merit is superior to even the best RT. I paint a little... I wouldn't paint a photograph or something that looked like one. What would be the point. I feel the same way about games, realism is fun but NOT hype realism.

IMHO I think the gut reaction of the people saying OH NO THIS LOOKS TERRIBLE. Isn't just instagram filter or clay face NPCs... its the lighting. Your mind knows there are too many highlights. Your mind KNOWS the shadows aren't right even if you don't "notice". Your mind knows there is something wrong about what your looking at.

I also agree that this tech has no place in DLSS. What does this tech have to do with super sampling? Its a completely different thing.
 
Several new dlss5 videos out now regarding the tech. In a nutshell, it’s an image filter with motion vectors. Nothing more. any and all lighting/shadow enhancements… aren’t. The tech is completely clueless as to lighting sources. Lipstick on a pig comes to mind.
 
WilyKit said:
Several new dlss5 videos out now regarding the tech. In a nutshell, it’s an image filter with motion vectors. Nothing more. any and all lighting/shadow enhancements… aren’t. The tech is completely clueless as to lighting sources. Lipstick on a pig comes to mind.
Click to expand...
And yet, I do not care if it looks good. As a consummate Reshade user, there are plenty of things I like that can be done with just an image filter. If this can make lighting look better, then I'm for it, regardless of how that is done.

We'll just have to wait and see. I'm certainly not going to say it will revolutionize graphics without having seen it in person, and without knowing what the actual requirements/performance cost will be. But I think it is really silly to hate on it.
 
Sycraft said:
And yet, I do not care if it looks good. As a consummate Reshade user, there are plenty of things I like that can be done with just an image filter. If this can make lighting look better, then I'm for it, regardless of how that is done.

We'll just have to wait and see. I'm certainly not going to say it will revolutionize graphics without having seen it in person, and without knowing what the actual requirements/performance cost will be. But I think it is really silly to hate on it.
Click to expand...
I know you don’t care. You made it clear multiple times lip stick on a pig in your eyes is better than no lipstick at all. I’ll stick with RT where lighting looks better and is more accurate. You can wait and see, after getting more details on how the tech works it’s already DOA to me.
 
WilyKit said:
I know you don’t care. You made it clear multiple times lip stick on a pig in your eyes is better than no lipstick at all. I’ll stick with RT where lighting looks better and is more accurate. You can wait and see, after getting more details on how the tech works it’s already DOA to me.
Click to expand...
But why? Why are you dismissing it is "dead" when what you've seen are short videos of a demo? Did you likewise dismiss DLSS upscaling after its initial demo with Metro Exodus, where IMO it wasn't all that impressive? It seems to me like the reaction people are having is less about its actual visuals and more "I hate AI, therefore I hate this." My argument to people isn't "You should love this!" or "This is the future of gaming!" it is just to chill out, and let's see what this ends up being. The fact that it works on the final 2D render is not an inherently bad thing and doesn't mean it can't improve visuals.

I guess I shouldn't be surprised, since this happens with every new technology it seems like, but I am annoyed that the knee jerk reaction is one of "Fuck this because it is AI, I hate it without any further consideration."
 
Sycraft said:
But why? Why are you dismissing it is "dead" when what you've seen are short videos of a demo? Did you likewise dismiss DLSS upscaling after its initial demo with Metro Exodus, where IMO it wasn't all that impressive? It seems to me like the reaction people are having is less about its actual visuals and more "I hate AI, therefore I hate this." My argument to people isn't "You should love this!" or "This is the future of gaming!" it is just to chill out, and let's see what this ends up being. The fact that it works on the final 2D render is not an inherently bad thing and doesn't mean it can't improve visuals.

I guess I shouldn't be surprised, since this happens with every new technology it seems like, but I am annoyed that the knee jerk reaction is one of "Fuck this because it is AI, I hate it without any further consideration."
Click to expand...
You cirtisized me in a now deleted post for not wanting to read your long winded reply, yet you do an exceptionally bad job at reading my short ones. I would much rather you give me the courtesy I gave you, which is telling you i'm not going to read it, instead of catching a glimpse of something you didn't like because it was anti-nvidia then responding with asked and answered questions.

First off I said "DOA to me" meaning, I personally am not interested in it, before that I illustrated my reasons, but I'll reiterate. The tech has zero clue what the scene is suppose to look like when it comes to lighting and shadows, it simply "guesses" at what is pleasing to the eye and applies it. Which is why I compare it to lip stick on a pig. Yeah, that pig with makeup will certainly "pop" but that shit aint supposed to be there. Taking a single frame with no input other than motion vectors then guessing what is pleasing to the eye is all sorts of immersion breaking. Developers, especially for SP games and ESPECIALLY scarier ones prioritize environment/atmosphere. They do this in large by using meticulously rendered shadows, lighting, reflections and sound. DLSS5 tosses 3/4 out the window.

Knowing now what the tech does, i'm simply not interested.
 
You must log in or register to reply here.
Back
Top