You must log in or register to comment.
DLSS takes sane people and turns them crazy.
DLSS uses AI tricks to improve framerate.
Previous versions used AI upscaling and frame generation.
This version now uses AI filters. It changes lighting and texture effects to make the image “better”, instead of just making the image faster.
It’s actively manipulating creative decisions about the style of a game. Which naturally upsets many developers and players alike.Yes. Adding on… It renders whatever game at a much lower resolution (to make game run better and save on resources), then uses phycology tricks to postprocess and overlay specific little things to make it seem not as low of resolution.
Nslopia decided that since they were using “Deep Learning” to “Super Sample” and do what I said above, they figured, why not use “Deep Learning” to make the image look “better”, too?
Putting it overly simply (ELi5)…
DLSS uses machine learning to do two things:
- figure out, in real time, what’s going on on-screen
- change it in some way
All that DLSS 5 changes is to use real life professional photography in stage 2, instead of trying to be subtle and use existing higher resolution videogame images/look. It’s an interesting experiment just for fun, but some disrespectful, greedy, clueless simpleton MBA with bad taste obviously saw it and thought “oh yes”.
Good explanation. For a bit more context:
AFAIK DLSS was the first “AI technology” to hit the gaming market when NVIDIA revealed it a few GPU generations back. Unsure if it came with the RTX 2000 series or later but somewhere along those lines. DLSS consisted of 2 major technologies, upscaling and frame generation.
Upscaling uses an AI model to take a lower resolution image (frame) and upscale it to a higher resolution, this cutting the required performance to rendering the frame to begin with. This improves framerate at the cost of bluriness in fast moving scenens and some ghosting artrefacts. This has been improved across the different DLSS versions but the issue still remains, although way less notable.
Frame generation on the other hand generates “fake” frames in between naturally generated frames to improve framerate and produce a more fluid experience. As the model needs extra time to produce this extra frame it also leads to some latency being added resulting in input delay, which is not ideal for fast-paced competitive games.
Both of these technologies affects the visuals of the game somewhat but still retains the original artists vision. Although it does influence the visual quality and brings some side effects such as smearing, ghosting and artefacts.
The new DLSS 5 debacle is a fresh can of worms. While previous DLSS versions have focused on retaining the original image and only improving performance this version now heavily influences the content of whatever is being rendered. The AI now alters the lighting, textures, geometry, etc in the frame and applies the most glorious of AI slop you could imagine, removing any sense of artistic direction and atmosphere from the game.
NVIDIA is attempting to damage control the situation by claiming that developers have full control over what is being touched by the AI model and what isnt. But the reality remains that this does influence the scene content with unpredictable AI generated content, thus simply removing artistic ability and feeling from games.
But we do not need to worry, NVIDIA CEO Jensen Huang himself has told us that “we are wrong” and that this is simply amazing so as every good citizen under a capitalist technocratic ogligarchy i believe we should just sit bow down and praise DLSS 5 as the godly diety it is. /s
It makes games run smoother and look cleaner.
5 replaces assets in the game with AI bullshit.
It makes games run worse and look horrible because devs now target dlss performance mode for 60fps on medium hardware. The game is rendered at 720p and blurred to death until you can’t notice the half rendered frame and pixelated effects that depend on it (and TAA).
It’s also a marketing ploy to lie to consumers, saying their cards get 3x the performance of the previous Gen, by generating fake frames in between real ones. Displaying garbage but because the benchmark say 100+fps, it must be better right? Even though there are more presentation errors, latency is through the roof and the result is insane.
DLSS 5 continues the trend of lies by simply putting a yassification filter on top, lies to the dev, and lies to the consumers.
It makes games run worse and look horrible because devs now target dlss performance mode for 60fps on medium hardware.
That’s like saying new graphics cards make a game run worse because developers make their games targeted at new hardware.
You’re conflating two issues. DLSS uses upscaling to improve framerate the the cost of image quality.
The decision of developers is a completely different issue.
That’s hella weird. 720p only upscales nicely on multiples, so 1440p or 2880p, but last I checked only about 15% of Steam users have QHD monitors. They should’ve gone 960*540 at the very least.
Of course, they shouldn’t half-ass their quality regardless. Frame gen is fine I guess, but not rendering at full res and using AI to distort the original is a huge issue.
Frame gen, in my experience, is worse than upscaling. At least upscaling has some uses (but about 10% of what they say in marketing materials). Frame gen just makes the game have worse latency for disgusting frames that barely look like the base game.
I disagree with your premise. It does make games run better which has helped devs get lazy and use it as a crutch. When devs don’t use it as a crutch but let it be a bonus it’s great.
Though I just replayed Just Cause 3 for kicks and think visually it holds up to just about any of today’s games and runs like a top with any fancy stuff.
I sure hope the games run better when enabled, it’s effectively running at lower resolution and upscaling. It’s just a fancy resolution select marketed as genuine improvements.
They don’t run better though. DLSS just makes fake frames based in guesses which looks like it is running better.
It is the same basic premise as LLM slop, it looks like something despite not actually being that thing.
Frame Gen makes fake frames. Standard dlss does no such thing which is what the question was. It uses temporal upscaling. Not the same thing.
Upscaling = artificially increasing the sampling rate through some sort of inter/extrapolation.
Temporal = it’s happening on the temporal axis.
Samples on the temporal axis are frames.
Therefore, temporal upscaling = artificially sampling more frames = frame gen?
No, that’s like saying TAA is frame gen.
Using temporal data to upscale is different than inserting frames that weren’t there to begin with.
TAA as in temporal anti-aliasing? Is that not frame generation? It’s interpolating between frames to create a frame that wasn’t previously there. Just like how spatial anti-aliasing generates pixels that weren’t previously there.
I think maybe we have a different idea of what “generation” means. I’m guessing your idea of “generation” is when it surpasses some threshold of information added through the process.
The early versions only did upscaling, but we are in the current day where DLSS does fake frame gen to increase fps on top of upscaling.
Only if it’s turned on. I have yet to see a game that forces it. You get to pick if it’s on or not.
You sound like you have never used a modern nVidia card. Not a slam, just saying it you had one you’d know it’s optional.
Keep moving those goalposts.
It’s more like running an ai instagram filter over the game, than replacement. Still unnecassary though.
DLSS is machine learning model trained by Nvidia to invent information to rendered game frames. In general there are two ways:
- You have two frames of game, model makes up third frame between those two frames, to make game run smoother. This has down side that input lag (you press button and it happens on screen) gets longer. But it increases your FPS (frames per second), which looks smoother.
- model makes up extra detail to the game frame. Previous versions have mainly being AI upscaling. Game renders in smaller resolution (less pixels), and then it is made larger, and model smooths and makes up details that originally wasn’t there.
Both of these have same downside, they can cause inconsistency between frames, and have flickering etc.
This is what DLSS 5 now takes to much more heavier “AI filter” level, because it is not anymore upscaling, but it now can add much more detail that originally wasn’t there. But because the model doesn’t have memory, it must always cause issue that the output will always be bit different. Curious to see how they have overcome that, or is it really so that with DLSS5 with every cutscene, the main character looks different.
Nvidia’s software bits for upscaling (game plays with lower resolution, GPU does funny things to make it look higher resolution), frame insertion (in a nutshell, creating fake frames in between real frames to make gameplay look smoother), and similar. Many of the features require game support though, so a majority games are unable to utilise the more advanced features. They can also result in a degradation in quality and increased latency.
DLSS 4 introduced multi frame gen, where up to three fake frames were made for every one. It doesn’t magically make you have 4x the performance (so no, the 5070 is not the performance of the 4090), but it’s still pretty neat tech.
DLSS 5 is being criticised as it’s “enhancing” video game graphics to look like AI-generated images that look more “cinematic” and “photorealistic”. To quote an IGN article, “I think it looks shit.” (https://www.ign.com/articles/nvidias-dlss-5-is-a-slap-in-the-face-to-the-art-of-video-game-design)
In other words, it’s removing the human element of game graphics and nobody likes it.
I’m so glad I use AMD now. I know DLSS is optional and they lead in raytracing efficiency, but supporting Nvidia’s choices by buying their product is just not worth it.
Skipping the what, it was covered.
Why? Gamer communities screech about everything. They have an absurdity vocal minority who are just some of the worst people to deal with on the planet. Nothing will make them happy. They’ll screech at a $70 game that gave them 100 hours of play time on principle.
They screeched about dlss1 and they’ll continue to do so. They’ll all be using it and it’ll be tweaked to look even better in a few years.
It generally doesn’t help the case if the one comment making it is the screechiest in the thread.
👌 gamers are a sane bunch.
How long before DF got death threats over dlss5? Thought so.
