Monday, March 16th 2026

NVIDIA Unveils DLSS 5 with Real-Time Neural Rendering

NVIDIA today unveiled NVIDIA DLSS 5, the company's most significant breakthrough in computer graphics since the debut of real-time ray tracing in 2018. DLSS 5 introduces a real-time neural rendering model that infuses pixels with photoreal lighting and materials. Bridging the divide between rendering and reality, DLSS 5 empowers game developers to deliver a new level of photoreal computer graphics previously only achieved in Hollywood visual effects.

"Twenty-five years after NVIDIA invented the programmable shader, we are reinventing computer graphics once again," said Jensen Huang, founder and CEO of NVIDIA. "DLSS 5 is the GPT moment for graphics—blending hand-crafted rendering with generative AI to deliver a dramatic leap in visual realism while preserving the control artists need for creative expression."
Bridging The Cinematic Gap
Since the dawn of GeForce, NVIDIA has strived to deliver the graphics horsepower required for game developers to create incredible, realistic worlds—where lighting, reflections and shadows obey the laws of nature.

From programmable shaders with GeForce 3 in 2001, to CUDA with GeForce 8800 GTX in 2006, to real-time ray tracing with GeForce RTX 2080 Ti in 2018, to path tracing and neural shaders with GeForce RTX 5090 in 2025, NVIDIA has delivered major architectural innovations and a massive 375,000x increase in compute to meet this challenge.

However, the rendering horsepower available to a 16-millisecond game frame remains a tiny fraction of that available to a photoreal Hollywood VFX frame, which can take minutes to hours to render. Real-time rendering cannot bridge the gap to photorealism through brute force alone.

NVIDIA DLSS was released in 2018 as an AI technology to boost performance, first by upscaling resolution, and then by generating entirely new frames. It has been integrated in over 750 games, becoming a gold standard for the industry. Launched at CES this year, DLSS 4.5 uses AI to draw 23 out of every 24 pixels seen on the screen. Today, DLSS is evolving beyond performance to transform visual fidelity in games.

Video AI models have rapidly learned to generate photoreal pixels, but they run offline, are difficult to precisely control and often lack predictability, with every new prompt generating bespoke content. For games, pixels must be deterministic, delivered in real time and tightly grounded in the game developer's 3D world and artistic intent.

DLSS 5 takes a game's color and motion vectors for each frame as input, and uses an AI model to infuse the scene with photoreal lighting and materials that are anchored to source 3D content and consistent from frame to frame. DLSS 5 runs in real time at up to 4K resolution for smooth, interactive gameplay.

The AI model is trained end to end to understand complex scene semantics such as characters, hair, fabric and translucent skin, along with environmental lighting conditions like front-lit, back-lit or overcast—all by analyzing a single frame. DLSS 5 then uses its deep understanding to generate visually precise images that handle complex elements such as subsurface scattering on skin, the delicate sheen of fabric and light-material interactions on hair, all while retaining the structure and semantics of the original scene.

DLSS 5 provides game developers with detailed controls for intensity, color grading and masking, so artists can determine where and how enhancements are applied to maintain each game's unique aesthetic. Integration is seamless, using the same NVIDIA Streamline framework used by existing DLSS and NVIDIA Reflex technologies.

Availability & Game Developer Support
DLSS 5 will be supported by the industry's biggest publishers and game developers, including Bethesda, CAPCOM, Hotta Studio, NetEase, NCSOFT, S-GAME, Tencent, Ubisoft and Warner Bros. Games.

"Bethesda has such a rich history pushing graphics with NVIDIA, going all the way back to Morrowind, with that incredible water," said Todd Howard, studio head and executive producer at Bethesda Game Studios. "When NVIDIA showed us DLSS 5 and we got it running in Starfield, it was amazing how it brought it to life. We've played it. We can't wait for all of you to do so as well."

"At CAPCOM, we strive to create experiences that feel cinematic, compelling and deeply believable—where every shadow, texture and ray of light is crafted with intention to enhance atmosphere and emotional impact," said Jun Takeuchi, executive producer and executive corporate officer at CAPCOM. "DLSS 5 represents another important step in pushing visual fidelity forward, helping players become even more immersed in the world of Resident Evil."

"Immersion is about making the world feel real. DLSS 5 is a real step towards that goal," said Charlie Guillemot, co-CEO of Vantage Studios. "The way it renders lighting, materials and characters changes what we can promise to players. On Assassin's Creed Shadows, it's letting us build the kind of worlds we've always wanted to."

DLSS 5 will come to games including AION 2, Assassin's Creed Shadows, Black State, CINDER CITY, Delta Force, Hogwarts Legacy, Justice, NARAKA: BLADEPOINT, NTE: Neverness to Everness, Phantom Blade Zero, Resident Evil Requiem, Sea of Remnants, Starfield, The Elder Scrolls IV: Oblivion Remastered, Where Winds Meet, and more.

DLSS 5 will arrive this Fall. See the first preview this week at GTC, and below, you can check out several examples of NVIDIA DLSS 5 in Resident Evil Requiem, EA SPORTS FC, Starfield, Hogwarts Legacy, and the NVIDIA Zorah tech demo.

Resident Evil Requiem
EA SPORTS FC
Starfield
Hogwarts Legacy
NVIDIA Zorah Tech Demo
Source: NVIDIA
Add your own comment

305 Comments on NVIDIA Unveils DLSS 5 with Real-Time Neural Rendering

#126
Visible Noise
umeng2002I assure you, Capcom did not do anything for that RE9 demo.
Citation?

If they didn’t do anything, that’s even better. Means this can be used at my option with minimal developer effort.
Posted on Reply
#127
nguyen
Would love to see DLSS5 in Witcher 3 and Cyberpunk 2077 :rockout:
Posted on Reply
#128
Naito
Excited to see how games that were designed with DLSS 5 in mind would look. As long as the devs have control over ambience, atmosphere and art style.
Posted on Reply
#129
evernessince
oxrufiioxoPeople are bashing this pretty hard online.

DLSS upscaling, online people hated it (now people buy Nvidia over AMD for it)

Framegen, online people hated it (people actually buying hardware seem to love it) I still don't like it.

DLSS5, only time will tell.

Anything AI related tends to get a strongly negative reaction online.

I had my wife watch every video and she mostly prefers DLSS5 which I find interesting.
I can see some people preferring it at first. The problem will then come later when all games start looking similar, as will the characters, as AI has that effect. Go and prompt an AI for an ugly person, it either won't be capable of doing it (because that's not in it's training dataset) or it'll give you it's idea of an ugly person that it will pull every single "ugly" character towards. The same issue happens across every concept. Any word you input the AI has a concept of, it'll pull the input towards (in this case the game data). Even worse if it doesn't have a concept of it, AI does horribly when something isn't in it's dataset as it's not capable of actually thinking.
Posted on Reply
#130
Visible Noise
Dr. DroEveryone hating on this haven't even bothered to read the announcement post in full, it's hilarious. Developers have complete control over the implementation
It’s understandable…Radeon users aren’t accustomed to having a choice about having features available to them.
Posted on Reply
#131
Dr. Dro
BoxCutter"Developers are gonna have complete control" this just comes across as preemtive deflection, developers that want complete control are not gonna rely on something that tempers with an imagine to such an extend. It' AI, it's by definiton non deterministic, if it were it would just be a shader. Also it's a vendor exclusive closed source feature, how is that "complete control"

Feature being bad and not sticking isn't the end of the world, but you saying that a feature sucks definitely is an end of the world.
Not mutually exclusive. Developers having control over the implementation of something doesn't mean it has to be supported on other vendors' graphics hardware.
evernessincePeople read the post. Devs can apply masks manually and control the intensity. The problem is that it doesn't solve the fundamental issues this technology has. You could apply it very lightly but then you are loading and running the full AI model for little benefit. Realistically it only makes sense when you are using it to notable effect, otherwise just go with raster lighting / rt. Using it more heavily has the prior discussed issues. I have to wonder how masks will even work. When you mask an AI image to control what gets painted, you have the huge issue that whatever isn't painted has entirely different lighting and the content doesn't align perfectly, hence by most AI editing programs have a mask edge region to apply some blur or blend method to make it look better. Have you ever tried using masks to edit AI content? Even doing it by hand, it can take me 20-100 iterations and manual editing in between to get it to look good. Of course that's to ensure the end result actually looks good instead of like AI slop. Let's assume devs mask just the faces, they'd look completely out of place compared to the rest of the lighting unless they have a 2nd AI come through and apply a controlnet with the background image as a style reference.

This would be really cool if 1. It didn't change the gemoetry or shadows at all 2. It had zero impact on the art style 3. It runs well on reasonable hardware 4. It is very near 100% consistent 5. Doesn't suffer from concept bleed / overfitting

Even with all the data in the world, the issues are with the AI model itself. If I generate 1,000 images of people in flux all with different prompts, chances are 1,000 of those images will have people with butt chins with the males and the females all having similar face shapes. Even if you name celebrities, the model will output an image that mixed the concept of that celeberty with it's understanding of the other words in the prompt. That is just the nature of modern AI models, concept bleed is a PITA. Each input can pull many concepts along with it aside from the main one. Face shape, eye shape, basically anything you can think of ,the model has a generalized concept of it will drag the output towards.
This just in, new, heavy feature currently in development is very heavy and quite buggy indeed and only runs on nearly futuristic tier hardware. More at 9.

Don't take this the wrong way, I'm not dismissing your very valid concerns, which I actually agree with. It'd suck if most games ended up having stable diffusion syndrome. But not only isn't the tech fully ready, but we don't even know to which extent it can be used with meaningful benefits. I'm more than willing to give it some time, we may even end up surprised with a solution that brings meaningful improvements, yet doesn't need a 5090 (or two) to run.
Posted on Reply
#132
Visible Noise
GuckyDLSS-RR works by replacing the denoiser with a better one. It is not hallucinating makeup on a face like they have shown in the demo...
Ray Reconstruction literally does denoising with AI. This is the new model for it.

There’s no makeup being hallucinated. It’s called physically correct lighting. You’re just used to games having bland, same intensity lighting.

Look in the mirror, are there not shadows under your eyes?
Posted on Reply
#133
BoxCutter
Dr. DroNot mutually exclusive. Developers having control over the implementation of something doesn't mean it has to be supported on other vendors' graphics hardware.
So not complete control, so nVidia just like lied? This really is just 5070=4090 all over again.
Posted on Reply
#134
evernessince
Dr. DroThis just in, new, heavy feature currently in development is very heavy and quite buggy indeed and only runs on nearly futuristic tier hardware. More at 9.

Don't take this the wrong way, I'm not dismissing your very valid concerns, which I actually agree with. It'd suck if most games ended up having stable diffusion syndrome. But not only isn't the tech fully ready, but we don't even know to which extent it can be used with meaningful benefits. I'm more than willing to give it some time, we may even end up surprised with a solution that brings meaningful improvements, yet doesn't need a 5090 (or two) to run.
That's fair, hopefully they make a lot of progress on getting the issues sorted out.
Posted on Reply
#135
Hyderz
Upon closer inspection, dlss5 details appears nice but overall it is very bright? Probably to show off hidden details… also I’ve noticed there is shimmering and ai failed to render on the players shirt on the fifa game… I wonder if this new dlss would take a performance hit or boost fps…
Posted on Reply
#136
FierceRed
It will obviously be tuned but initial impressions are
  • It clearly changed Grace's face, especially the lips
  • It's too sharp/oversharpened
  • Every scene I kept thinking "Where the fuck is the sun in this?" since it actively changed the sun's radiance intensity for most of the games
  • It almost looked like a porn star beautification filter
More style than substance so far but it is a GPU-seller. Interested in what AMD fires back with.
Posted on Reply
#137
Visible Noise
BoxCutterSo not complete control, so nVidia just like lied? This really is just 5070=4090 all over again.
The Venn diagram between Nvidia hating AMD users and people hating on DLSS 5 is a 99% overlap.

Paging Hecate, where are you?
Posted on Reply
#138
BoxCutter
Visible NoiseThe Venn diagram between Nvidia hating AMD users and people hating on DLSS 5 is a 99% overlap.

Paging Hecate, where are you?
I hate AMD too if that makes you feel any better.
Posted on Reply
#139
RejZoR
Visible NoiseRay Reconstruction literally does denoising with AI. This is the new model for it.

There’s no makeup being hallucinated. It’s called physically correct lighting. You’re just used to games having bland, same intensity lighting.

Look in the mirror, are there not shadows under your eyes?
That's absolutely not what's going on here. Her face literally looks like every single "generate me a sexy blonde" and generative Ai spits out exactly this looking blonde woman.
Posted on Reply
#140
theglaze
The problems are obvious, even with YouTube video compression.

Just go frame-by-frame, and every DLSS 5 ON example showcases the same ol' DLSS problems:
  • adds motion blur and ghosting
  • muddy texture clarity
  • poor geometry stability
  • lacks consistency to previous frames
Just more NVIDIA marketing slop

Posted on Reply
#141
oxrufiioxo
evernessinceI can see some people preferring it at first. The problem will then come later when all games start looking similar, as will the characters, as AI has that effect. Go and prompt an AI for an ugly person, it either won't be capable of doing it (because that's not in it's training dataset) or it'll give you it's idea of an ugly person that it will pull every single "ugly" character towards. The same issue happens across every concept. Any word you input the AI has a concept of, it'll pull the input towards (in this case the game data). Even worse if it doesn't have a concept of it, AI does horribly when something isn't in it's dataset as it's not capable of actually thinking.
Yeah, I only liked it ok in Starfield but more realistic characters can't save how wooden the NPCs are in that game, the soccer game is fine if it's improved, and Leon looks ok but the default model is already more than adequate. The rest I'm pretty negative on. AC environments looked more pleasing but it was a pretty brief snippet.

I think this could be good with a lot of developer intervention but we all know a lot of developers will just tack it on, kinda like RT was done in most games. The RE games are a good example, up until requiem RT was horrible in every release from them.

Until I see a shipped version of this I will probably remain indifferent about it. I definitely do not want every game to use it and it's up to Nvidia to make it good enough to change that sentiment. I'm sure we will see AMDs version next year but their RR even in sponsored titles like crimson desert is much worse than the Nvidia alternative.

The one thing I will say is regardless of if I like the look or not them doing this in real time is pretty impressive, I just wish that effort went into making better GPUs by default at prices the average person can afford lol.
GuckySince DLSS3 it is usable and I used it since.
I still see the downside of it, like the "halo" around objects. But it is better then the alternative like lower FPS and TAA.


FGx2 is usable and I use it occasionally.
But MFG x3/4 and higher is not usable. To many graphic errors and input is like walking through mud.

Still DLSS 5 like they have shown is not Upscaling or frame generation, it is more like RTX Remix, Reshade or just an AI-Filter.
I would not like it as part of DLSS. But they can make it work outside of DLSS.
I've liked DLSS since 2.0, but not because I think it's the most amazing thing and Nvidia's gift to gamers. My positivity towards it is mostly due to how badly TAA is implemented into most games.

Framegen even in 2x is just not for me with a mouse anyways. The one scenario I like framgen in is on a large oled using a controller.


I was just pointed out that the online sentiment towards anything Nvidia releases is usually very Negative, and that does not seem to translate to the average person buying their hardware.
Posted on Reply
#142
Harun8
Looks nice for whoever actually likes photo-realism. AAA games are going to slop with this obsession with photo-realism imo. This just seems to get rid of whatever little art direction was left

However in simulation games specifically, such as flight sims, this looks like it could be awesome
Posted on Reply
#143
Visible Noise
theglazeThe problems are obvious, even with YouTube video compression.

Just go frame-by-frame, and every DLSS 5 ON example showcases the same ol' DLSS problems:
  • adds motion blur and ghosting
  • muddy texture clarity
  • poor geometry stability
  • lacks consistency to previous frames
Just more NVIDIA marketing slop

Good thing this isn’t a released version then.
Posted on Reply
#144
wolf
Better Than Native
RejZoRI have honestly never seen a more universally hated technology by basically entire world... Another fine example of DLSS 5...
It's super easy to hate on when people blantantly strawman it by creating their own ridiculously exaggerated, memeified pictures. Sarcasm or not, even you said it's an example of DLSS5.

Is everyone ignoring that developers get control of this? Let me repeat it.

Developers get control of DLSS5.

That means they can use it, or not, in ways they see fit that match their artistic vision.

Maybe the simplest solution is to fast forward to AMD doing the same thing so everyone can agree its good.

Not surprised to see the same few crap on it and DF tho, it wouldn't be TPU without the regular dose of Nvidia/DF derangement syndrome airing itself.
Posted on Reply
#145
oxrufiioxo
Harun8Looks nice for whoever actually likes photo-realism. AAA games are going to slop with this obsession with photo-realism imo. This just seems to get rid of whatever little art direction was left

However in simulation games specifically, such as flight sims, this looks like it could be awesome
First thing I thought as well is Racing games, and flying games, basically anything without humans could be pretty darn good lol.
wolfIt's super easy to hate on when people blantantly strawman it by creating their own ridiculously exaggerated, memeified pictures. Sarcasm or not, even you said it's an example of DLSS5.

Is everyone ignoring that developers get control of this? Let me repeat it.

Developers get control of DLSS5.

That means they can use it, or not, in ways they see fit that match their artistic vision.

Maybe the simplest solution is to fast forward to AMD doing the same thing so everyone can agree its good.

Not surprised to see the same few crap on it and DF tho, it wouldn't be TPU without the regular dose of Nvidia/DF derangement syndrome airing itself.
Especially when Alex clearly pointed out every weakness of DLSS 4.5 and they lament that Nvidia still releases 8GB cards.... Here's the thing something can not be perfect but still better than every alternative it doesn't mean deficiencies shouldn't be pointed out.


My reservation for DLSS5 isn't from the technology itself, it's after seeing how the majority of developers implement frame generation as a performance crutch, or how RT even after almost a decade still get's really bad implementations. I think this will be used as a lazy way to do art and the risk of games looking similar and having no artistic soul is high.

Hope I'm wrong but I guess in a few years when even consoles can do similar things we will know for sure.
Posted on Reply
#146
remekra
wolfIt's super easy to hate on when people blantantly strawman it by creating their own ridiculously exaggerated, memeified pictures. Sarcasm or not, even you said it's an example of DLSS5.

Is everyone ignoring that developers get control of this? Let me repeat it.

Developers get control of DLSS5.

That means they can use it, or not, in ways they see fit that match their artistic vision.

Maybe the simplest solution is to fast forward to AMD doing the same thing so everyone can agree its good.

Not surprised to see the same few crap on it and DF tho, it wouldn't be TPU without the regular dose of Nvidia/DF derangement syndrome airing itself.
Developers are not some mythical creatures that strive to work hard to make the best game they can, with great artism in mind. They are corporations now. If higher ups that signed a deal with nvidia tells developers to use DLSS5 or implement it they will. If it will cut costs that are associated with making better models, with better textures, better lighting. Why do all those things when you can have DLSS5 do it for you. In the process it will make everything look like the same AI generated images, like we have now. It's literally Vibe Rendering.
How, after so many years of those same developers completely neglecting performance and optimization because you have DLSS to improve performance by upscaling from 720p, can someone really believe that this tech will do anything else than just make the games look like same AI slop is beyond me.

I rewatched the scenes on youtube like 10 times now, it simply looks uncanny.
Posted on Reply
#147
Battler624
dgianstefaniYeah we're at the point for some years now where RTX which had a bad reception with Turing is now looked back on favourably because it's understood it was a forwards looking architecture, that allows it to take advantage of 2026 software that has become a major value add. Maybe this is another similar jump. I remember editing the Neural Rendering parts of the Blackwell release and thinking to myself that it will be a hurdle to get timely adoption because some architectures drag their feet and devs can be lazy.
Nah it was just shit at the start and got better with time, not because it was a forward-looking tech or whatever. Neural rendering will be shit at the start and will start looking good by 2030.
Posted on Reply
#148
AusWolf
Finally, some new tech that makes games look better, not worse! :)

The only negative is that there's no word on hardware support.
Posted on Reply
#149
wolf
Better Than Native
remekracan someone really believe that this tech will do anything else than just make the games look like same AI slop is beyond me.
Is it really a novel concept that others hold different opinions to you?

Just like all previous DLSS tech, or indeed genAI, my advice would be that if you don't like it, don't use it.
Posted on Reply
#150
oxrufiioxo
AusWolfThe only negative is that there's no word on hardware support.
RTX 6090 AI extreme edition.
Posted on Reply
Add your own comment
Nov 20th, 2026 15:45 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts