Tuesday, July 21st 2026

NVIDIA Shows DLSS 5 Progress and Technical Details at SIGGRAPH 2026

NVIDIA has once again showcased its DLSS 5 neural rendering at the SIGGRAPH 2026 event, currently taking place in Los Angeles. Despite the initial negative reaction from gamers, NVIDIA is continuing to develop the technology, offering improvements and a new demonstration of how DLSS 5 is intended to function. One of the most interesting aspects is that game developers can choose from three DLSS 5 models, each providing a different level of structural intensity, global illumination, texture detail, and more. These models, labeled Model A, Model B, and Model C, can be tailored to specific scenes or character needs, allowing developers to mix and match according to their requirements. Each model carries a different parameter count, meaning that some are more compute and memory-intensive than others. However, they generally fit and run fine on a single GPU without any problems.

Another noteworthy feature is the preservation of artistic intent, which was heavily criticized in the original DLSS 5 demo. Each individual object, scene, and game setting can be customized according to the developer's vision. This is one of the challenges NVIDIA faced with DLSS 5, as maintaining visual identity, object semantics, and character poses, along with output lighting semantics, motion, and art direction, is crucial. Within these strict parameters, DLSS 5 applies material photorealism to enhance visual appeal. Below you can even see the customization of the applied model, where there are adjustable parameters that can be done per scene and per character.
The second challenge that the company faced is temporal streaming. Traditional generative AI video generators process in chunks, reasoning across a frame and generating a batch of frames at once. However, real-time rendering requires immediate generation, frame by frame. Since DLSS 5 uses motion vectors from the game engine, it can generate additional textures instantly, without delay. This results in no shimmer or image corruption, as DLSS 5 know exactly where to attach itself and perform its texture generation instantly.
The final challenge for DLSS 5 is speed. To render a 4K frame with 8.3 million pixels at over 60 FPS, NVIDIA achieves this in less than 16 ms, with most of the processing time consumed by the game engine rather than the DLSS 5 model. The model is so efficient that users will not notice any delay during gameplay, and it operates on a single GPU. While NVIDIA highlights VRAM efficiency, it remains unclear whether 4K gaming will be limited to high VRAM GPUs or something else. As the launch is expected this fall, the company will provide general system requirements and further details, but the improvements compared to the original demo seem promising for now.

Source: NVIDIA at SIGGRAPH 2026
Add your own comment

153 Comments on NVIDIA Shows DLSS 5 Progress and Technical Details at SIGGRAPH 2026

#76
MTS-C
tsurugiDLSS 4.5 RR can solve your problem. coming in August
I highly doubt it.
Posted on Reply
#77
tsurugi
evernessinceReshade can gain access to things like depth buffer and there are full reshade ray tracing shader available.

You are definitely underselling it.
I really don't know much
evernessinceIt's not that said effects aren't traditionally achievable. RT, PT, subsurface scattering all are in realtime without the need for DLSS 5.

It's the "quality" of those effects. AI has a fixed cost for high level of lighting fidelity. The problem is that it carries of large variety of other issues with it. Right now those lighting effects are bundled, you can't choose just subsurface scattering. AI has bleed issues as well, where it'll pull outputs towards it's common understanding. It's why you still see the front facing lighting effect, because AI is trained on images with that kind of lighting.
it cannot be controlled precisely, so masks and sliders are needed to filter out excessive effects. When Nvidia first announced DLSS5, masks and sliders were already expected, and this showcase confirmed it.
evernessinceThe problem of not being able to unbundle specific effects is inherent to AI. The neurons are all connected and one input can pull on many neurons. I have to assume as well that in order to keep performance reasonable, understanding is limited. It's not going to be on the level of a model dedicated to identifying things visually. Originally it probably wasn't identifying things at all but merely coloring pixels as that's all it had beside motion vectors. Them feeding it more game data will absolutely help. The point of a model like this isn't to deeply process the data, it's to get it through the neural network as quickly as possible with the highest possible output aesthetic quality.
That's just how it is, this is indeed a drawback of DLSS5. It is a genAI positioned in post-processing stage, and if this is the best current technology can do, I have no problem with it.
Posted on Reply
#78
JustBenching
ChomiqThe point is that we don't get progress in raw RT and raster performance and instead we get crutches like DLSS.
DLSS isn't a crutch, it's one of the biggest advancements in image quality that ever happened. 4k dlss looks so much better than 1440p native while having the same performance. That's a net positive, is it not?
Posted on Reply
#79
medi01
Recus(FSR x8 frame generation)
Was bullcrap when The Filthy Green did that and remains bullcrap when AMD implemented it.
AleksandarKa different level of structural intensity, global illumination, texture detail, and more.
My god, they still double down on lying about it. "Illumination" my jensen.

Posted on Reply
#80
MTS-C
JustBenchingDLSS isn't a crutch, it's one of the biggest advancements in image quality that ever happened. 4k dlss looks so much better than 1440p native while having the same performance. That's a net positive, is it not?
DLSS quality at 4k renders at 1440p....."so much better". Y'all really drank the koolaid.
Posted on Reply
#81
RejZoR
Only reason DLSS5 can do more than ReShade is because it taps into depth buffers directly through official engine implementation where ReShade taps into depth buffer through interception, so it often doesn't work or doesn't work well and is limited to screen space, but regardless, it's free and makes insane visuals that you can craft yourself the way you like it with countless shaders. Some really high end shaders like Marty McFly's RTGI adding global illumination using ray tracing are actually pretty insane. Years ago I've played few Tomb Raider games with his RTGI and it added lighting almost the way you'd expect it from ray tracing even if it was just screen space based which means you can't see effects that aren't on screen as it can't calculate them off screen because they don't exist for the ReShade. But still, the effects were insanely good.

My traditional go-to is CAS, FakeHDR and Ambient Light shaders. When tuned right it gives sharper image with darker less washed out colors that are annoying me so much in all games and not even native HDR solves it properly. This kinda does and I really like it.

What I'd actually love more than DLSS 5 would actually be official API for ReShade so people could do it for free using ReShade with official engine integration so you could use it in multiplayer games (without depth coz then you could wallhack) or without a lot of fiddling officially. Then gamers would have total freedom and full ReShade support, theoretically not limited to screen space and not limited to one graphics card brand.
Posted on Reply
#82
Dr. Dro
medi01Was bullcrap when The Filthy Green did that and remains bullcrap when AMD implemented it.


My god, they still double down on lying about it. "Illumination" my jensen.

At least try to be honest in your argument, this is an old screenshot from the reveal. Who am I kidding, you're already throwing names.
Posted on Reply
#83
CornyOC
SLI will return, one GPU for rendering and one for AI features.
Posted on Reply
#84
Lionheart
RecusWatch AMD fanboys attack this but suddenly it's must have feature when AMD has alternative. (FSR x8 frame generation)
2nd comment in and we already got an instigator like you with the "fanboy" crap.
Posted on Reply
#85
JustBenching
MTS-CDLSS quality at 4k renders at 1440p....."so much better". Y'all really drank the koolaid.
Having eyes = coolaid

You have to have terrible eyesight in order to not notice the obvious difference. Just compare screenshot Oz from tpu itself, they have plenty of side by sides. 1440p native looks like garbage next to 4k dlss.
Posted on Reply
#86
Trunks0
JustBenchingDLSS isn't a crutch, it's one of the biggest advancements in image quality that ever happened. 4k dlss looks so much better than 1440p native while having the same performance. That's a net positive, is it not?
Yes/No. I mean DLSS's reason for existence was to help with performance so they could manage RT better and make use of the AI features nVidia wanted in the architecture that where primarily for enterprise use. It mostly panned out, but it took awhile for it to really hit it's stride. I'd hardly call it the biggest advancement in image quality that ever happened though... just off the top of my head for a few...

- Hardware texture filtering
- Shaders in general (Vertex, Pixel, Programmable, Mesh, Geometry etc etc etc)
- Hardware Tessellation/Displacement mapping
- Texture compression
- Detail textures
- Fullsceen Super-Sampling Anti-Aliasing
- 32-bit color
etc etc etc etc

DLSS probably wouldn't make a top 10 of biggest advancements in image quality ever. It's great, don't get me wrong... but it's at best the greatest advancements in Temporal AA and upscaling. Maybe one of the greatest use's of AI in gaming so far 👍
Posted on Reply
#87
JustBenching
Trunks0Yes/No. I mean DLSS's reason for existence was to help with performance so they could manage RT better and make use of the AI features nVidia wanted in the architecture that where primarily for enterprise use. It mostly panned out, but it took awhile for it to really hit it's stride. I'd hardly call it the biggest advancement in image quality that ever happened though... just off the top of my head for a few...

- Hardware texture filtering
- Shaders in general (Vertex, Pixel, Programmable, Mesh, Geometry etc etc etc)
- Hardware Tessellation/Displacement mapping
- Texture compression
- Detail textures
- Fullsceen Super-Sampling Anti-Aliasing
- 32-bit color
etc etc etc etc

DLSS probably wouldn't make a top 10 of biggest advancements in image quality ever. It's great, don't get me wrong... but it's at best the greatest advancements in Temporal AA and upscaling. Maybe one of the greatest use's of AI in gaming so far 👍
The main difference is most if not all of these advancements could not work on older gpus, while on the other hand you get better and better upscaling models on 8 year old gpus.
Posted on Reply
#88
Trunks0
JustBenchingThe main difference is most if not all of these advancements could not work on older gpus, while on the other hand you get better and better upscaling models on 8 year old gpus.
Sure, but that's really just shifting the goal posts. I"m not saying DLSS isn't great, it is... but you made a blanket statement that was just a little silly, it's all good :)
Posted on Reply
#89
oxrufiioxo
This really doesn't change my perception of this technology. The two things I care about still haven't been answered, How does it hold up in actual gameplay? If there are noticeable artifacts, what's the point? And what is the performance cost? The fact that they keep saying they'll get it down to a single GPU doesn't really say much if "a single GPU" means a 5090 or nothing.


That being said, this presentation isn't as bad as the original, but it still doesn't show anything that would justify taking a massive performance hit.
Posted on Reply
#90
MTS-C
JustBenchingHaving eyes = coolaid

You have to have terrible eyesight in order to not notice the obvious difference. Just compare screenshot Oz from tpu itself, they have plenty of side by sides. 1440p native looks like garbage next to 4k dlss.
We've already been over this. Do it in motion and you'll see the difference. If you can't then you're either lying to yourself or you're the one who needs glasses. DLSS is a solution to the problem, the problem was RT that nobody asked for. We shouldn't need to render at lower resolutions to achieve what we already had prior. If DLSS is fine for YOU then so be it, but don't pretend it's better when it's not. These technologies are a crutch to try and retain performance figures we already had before Nvidia tried shoehorning RT too early into the industry. The only advantage DLSS brings is DLAA but that's a mess trying to cover an even hotter garbage that is TAA. A spade is a spade is a spade, and trying to say "4k DLSS performs like 1440p is net positive" is laughable because of course 1440p performs like 1440p. Does it look slightly better than native 1440p? It should when you add some tech to it but better than native? Don't kid yourself.
Posted on Reply
#91
mrnagant
So looking at those DLSS 5 photos, there is really nothing going on. Yes, there are differences, but it already looks perfectly fine without it. So seems like loads of power is going to be needed for very minimal net positive change in presentation if litterally anything. And that presentation will be 100% subjective, but still look good on or off. The lighting is a little brighter, but is that needed in this scene? Should the scene be a touch darker? It is ok for stuff to be a touch darker. So in these screen shots you can't really tell if a slightly darker look is more appropriate. But then it may not even really matter for this particular scene.
Seems like we are hitting diminishing returns, each iteration requires a factor of additional horsepower, and the differences are more subjective heavy than objective heavy.

Kinda like going from 8x MFG to 12x MFG. 1x to 2x to 4x are huge and require some good AI power behind them, but then the more you add it just diminishes your returns.

Games need to start coming out with Game Maker Mode settings preset, like TVs have Film Maker Mode, so you can play the game to the original intent of the developer.
Posted on Reply
#92
Youseethis
JustBenchingHaving eyes = coolaid

You have to have terrible eyesight in order to not notice the obvious difference. Just compare screenshot Oz from tpu itself, they have plenty of side by sides. 1440p native looks like garbage next to 4k dlss.
MTS-CWe've already been over this. Do it in motion and you'll see the difference. If you can't then you're either lying to yourself or you're the one who needs glasses. DLSS is a solution to the problem, the problem was RT that nobody asked for. We shouldn't need to render at lower resolutions to achieve what we already had prior. If DLSS is fine for YOU then so be it, but don't pretend it's better when it's not. These technologies are a crutch to try and retain performance figures we already had before Nvidia tried shoehorning RT too early into the industry. The only advantage DLSS brings is DLAA but that's a mess trying to cover an even hotter garbage that is TAA. A spade is a spade is a spade, and trying to say "4k DLSS performs like 1440p is net positive" is laughable because of course 1440p performs like 1440p. Does it look slightly better than native 1440p? It should when you add some tech to it but better than native? Don't kid yourself.
In motion dlss is like this (grand theft auto 5 enhanced)



The car has many many exhausts.





The artifacting is less pronounced but still visible in motion. Just look at the rear bumper.



EDIT, to be fair...



At this point it's becoming hard to tell.



Posted on Reply
#93
wolf
Better Than Native
Dr. DroWho am I kidding, you're already throwing names.
A few TPU posters seem determined to do more damage to AMD's growth than Nvidia ever could. It's hard to imagine Lisa Su looking at a forum full of insults and thinking, "Yes, this is how we win new customers." If your argument needs name-calling, you're probably not helping the team you think you're helping.

It's awfully like the stereotypical militant vegan who spends more time attacking meat-eaters than positively promoting veganism and positively engaging with non-vegans.

People don't just reject the message, they decide they don't want to be associated with the messenger.
Posted on Reply
#94
wheresmycar
Every time someone says "slop" my stomach inflates another inch. I literally ate 4-5 hours ago, went for a 20 min walk and felt great. Then I opened this thread, I'm bloated again.

I've reached a point where slop is officially my most hated word. It sounds like someone took a dump and slipped on it.

Getting back to business: If the full DLSS 5 feature set lands on 50-series, thats a big plus in my book. I haven't really felt the need to use DLSS at 1440p so far but the direction the industry is heading makes DLSS (and other upscalers) less like an optional extra and something we'll all end up relying on. At the moment for me its either native quality or where applicable, upscaled implementation which is extremely close to native.
Posted on Reply
#95
MTS-C
YouseethisIn motion dlss is like this (grand theft auto 5 enhanced)



The car has many many exhausts.





The artifacting is less pronounced but still visible in motion. Just look at the rear bumper.



EDIT, to be fair...



At this point it's becoming hard to tell.



These images show part of the problem, but I have found it's almost impossible to capture the full extent in still images so a lot of it gets a pass in debate vs reality. Ghosting, blurring, noise, flickering, only to try and be covered up further by more tech.
Posted on Reply
#96
JustBenching
YouseethisIn motion dlss is like this (grand theft auto 5 enhanced)



The car has many many exhausts.





The artifacting is less pronounced but still visible in motion. Just look at the rear bumper.



EDIT, to be fair...



At this point it's becoming hard to tell.



If I post 10 pictures from 10 different games (not just one) that native has ghosting, blurring, dancing backgrounds etc. do I win? It's such a dishonest argument to use, everyone can cherrypick a scenario here or there. On the bigger picture, 4k dlss embarrasses 1440p native while offering similar performance. So if you are not implement DLSS one way or another, you are losing image quality

Mini comparison from TPUs screenshots, fps is similar (35 vs 36). Just look at the billboard, the health warning text. Horrifying


And this is the difference between 4k native and 4k dlss

Native = 18 fps


DLSS = 35 fps



Double the framerate, better image quality. I guess im just blind, ill just check with the doctor
Posted on Reply
#97
EsliteMoby
JustBenchingDLSS isn't a crutch, it's one of the biggest advancements in image quality that ever happened. 4k dlss looks so much better than 1440p native while having the same performance. That's a net positive, is it not?
DLAA 4 on many UE5 games looks so blurry it's worse than in-game TAA. Why everyone is praising this technology is beyond my understanding.

Also, DLSS 5 would just be another temporal slop.
Posted on Reply
#98
Frizz
EsliteMobyDLAA 4 on many UE5 games looks so blurry it's worse than in-game TAA. Why everyone is praising this technology is beyond my understanding.

Also, DLSS 5 would just be another temporal slop.
NVIDIA actually gave us something why you hating so much on this lol.. we got given ketchup for our food, we can use it or not use it. Some people don't like ketchup at all, some flood their dish with it, its subjective.. It really depends on the game for me, STALKER 2 for example looks like poo poo without DLSS. Stellar Blade running native even with TAA introduces jaggedness, DLSS smoothens out the details at DLAA.

Also not everyone plays on a beefy graphics card, Laptops and handhelds need DLSS and Framegen.
Posted on Reply
#99
Onasi
EsliteMobyDLAA 4 on many UE5 games looks so blurry it's worse than in-game TAA. Why everyone is praising this technology is beyond my understanding.
First, unlikely. Default UE TAA and TAAU is, frankly, awful. Secondly, a lot of blurring in modern UE5 titles has less to do with any AA and more with just how undersampled to shit everything is. Try and find a UE5 title that allows fully disabling AA (or do manually through .ini tweaks). The game will, most likely look absolutely fucking atrocious. Resolving this shit via any temporal algorithm will lead to blur.
Posted on Reply
#100
qlum
Is it just me or does dlss5 still look worse, less grounded?
Posted on Reply
Add your own comment
Nov 19th, 2026 12:23 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts