DLSS 3.5: Ray Reconstruction 

Soon he will become a downloadable cursor skin

"aaaaa..."

It is also hilarious 😂

and i love the noises he makes when resizig

100%, it's my favorite quirk of these videos :)

Yep. Along with the destructible physics environments we were promised in like 2008.

Must have been wind

Games that look pretty are easier to market.

In 1995? Unreal Tournament 99's AI bots destroys a lot of modern games.

Enemy AI is left "dumb" intentionally because it would be a pain in the ass to beat otherwise. Computers can beat the best chess players in the world, they're definitely capable of beating anyone in games if left unrestricted.

I was dying when he was bouncing like rays at the beginninng xD

It's functional and creative xD

he needs to edit himself in a professor x wheelchair. preferably the yellow hover one.

I find it super annoying. 2nd video I watched from this guy and can't see myself following him.

Don't let the door hit you on your way out.@@alexgr0111

Damn 😂

LMAO

and never mention it again for another year 🤣

They lost an opportunity to build something faster and universal with FSR.

@@darudesandstorm7002 And then a year after that they'll finally release it, which will guaranteed to be second rate in comparison.

I was a little disappointed when I tried Cyberpunk Overdrive when I saw the artifacts. I expected it to look way cleaner. I hope this fixes that at least in part.

@@Dionyzos Same here. Its honestly a huge reason why I didnt bother using it despite it being so damn good looking. The artifacts/noise when driving or just moving was annoying. I am a person who gets distracted really easy and looks around the screen a lot at random shit so I notice the artifacts a ton.

@@Dionyzos It's a growing field, and still in its infancy, so we have to be patient, especially when we don't know how it can be improved 😅 I'm still shocked we have real-time path tracing in video games. CDPR accurately stated that it's a 'technology preview'. The real deal is (hopefully) coming with Phantom Liberty 💪

@@Chasm9 When I was in school for Game Art Ray Tracing was used in pre-baked lighting but it took an eternity to render and still looked noisy, you'd have to smooth it out in photoshop still. To see it done in real time is ridiculous to me. This was in 2008.

I also hope they fix the artifacting of DLSS when used with TAA, it's subtle most of the time but really ruins the visuals of Cyberpunk for me sub-4k. I think they "fixed" it but still seems to affect some (myself included). There's no real 100% fix to it either that I've found, I've gotten it to stop but it always comes back with game/driver updates or seemingly at random. Anyone who wants to play through cyberpunk immersively in VR is SOL because of this. The game needs to be run in native while also displaying the VR image and the only cards that can keep up are the 90 series gpus without DLSS, and even then they struggle.

@@LibangF15Nvidia guy in the presentation said it's coming to all RTX GPUs.

"Feautres" huh? I prefer the word "gimmick". We will have 15 games within 2 years if we are lucky that actually use it, and 1/2 of those will be Cyberepunk / Quake 2 / Half Life 1 / etc.

Lil bro deleted his comment proving he didn't watched the video 💀

@@TheVanillatech if it improves the ray traced quality and fps in games, why are you complaining?

@@TheVanillatech "Gimmick" I dont think you know what that word means. Maybe AMD should start to be the first to implement "gimmicks" instead of always playing catchup.

Sad part is that without FG you would most likley not be able to play any raytraced games on lower hardware anyways since DLSS3.5 wont be adding any FPS as the guy in the presentation said, only reason it might get a little better perfomance is because of the many denoisers taking some performance in that scene! It's a catch-22.

@@takemeseriouslyplx2124 the 20xx series was an overpriced mistake since rtx works like shit on them. The 30xx series has decent rtx performance only on the top end. But now Nvidia is showing interest in path tracing, meaning that even their 40xx series is being humbled. What I wonder is if they're showing interest because it's the obvious next step or just so they can showcase their AI cores...

However, only the more powerful 4000 series models can comfortably render RT or PT)))

@@AlucardNoir RTX 20 series was not a "mistake" in the way you describe. I can't really talk about the price easily because mining and other reasons affected the prices. But specifically talking about performance, you HAD TO START SOMEWHERE. It's the chicken-and-egg problem. You won't have great software support without the hardware to drive it. And if NVidia put on a crap tonne of extra transistors to make RAY-TRACING better at the expense of traditional RASTER performance then people would complain about the FPS in non-traced games... The ONLY way to do this is what NVidia actually did. You bite the bullet, put in enough ray-tracing hardware to get things started and let the game devs start adding it to EXISTING games. Optimization comes later. If YOU have a better way to do it that makes BUSINESS sense I'd love to hear it. And this is NOTHING NEW in the computer World. For decades we've had new features that weren't efficient. Tessellation. Various anti-aliasing methods. Even AUDIO when it would run on the CPU could kill performance until that became a solved, non-issue. So the trend is this: 1) New feature. Has issues. Need high-end to "brute force" it. 2) Optimizations. Newer, more powerful cards. New feature filters down to mid-range. 3) Further optimizations and newer cards. Lower-end hardware starts running the features.

@@AlucardNoir good, I love it when tech companies push the boundaries and are innovative, bring on the new tech and lets see what wonders it can achieve.

more like NvidAI

That is kind of becoming their primary focus now.

They know they are fast approaching the limit of the economics feasibility of how much hardware can be push so they have to find a way to push it. This also mean they don’t have to compete with amd on pure silicon which is ahead of them in chiplet design

As long as the CUDA API is the industry standard.

@karl5010they are. You just to much up Jensens arse to pay attention

Stick to console gaming if this is to much information for you to take in.

@@WidePhotographslmao let the man have his cake & icing my friend

​@@WidePhotographsI think he is just referring to why can't all RTX cards utilize dlss 3.0 yet they can utilize certain key features provided by 3.5

best they can do is pushing a shitload of fake frames ignoring your input and the games response to that.

@@JustSkram Because all RTX have raytracing capable hardware but only 40 series have frame generation hardware probably

Second one, RTX30xx series does not have FG, it is exclusive to 40xx series

It doesn't reduce blur, that's down to the fact that DLSS renders at a lower resolution and then upscales. This tech is only focused on improving the ray traced lighting.

@@giglioflex Games use less and less ray samples as you reduce the internal rendering resolution, so using DLSS Performance instead of DLSS Quality results in a far noisier and less defined image. You could describe it as surfaces sort of boiling when the denoisers try to compensate. DLSS 3.5 is supposed to solve that.

​@@guspazso it's gonna improve DLSS and not only RTX ?

If you have the performance headroom you can sort of emulate DLAA with DLDSR+DLSS for higher image quality

Can't edit my comment but the point of that would be to get around the limitation that ray reconstruction in 3.5 is currently limited to DLSS and not available using DLAA, like Owen mentioned. As a bonus you could potentially get even better image quality depending on your hardware.

AMDone

Yeah, that one is a bit of a bummer. AMD falling further behind isn't a good thing for consumers in the long run.

It's an improved denoise. It's inherently a slight improvement over what was already there.

Shaders are what give things a 3d feel, not raytracing.

@@giglioflex as a movie buff, i can confidently say that an improved denoiser is NOT just a "slight improvement"

Try Reshade with Superdepth3D, then the game won't just feel 3D, it will be! Of course you must have the HW for it, but for example CP2077 is godlike converted to real 3D. You won't even mind the lowered resolution or little ghosting + artifacts, all those minuses which accompanies the 3D effect/conversion. It's just so immersing. Or Subnautica (kind of indie underwater diving game) converted to 3D is very scary (it's scary even in 2D lol).

Their AI denoiser seems to like adding more light to the scene, than what is there already. It's like bumping up the saturation on your TV. Looks move vivid, but does not actually make the image any "better". One positive is that it'll now upscale the reflections too, but I'd temper your expectations.

@@kadupse Some raytracers reduce the amount of noise before post-processing via multiscale rendering, maybe it would also work for this rendering method.

@@kadupseExactly! in Cyberpunk, It also shows up as nasty ghosting trails. Reminds me of GTA Vice City lol

@@kadupse very immersion breaking in Lego RTX

Ya, it's a processing power issue. When it's obvious it arguably doesn't make sense to even have ray-tracing. At some point you're better off sticking with more traditional raster techniques. You may use Screen Space Reflections for example, rather than Ray-Tracing and then have the reflections disappear at certain locations or camera angles but better that sometimes than having the reflections have a grainy flicker that catches the eye. I can tune out certain raster issues. i can't tune out the grainy flicker of some ray-traced effects that don't have sufficient rays (and/or sufficient denoise etc) and look noisy.

​@@odytricetbf in cyberpunk the default lighting solution is that noisy I'd rather play 30fps Ray tracing than use that mess

He is a man of focus

Real MVP

Daniel is the one take king 👑

Between two math lessons.

Good teachers can improvise well :)

It will be available with Phantom Liberty update.

They have never just "not made something available". DLSS 3 is available for all RTX cards too. The ONE specific feature that uses the hardware only 40 series has is the only thing platform specific. As is with DLSS itself. Everything nvidia GPU with tensor cores has DLSS. They have never artificially limited in any of their proprietary software within their own ecosystem.

And you'll still be poor and mad

and down to 96 bit bandwitch

@@TTx04xCOBRA No, I call one "sucker" if pays prices taken out of someone's ass. I can afford two 4090s. Do you think I will pay such money for the PC part only to play a game on it? 😆 I have 5 other expensive and way more attractive hobbies I can spend this money on. It is the case when the steering wheel, a few 4K displays and hydraulic suspension for the seat make a racing gaming setup. The setup costs more than the entire season of real racing (car included) at some low-cost stock racing cup. In Europe, we have races with older cars which can't cost more than 1000 or 2000 euros. I'm sure it is the worst experience than gaming with 4090. I don't care if you call me "poor". Go back to your mom's basement and play.

@@HanSolo__ you're broke

yes. it's a free upgrade for all RTX GPUs, turing included.

You can! If you look at the chart it says Ray Reconstruction (All RTX GPUs)

it would be logical you can since its will works on all RTX cards, and not only RTX 40 series like frame gen.

If you can play with RT, it will just look better with the same FPS or slightly faster. Path tracing already looks sick, and now they're making it better.

Awesome guys, thanks for the replies :)

Nah AMD will never get close with ray tracing.

apparently there are some "AI" cores on the Radeon 7000 series, but yeah AMD software is way behind Nvidia. They can't really humble them the way they did Intel in CPUs because Intel got complacent for nearly a decade, Nvidia isn't making that mistake, they are just greedy and AMD is following them in that regard as well

Shhh, aware and informed customers are bad for business...

@@thelegendaryklobb2879 Hehe. Tho if these don't become universally supported technologies that developers can rely on they will die out (like physx etc) and that would be even worse for business. That's why Nvidia worked with Microsoft on DirectX Raytracing

kys

The main feature that came along with DLSS 3, Frame Generation, is still limited to the 40 series.

Awwe youre mad lmao

Salty😂

Frame interpolation is nothing new, TVs do it with "frame smoothing"(the soap opera effect) , what Nvidia is doing is actually doing that in-engine instead and inserting an AI generated in-between frame between two frames, so it works great on high end cards because theres more frames to work with, and it's not as good on lower tier cards... You still need a good base performance anyways

It's not just an opinion. This is the actual reason that DLSS 3 is not enabled for older Nvidia cards as well. The hardware structure of the 4000 cards is different and was designed with this technology in mind. AMD's FSR works with an entirely different method that has nothing to do with tensor core algorithms.

Frame interpolation isn't voodoo magic or something Nvidia created. TVs have been doing it for a decade or more using low-end ARM chipsets. Turing and Ampere cards have the so-called 'Optical Flow Accelerator' that Nvidia are offloading the work to. They simply claim that the Ada one is improved, and the previous generation ones wouldn't offer as good of an experience. Be sure to note that they're not saying previous gen cards CAN'T do it. Just that it'd (allegedly) have more artifacts, and so they won't enable it.

@@CaptainKenway Nvidia has said that the Optical Flow Accelerator present in 30 series cards is not fast enough to get any fps benefit out of frame generation. Of course DLSS is backported because DLSS was designed with hardware in mind starting in the 20 series. They can do it but it's not worth enabling because the entire point is to get a significant fps boost. You are correct that frame interpolation is not new. What is new is the proprietary method they are using to do it and the quality of the image that algorithm produces because of the custom hardware modules on the card specifically designed to make this feature work. This is what Nvidia claims. I don't really care either way if it's true or not as I hate Nvidia and will never go back to using hardware from them ever again. I think them bundling all these features under DLSS when they are really separate features is just another example of how anti-consumer Nvidia really is. They do shit like this all the time since the beginning of their company. Insult to the intelligence of their customer.

"works as well"? It doesn't work well, it creates a ton of input lag. I can get the same effect on my smart tv. Frame interpolation is nothing new.

No they can't. SS effects can't show anything off-screen, and have horrible artifacts at the edge of the screen and in combination with occlusion. The AI/lower precision calculations aren't approximations in the same way the old school effects are approximations. The old tricks are doing the wrong calculations to get to somewhat realistic results quickly. The new approximations perform the correct calculations at a lower precision or frequency. Sure, they both have their artifacts, but the new effects will scale with new hardware to greatly reduce those. The artist effort and workflow with the new effects is also greatly reduced. We're not quite at the inflection point yet, but ray tracing will be the norm and the noisy artifacts will go away.

You would have to be incredibly ignorant to think screen space effects can reproduce path tracing effects outside of canned and cherry picked still comparisons. The moment something moves outside of your screen space, it's game over.

@@Niosus That's not entirely accurate. The new denoiser uses AI as does DLSS to upscale. In both cases you have an AI making an educated guess. The only difference is the AI is better at guessing than a algorithm.

@@Niosus I specifically said the effect in the slide deck (the Cyberpunk 2077 comparison screenshot). No RT, nor AI is needed to have an on-screen, bright purple object bleed color onto the concrete below it. And no, the new effects won't scale with the hardware - not if there isn't a fundamental change in GPU architecture design. RT/PT and the algorithms employed, i.e. by UE5, leave more than half of the ALUs empty, doing nothing due to lane masking, as modern shader code essentially became a cascade of "if-else", or worse, "switch" statements - with GPUs near-always having to execute every branch sequencially. We're facing at least another 5 years of games running like shit in relation to raw HW performance and looking blurry as f*ck due devs not optimizing anything anymore and due to upscaling, denoising and other such reconstruction and image synthesis techniques. I'm not having it!

@@Psychx_ Until you look down and all your lighting is gone. Or until there is an object in-between the ground and the light source, and you get this weird shadow around it. In most games, this object could even be your weapon. Do you have any source on your UE5 claim? Thousands upon thousands of engineers at Epic, Nvidia and other places are working on this topic. You're expecting me to believe that a random RU-vid commenter is the only one to figure out that they're only using half the GPU and nobody has bothered to fix it and literally double performance for free? I find that extremely hard to believe. I'm not a graphics engineer, but I am a software engineer. In my experience, every time someone proposes a simple solution to a complex problem, it's just wrong. You can say a lot about Epic and Nvidia, but not that they're technically incompetent.

You shouldve bought Nvidia dude

@@TTx04xCOBRA well, I'm using RTX 3080 now, but the reason I'm waiting for FSR 3, is for the actual competition, and extra option for those couple games that only has FSR

@@setiadisatria1038 FSR 3 is heavily rumored to only work on RDNA 3 GPUs

@@TTx04xCOBRA ahhh, well that's add another reason for people to buy 7000 series card huh? Finally AMD turns to the darkside, just like Nvidia mwehehehehe. Welp, we'll see how it turns out, if it's indeed interesting then..... It's time to actually start the technology wars then, which is GOOD. Might not be good in the short run, but probably good in the long run

@@setiadisatria1038 What technology war? Name a technology AMD released in the last 10 years that wasn't a clone of an Nvidia technology released prior to it? There is no competition, look at the market share and steam hardware survey numbers. Nobody buys amd. Even people who talk shit about nvidia buys nvidia. Btw competition is not necessarily good for the consumers, despite the clichee. Look at cpu market, since the competition sharpened, amd has now raised the priced beyond intel prices in the dominant era. So, (1) don't fall for clichee's and (2) buy what serves your interest best, don't think about anything else.

The Unreal Engine has its own upscaler too.

XESS is not hardware agnostic, there are two upscalers called XESS, one for XMX cores, and one for DP4a (other GPUs). Hence why XESS is better on intel. Until AI accelerators become standardised (fat chance) this will continue.

There's alway been proprietary hardware and software in the PC market, just as much as there has been in the console market. Intel CPUs have always had different architecture and have used different sockets and motherboards than AMD CPUs, for decades. The same is true with Nvidia GPU's and AMD GPU's. Both have always had a different set of proprietary hardware, features, drivers and other software. It's just the way things are. If any of them come up with a new idea, innovation or patent, then they should have the right to choose whether or not to share it out or to close it off from any competition and instead use it as a selling point. For example do you think AMD ought to let Intel use thier Infinity cache and 3D V-cache? Or even let Nvidia copy thier MCM design? Or should they reap the rewards and force the competition into coming up with thier own solutions in order to compete against thiers. Like it or not, proprietary product's and IP's are a very good way of forcing companies to innovate, come up with thier own ideas and compete against each other. And heathy competition is always good thing. From a business prospective making something 'open source' also isn't going to earn you a dime and is also helping out the competitions product, just as much as it's helping your product.

@@fafski1199 @fafski1199 no, you are missing the point by a mile. AMD, Intel and even Nvidia never tried to push proprietary APIs into games until DLSS. XESS is obviously heavily optimized on Intel hardware, or rather the other way around, and so, I guess, FSR on AMD. Despite this, they can still run on other hardware if the developers chose to implement that API in game. I can still test XESS on Radeon or RTX GPUs if I want, or FSR, and see how it compares with other upscaling API if I like, but I will never be able to do the same with DLSS, by design, meaning that code is for exclusive benefit of Nvidia owners, and I pay for it even if I don't own a GTX Card. I don't care what is in the drivers of a Rizen or Intel main board and CPU: the program will work exactly the same on both systems. Like in the old Voodoo 3DFX API times, DLSS its brand tied and Nvidia pushes for developers to code for their hardware specifically. You can see the difference: I can write code for FSR or XESS, and I guess the API designers will heavily optimize their hardware to run better on their architecture, but they don't cut, purposely, off other brands. If AMD is smart enough, it can freely design a GPU which is better optimized for XESS then Intel ones as it's free use software. Intel and AMD will never be able to design a GPU able to run DLSS code anyway because of legal constraints, exactly as it happened for Voodoo dedicated API. The developer should set the API as an overlay for their game to work and it's up to the drivers and the hardware to optimize for it.

​@@LeptospirosiYou don't need to specifically code anything for DLSS... DLSS, FSR 2 & XeSS make use of the exact same inputs that're generated by the game engine. The only thing you need is the SDK's for the upscaler in question but that's not exclusive to DLSS. FSR 2 & XeSS also require their respective SDK in order to work. What is different is that DLSS does not have any inferior fallback for any generic GPU without "Tensor Cores".

It will look pretty realistic bro

could probably do 6-10 ray bounces in fully pathtraced games with 4k textures and 8k screens, though the 8k screen is a bit overkill, more fps at 4k resolution is better, imagine path traced 4k games running at 240fps

first slide shows slight performance increase

@@GewelReal but if you watch the video daniel said that some scenes might but is not the norm this is just a image quality option basically shouldnt lift performance in 90% of the time

From the CP 2077 it seems to do both

same performance but sometimes better. But it mostly just improves quality.

Without Nvidia you would still have a patato in your pcie slot..

Indeed, that sounds like very limited bounces for GI (or none at all & just RT shadows).

Multiple bounces are heavy to compute, so they are limited to 1-2 in games. Portal RTX lets you increase the bounces to see what it can look like.

Someone has to be first mover though, and these features require hardware other than the system CPU to be able to run at a usable performance. Likely rules out a software company led approach. Are you suggesting you’d prefer for these features to just not exist at all? Someone has to be first (it just happens to be Nvidia in this case), then others will follow with their own and it will become a standard with support then added to DX, Vulkan etc.

Are you seriously whining like this? Nvidia is typically the one pushing the tech in this market. AMD doesn't do it. Intel doesn't do it. Someone has to do it first, and it's up to the others to match or exceed that.

who is gonna rent a supercomputer to train AI models for stuff like this and make it open source? AMD hasn't made anything on this scale of quality yet. Microsoft isn't gonna do it for free either, though they have a better chance of doing it for DX12 or a supposed DX13 but don't expect this in Vulkan, Khronos group don't have that kind of money and Apple doesn't care about games either

Yes.

😅

No frame gen : D hehehehe poor AMD

Just 2 more weeks!

Lul. Ever the elephant in the room. "Shhhh, guys, it'll be fine, we've got this new trick just you wait!" No shame.

4080 never was/is the right choice

@@mornedhel1 I can always back but explain

worst price to performance of current gen GPUs if you have the money just get the 4090 @@DaVizzle_Bro

​@mornedhel1 I can understand if your point of view is the one suggesting I invest my money in AMD because I'd be getting more Bang for my buck. But you have to realize, which I think a lot of people are overlooking, is if I get the AMD that is equivalent to the 4080 in terms of performance (ray tracing not withstanding) Then I'm going to have to fork over another 100 to $200 for a 1000 W power supply since I don't have that much with my current setup. Now granted the premium Starfield bundle is VERY inticing with AMDs current promotion I have to say it is still not exactly winning me over. If AMD could come up with a better solution to their FSR ANDDD more importantly to their very big problem with Ray tracing then I might actually be won over since the one thing I will concede AMD destroys Nvidia on is VRAM Capacity which is starting to shows cracks in foundation when it comes to the most recent AAA games at 4k. Keep in mind also that my decision is extremely based off of the fact that I have a 55 inch 4K TV so I have to play games at 4K If I want them to look even halfway decent and not like a PS2 port.

The 4080 is a fantastic GPU, if you have the money. Compared to a 3080 the price to performance is about 15% worse, but it is 50% more powerful. Saved me from having to upgrade my 750W power supply, runs much cooler, and you get the cutting-edge features

Good for me, i have a 4090.

@@pliatI'm sorry for your loss. Nvidia swindled me with the 980Ti. We learn the hard way I guess?

@@thisguy3500 i don't feel very swindled tbh, it's an excellent card.

@@pliat RDNA4 high-end GPUs will come later than those Navi44 and Navi43 because they are aiming to 4x performance of RTX 4090. BTW. RX 7900XTX match 4090 in most 4K titles.

@@pliatExactly how I felt at first, then 5-6 months later the 1080 released. At least with the 4090 there won't be a replacement until 2025, if that's to be believed. They intend to make people salivate for the $2200 RTX5080.

Yes, RT is slow so it forces the use of very low resolutions and a lot of tricks to compensate for it (denoising, upscaling, frame generation, reduced ray bounces, higher resolution rasterized base images, ...).

@@vitordelima The way it was explained is that the rays are driving the image creation rather than a effect on existing textures. Am I incorrect in that understanding?

@@robertmyers6488It should be this way but the denoiser becomes confused with the lack of information and uses a rasterized image with only the geometry and textures to guide itself. This is shown in the slides.

@@robertmyers6488 I alway love it when wannabe smart asses like you talk about stuff they have zero clue, reminds of the people who are afraid of vaccines despite professionals telling them they are safe.

I think native is not using denoisers at all because those go on with dlss and you are at 20 fps so maybe the ray hasnt even show yet hahaha

@@allansolano5587 denoisers should be there in the game engine, no way a RTX 4090 can get those framerates while rendering all that rays (it would be more like a frame every 5/10 seconds not 20fps)

Because it's the hardware accelerated DLSS reconstruction algorithm that processes the denoising. Native only works with a simpler, let's say, stupid denoiser. It should be possible to do this without upscaling in the future, just like DLAA.

@@Dionyzos ok makes sense

There's a whole slew of graphics features we enjoy nowadays that we're originally pushed by Nvidia in a proprietary feature. That it's proprietary now doesn't necessitate it being proprietary forever & since they don't need to make it compatible with everything & everyones hardware you can push things to the masses faster & with lower research costs. It's not ALWAYS because corporations being evil overlords...

Have you been to Las Vegas at night? That's polished concrete. It's going to reflect color.

@@Draanor You think there'd be polished concrete in a dystopian city? I'd have to agree with the OP, the material properties were incorrectly set.

Well given the first image without RT is polished concrete, I don't particularly care about your art design choices, the asset in the screenshot is polished concrete. Complain to the devs. And yes, polished concrete seems right at home is dystopian Las Vegas style city... @@giglioflex

You should look at nvidia's presentation to better understand why the lighting is so different in the screenshot. There's a video at the end where they show all 4 variants of the scene in motion. It's a scene with dynamic heavily colored lighting, so the lighting is changing constantly and is mismatched between all 4 variants, that's why they look so different in the screenshot. What I noticed right away though was that the scene with DLSS 3.5 changes raytraced lighting from the light sources in the scene much much faster, because of (I presume) faster denoising process which doesn't need as much frames for better accumulation of results. The difference is night and day in my opinion. Without DLSS 3.5 the secondary lighting lags behind by at least half a second and you can see how slowly it reacts to the changes in light sources. With DLSS 3.5 it happens almost instantly like it should. Basically DLSS 3.5 effectively provides faster light bouncing in the scene, which is much closer to what happens with light in reality (instant change). This is very promising. It's not yet completely lagless, but getting close to it. Also the reason the lighting looks more muted in the scenes w/o DLSS 3.5 is because it doesn't have enough time to fully saturate the scene because the light source changes faster than the GPU is able to fully process changes in RT lighting. You can see it clearly in the video, how it slowly and gradually changes from yellow-ish to blueish and back again, but never has the time to change completely to the new lighting state before the light sources change again, so it's always sorta kept half way. In DLSS 3.5 version because the light is bouncing much faster, the scene saturates fully almost instantly like it should.

I thought that same thing...

My understanding: Before denoisers composit an image, engine needs to cast and sample rays for lighting - this process takes time and you need many rays for good looking results, ReStir helps with this issue, but the result is still noisy and at this step NVIDIA's new denoiser starts to create an image that in theory will better represent something like a game. Denoisers fill in missing pixels that's why the result looks different

@@antont4974 I see, thanks for the explanation. ^^

I think in another generation or two, Intel is going to be a serious competitor based on what we've been seeing with their driver improvements.

intel has spyware drivers same for ngreedia.

NVIDIA should spend more time making their GPUs better instead of forcing us to use an upscaler and fake frame generator to give the illusion of smooth gameplay.

And still you can do the ”same” thing with traditional ways… With Nvidia 1080 and other older GPUs… By hand coloring the images etc…

@@antimsm6705 this not a illusion of smooth gameplay whatsoever. This is genuinely a way to improve Ray tracing as a feature which will be used more and more in coming games. Sure frame generation is a gimick but this isn't. DLSS 2 is definitely not a illusion. It's a very good technology. The game developers are simply exploiting this feature to hide their garbage optimization. It's not all Nvidia's fault. (I do agree that the 4000 series is bad)

@@haukikannel That is not what ray tracing means at all....... It's completely different from what u are saying. The very point of ray tracing is to remove the need for such hand tuned traditional methods and decrease the effort on developer side and increase the workload on hardware side to making realistic lighting. U are right such *effects* can be recreated using traditional methods but it takes a insane amount of work to do it and is very time consuming. Even then it won't be completely realistic. Nvidia is trying to make lighting in games become realistic without having to put a insane amount of work on it and more accessible to every developer

I disagree. A proper GI implementation of RT is actually making a big difference in immersion. You are probably talking about implementations that can barely be called "RT" like some shadows and some reflections here and there, which is most of the "RT" games today to be frank. But its still a step over pure raster, its like the new "ultra".

Pretty much the same as frame generation, as long as one method becomes available as open source or multi format...

but why? are you not entertained?

Agreed

Ray-Tracing has been added to many games AFTER the game was designed WITHOUT ray-tracing in mind. And many raster shadows and reflections look GREAT already when used in appropriate situations. If you have Screen Space Reflections for example, then maybe the MOUNTAINS reflect on to a lake. It looks great. You don't really register that if you're far away. And if the game never lets you get CLOSE it's a non-issue. But if you DO get close then those reflections can disappear. There are similar issues for all shadows and reflections... also, in the FUTURE it will eventually be far, far simpler to design a game if you ONLY design it with ray-tracing. Because you don't need to keep baking the lighting. You're building the game, you move the objects around in real time, you change the light color etc and the game engine just re-calculates all the bounces (so shadows/reflections are just automatically created like in real life). So part of this is using ray-tracing where it makes SENSE visually that justifies the performance hit, and part of this is about making it easier for the game developers (eventually). ALSO... there are certain things game devs SIMPLY DO NOT DO because raster techniques fail. One example would be a crystal cave with a lot of reflections AND refractions. So light bouncing off and going THROUGH a lot of objects. you can create those areas, but they don't look like they would in real life. And they don't currently do it with ray-tracing because it would cost too much. So there ARE situations that can only be done WITH ray-tracing and only WHEN it's computationally feasible. (Pixar knows how computationally expensive every frame is when generating animated movies. When they did "Frozen" the most computationally expensive scene, which was calculated with path tracing, was in the frozen castle due to all the light bounces needed to make it look correct.)

@@photonboy999 You remind me how musician theorists talk about why one song is great... and another is not so great... while I just like both. Sorry folk. You seem to not get it. You want me to appreciate that one is better than the other and follow your opinion. I just DONT. For me the "better" graphics often just isnt better than the already very good one. I like both. For me they are just different.

Yeah, the best way would be upscaler API included in DirectX12 and Vulkan and supported by all vendors Doesn't seem to be going in this direction now, though

It's denoising.

Nah, that's not true anymore. Intel has its own "tensor cores" called XMX and uses it as an excuse to vendor lock best upscaling method of XeSS to Intel only GPUs. AMD has AI acceleration tech included in radeon 7000 series. Not used for anything except for compute/AI/ML right now. But there's a suspicion it'll be used for FSR3. So yeah, Nvidia is just doing their usual vendor lock schemes.

@@ZoragRingael There’s no standardisation on the hardware though. Unless you are suggesting that Nvidia don’t invent new ways to break from the clock speeds/fab/bus grind to get more performance, there’s not really a viable choice other than to develop for their hardware. Also sounds like it’s an industry-wide problem if Intel are also using their XMX in a way to lock in and AMD may follow for FSR3. Maybe we need DirectX AI/ML in 13 these new features can build on top of.

@@alwatts there weren't any standards for RT either, until Microsoft came with DXR and Vulcan community came up with Vulkan RT Same needs to be done with upscaling tech. --- >Unless you are suggesting that Nvidia don’t invent new ways to break from the clock speeds/fab/bus grind to get more performance, Nvidia didn't. Nvidia is now using newest and smallest node and hugely benefits from it. As for RT/AI tech in their RTX GPUs, the reason for that is following: - Nvidia uses same architecture for both gaming and enterprise/server/compute - Nvidia introduced tensor cores and RT tech for their enterprise consumers. DLSS and gaming RT were afterthought, because tensor cores and RT stuff take die space gimping Nvidia GPUs raster performance. So Nvidia needed a way to turn it into their advantage. AMD doesn't have that problem since they have separate architecture for enterprise/server/compute which includes AI/ML. If anything, it's AMD who's trying to overcome limitation of tech processes by utilizing chiplets tech.