NVIDIA Did It: Ray Tracing 10,000 Times Faster!

[@Rock1290T]

It's incredible how rapidly they were able to get pathtracing onto consumer hardware!

[@RicardoM492]

Where the money is, the solutions comes after

[@StiekemeHenk]

I mean, the games they introduced were quite simple with nearly no polygons.

[@ollllj]

took em 30 years too long if you ask me

[@link99912]

@@StiekemeHenk The performance cost of increased geometric complexity scales logarithmically with pathtracing, rather than linearly with rasterization. Meaning geometric complexity is not nearly as much of an issue- the pathtracing part itself is most of the performance cost. The proof is in the pudding, too. Games like Cyberpunk 2077 are now nearly completely path traced, and easily running in real time. You can also see some of Nvidia's recent tech demos with fully path traced fields/jungles and the like. (I think it was one of the GTC or Siggraph presentations, but I can't remember off the top of my head).

[@Chuck.1715]

exactly 3 years, and almost three generations of GPUs, to beat GTX series GPUs in regular tasks appart from RAY tracing

[@exapsy]

I never actually realized we never simulated the "wave" part of lights. Crazy. Such a small detail that creates such a big difference.

[@Roxor128]

I wonder if these results might be useful for audio as well?

[@chasethevioletsun9996]

@@Roxor128 How? DSPs have been processing audio in real-time for decades.

[@TheEclecticDyslexic]

​@@chasethevioletsun9996I think he is talking about audio in games and having it properly reverb inside a space based on the polygons in the scene.

[@chasethevioletsun9996]

@@TheEclecticDyslexic that's an interesting idea, and already doable. There is software and hardware for analyzing performance spaces like that. I have doubts as to how artistically desirable that would be in a game though. Demo scene or indie, sure. Edit: I got the paper for this and the notion of applying it to 3d sound spacialization is actually pretty intriguing.

[@smackanoodle]

we've never had a rendering program that uses wave mechanics to simulate light before no, but you can simulate the wave part of lights haha, you could probably do it in desmos if you wanted.

[@viktortheslickster5824]

With the focus of tech journalism being on AI and LLMs, it's great to hear Karoly talk about a different subject which happens to be his passion - Light Transport Research!

[@thethiny]

It's his area of expertise. That's why.

[@sammikinsderp]

I 100% agree. This is just as exciting as any LLM news!

[@claytoncowen3428]

​@@sammikinsderpwheelbarrow was wax ex ex ex ex ex ex cc RC FC ex ex acc renewed web EV RV for the antennas the same day I think it

[@sammikinsderp]

@@claytoncowen3428 you having a stroke?

[@xl000]

tech journalists love LLM because they can type anything they want in the prompt and it will give them free material of good quality, especially if the question is vague enough that the answer is unverifiable. That's why. It also tells a lot about the person asking the question. When it was in the news for the first time, the only question that one Fox News host could think of was: Who is Will Cain (her colleague), just to get basically the first paragraph from his wikipedia entry.

[@Ilya_Yatsko]

What a time to be alive! Crazy how far technologies have come.

[@PuppetMasterdaath144]

degenerate

[@polygonalcube]

One video where he didn't say it.

[@Kapanol97]

Imagine in a decade or two, lol

[@jackbequick]

@@Kapanol97 Literally can't

[@sigmamale4147]

Truly amazing how they inflate gpu prices because of stupid features !!

[@agustinlado]

Saw this at the Nvidia Keynote, and I couldn't believe more people weren't talking about it. The pinnacle of computer graphics has been achieved in real time and I didn't see any mention of it whatsoever; how could this be!?

[@alexmcleod4330]

It isn't really realtime and it isn't really a pinnacle. When the paper mentions realtime or interactive framerates, it's with a single sample per pixel, which is extremely noisy, but still indicative of how it'll look if you let it cook for longer. For example, their fully-converged render of a compact disc took *a whole day* to render with their method. The paper says 42ms per SPP, which is the realtime part, but two million SPP for the final image, which is over 23 hours. So they've introduced a faster and more accurate way of rendering those spectral shading effects, which is commendable, but the only people who are going to keep talking about it are people who really really want to render compact discs.

[@jazzman7320]

@@alexmcleod4330your short-sightedness is only impressive to the short-sighted. You’re leading a pack of blind people in your belief that this tech is only good for rendering CDs. You probably would have told us that computers are only good for storing text and breaking codes.

[@Slav4o911]

For some reason the narrative is.... we have to hate Nvidia.... that's why nobody mentions any of their revolutionary technologies.... we even have to abandon ray tracing.... "because it's a gimmick" according to certain people.

[@daveinpublic]

Truly is a great breakthrough. Really is a great time to be alive!

[@alexmcleod4330]

@@Slav4o911 this doesn't have much to do with NVidia. It's an academic paper, and the source code is published. I'm sure NVidia have sponsored it one way or another so they can put their logo on it, but it wasn't invented at an NVidia facility and it doesn't even require NVidia hardware. If it was 'revolutionary' in any respect (it isn't, but if it _was_), it would be proprietary.

[@MuhammadHosny0]

the cuts in audio hurt my fucking brain wtf

[@nixel1324]

Honestly, all the videos about AI advancements were great, but videos like this will always be my favorites: videos about hand-crafted advancements in semi-niche topics (especially ones related to light transport, of course!)

[@Kknewkles]

I asked for a graphics papers video, and you delivered 🙏

[@TwoMinutePapers]

🙌

[@pawnix4122]

I have never built my own path tracer, but I have built a fully functional PBR render inside 3ds Max. With this said, I would imagine it is MUCH easier to render the rastorised image first, then render a few hundred path traces rays of light and superimpose the path traced layer on to the raster layer for a very fast and good looking image, depending on the number of rays per frame. This is as opposed to attempting to turn Cycles into actual 100% real time with 0 artefacts and all that. There would be more to it than just composing two renders, but it is much too difficult to write out in a youtube comment. I'd have to write a paper on the subject. For example, I could render the raster version without transparency, and or with limited transparency, and in the path traced layer, I receive the reflective properties of glass and transparent values on grass or hair cards and so on, including baked transparency on textures and what not.

[@st0ned_pixel]

Nagyon élvezem mindegyik videódat! Folytasd. Nagyon szépen köszönjük!

[@nathancreates]

It's crazy how all the Nvidia papers seem to be making such incredible advancements to light simulation. I see a new video with Nvidia in the title and I know it's gonna be a good one.

[@Ila_Aras]

This is incredible! I always rush to watch any videos raytracing related, or or light simulations in general. This is incredible work.

[@Linventor]

can't wait to see this tech appear in games

[@deadpianist7494]

Looking at gaming industry right now, Sorry to disappoint you that it will take 1 or 2 decade to get this in games 😢

[@Linventor]

@@deadpianist7494just in time for Star Citizen.

[@intermsofreality]

You're gonna be waiting a while.

[@CremeeCactus]

​@@deadpianist7494I think a decade sounds right but two would be way too far. Gaming industry only cares about "graphics" nowadays. It's soo much overload, to me graphics could be shite and I'd still play it as long as the game was fun. Nowadays everyone pushes out over polished pieces of shit games. It's like your playing hyped up demos.. that never fully came to life. We are just being fed trailers of what they would want the game to be like. Sorry if I ranted or didn't make sense, but with the speed and over emphasis on "graphics" I think we will definitely get this in 7-10ish years

[@MegaLol2xd]

Can't wait to buy rtx6090 lmao

[@AllisterVinris]

I imagine the discussion in Nvidia research lab. Cultured scientist :"Bro did you know Bungee gum had the properties of both gum and rubber ?" Other cultured scientist:"Yeah bro, I knew. And did you know light has the properties of both wave and particle ?" CS: "Wait ..." OCS: "Oh shit !"

[@AllisterVinris]

@@googleuser4720 Yeah well why are you telling me though. It may not actually be 10000, but an upgrade is an upgrade anyway

[@ABaumstumpf]

Hisoka Scientist ... that would be frightening.

[@AllisterVinris]

@@ABaumstumpf Indeed, AI would be the least of our worries I think

[@666Azmodan666]

it's all old "claustics" is 20 years old if not more, the only novelty is gpu power giving the ability to transfer it to real time in a castrated version.

[@pneumaniac14]

I was just thinking yesterday what it would be like to have a wave simulation instead of ray tracing, but decided it just wasn't a reasonable thing to compute, and would obviously take drastically longer, and any corners cut would dimish the quality too much. This video feels like some sort of divine irony for me.

[@TheMrKeksLp]

This video felt a bit unclear. In the paper for the wave simulations it still said that the CD took 42 ms to render. Certainly an improvement but at least an order of magnitude too slow for realtime applications still. And honestly, why would games even care about physically accurate "wave like" rendering? We can barely do 1 secondary visibility ray trace per pixel on current hardware! Most games trace half a ray per pixel on average for reflections for example, and that's just one query, not even full path tracing yet, where the light bounces multiple times

[@alexmcleod4330]

Yeah, I think Károly got confused there... In the paper it says the CD took 42ms per sample per pixel, but cumulatively it was two million samples for the final frame, which is over 23 hours. Their other examples took minutes or hours per frame, as is usual with path-tracing, because it's still just path-tracing, but with added waviness to its ray bundles. I don't know how many frames per day would be acceptable for a modern game, but I'm guessing most players would prefer at least a few frames per minute.

[@cortical2digital]

I wonder if this can be applied to sound propagation?

[@jefflayton153]

I had the same thought

[@goochipoochie]

Humans have a very limited hearing capacity, it won't make a difference to how we perceive it Our vision capabilities are much more detailed hence the focus around them

[@Humanaut.]

​@@goochipoochiewill still be cool to also have advancements on the acoustic front. Also there is quite a large range within humans: professional musicians or deaf people will have a much more precise hearing with more depth and nuance than the average person. But yes, generally I also used 15$ cable earphones because for everyday use they do just fine and it doesn't matter if I wreck them.

[@JorgetePanete]

There's a Microsoft thing for UE5 that makes something for wave accoustics, I don't know how

[@grahamthomas9319]

Simulated sound design is an interesting idea. I have found most programs focus on the visual aspect. I think a lot of work could be done in this realm. The human brain filters alot of information as it comes in. In our experience our focus determines what we hear. In something like a more guided form of media, film, sound design allows you to intentionally carry focus, however if you wanted to truly simulate reality then deep layered generated audio could be really interesting. I have a feeling it might be one of those things, that if you experienced it done right, you couldn’t go back.

[@noot2981]

Excellent video once again! So exciting to see these kinds of breakthroughs in virtual worlds. As a request, I would love to see a comparison of these state of the art techniques and the level at which industry is operating right now. How big are the differences between what gamers for example, use every day and what is under development now? What needs to be done to get to that point? Etc.

[@joeyhayden8055]

Can’t wait to see this implemented in games!

[@MonsterCreations662]

we have to wait just i think some months or maximum 1 year only

[@AlexAnteroLammikko]

@@MonsterCreations662 Definitely longer than that. A lot longer.

[@jackbequick]

@@MonsterCreations662 it took 42 ms to render a tiny CD disk, we ain't seeing this in games for at least a couple years

[@Druark]

@@jackbequick For anyone wondering, the 42ms this took to render is equivalent to 23FPS in a game. Hardly playable except in the slowest kinds of games. As he said we still have a while to go with optimisation and better hardware first.

[@Druark]

@spooderderg4077 True they wouldn't render it to the same fidelity in a game but RTX is not magic that just fills in the blanks perfectly. You'd end up with lots of artifacts just doing it that way.

[@tomtomkowski7653]

And just like this your new shiny rtx4090 is a 1080p GPU xD

[@bluesillybeard]

I doubt ordinary rasterization will ever go away, especially in titles where graphical fidelity is not important, and it's just the fastest way to draw things in general. But, the sheer flexibility and power that ray tracing has will certainly take computer graphics to a whole new level.

[@himan12345678]

Especially with the possibilities of AI rendering. Likely the only thing that will be needed will be a matcap reference with a "neural render engine" and the result will be realism for rasterization "prices". Not to mention the concept of neural rendering can be much more stylistically diverse than ray tracing.

[@goldbomb4268]

Every time this channel pops into my feed, turns out raytracing somehow got a 1 million magnitude speed improvement. Hopefully this means Cyberpunk will finally be able to run at more than 60fps with raytracing enabled on my fucking 3080ti

[@MsJeffreyF]

This video really opened my eyes to how far away we are from having realistic looking renders. The new algorithm's results look nothing like the incredible real-life photos shown earlier in the video

[@HISEROD]

Definitely a big difference, however, this isn't really a fault of the renderer, but a lack of texturing, lighting, etc. experience on the part of the researchers imo (they are primarily scientists and programmers after all). A skilled 3D artist could definitely match or even outperform the aesthetics of the real photos without sacrificing physical accuracy if given access to the same render engine.

[@arekb5951]

in other words, with these screenshots they were trying to showcase the tech, not the current possibilities of photorealistic graphics rendering in general.

[@stickguy9109]

​@@HISERODYeah you could fake those with some clever shader magic and nobody would know the difference

[@arekb5951]

@@stickguy9109 you can say the same about ray/path tracing, you can use a lot of gimmicks to fool eyes of players, but 1. there are situations where normal rasterization just falls apart compared to ray/path tracing, 2. using gimmicks requires a lot of expertise and development time. Path tracing allows you to just set up materials and light sources and it takes care of the rest, producing very accurate lighting.

[@MsJeffreyF]

@@stickguy9109 I'd be curious to see some fakery that looks as good as the photos, got any suggestions on what to search for?

[@elderado]

bro became asphyxiated looking at ray tracing run with 20 more fps

[@eternaldarkness3139]

Any chance Nvidia can simulate a decent video card at a reasonable price?

[@atomictraveller]

wanna borrow my tandy coco?

[@Darhan62]

To do a rainbow boa, you have to have a 3D model that accurately simulates the shape of its body, the way it moves, the size, shape, and arrangement of scales and the way they flex as the snake bends its coils, etc... They've made *some* progress on iridescence, but if you've seen a real live rainbow boa right after a shed, or even good photos and videos, you know its iridescent effect is far more beautiful than anything we can simulate yet. I guess we'll have to wait and see where we are two more papers down the line.

[@4crafters597]

So, how long until this ends up in Blender / UE5 etc?

[@XeclipseXZ]

This is huge. What I mean I am working as an RF/Microwave Engineer. Most of the time I work it is just a comupter solwing wave physics. If the wave physics simulations gets more and more mainstream on GPUs it would cut our simulation times down aswell!

[@Cineenvenordquist]

Close enough for THz work sounds pretty natty.

[@bababonzio]

Does anybody know from when the render engines started to simulate light as a wave and not as a "ray"? (I mean, I don't even know if there are some yet)

[@7kortos7]

nope, this is new tech as far as I'm aware.

[@bababonzio]

@@7kortos7 lesssgooooo

[@Cineenvenordquist]

Still not a chiral wave.

[@chosen_oNEO]

Indeed "what a time to be alive" amazing freaking job

[@notCAMD]

So Nvidia's 50 series will be WTX instead of RTX?

[@notram249]

Probably

[@napoleonbonaparte525]

Ok, it can be render in real time but on what pc?

[@Loli_Awakening]

Your Nacho Libre impression is really good.

[@goodtothinkwith]

Now that the method is there, it is amenable to further optimization with AI. That makes it even more exciting that it was “hand crafted” to this point :)

[@thethiny]

AI doesn't optimize, it just trades less meaningful data for speed.

[@thethiny]

@Johnithinuioian skynet won't happen until gpt5 comes out with the ability to render stuff in real time using mobile phone cpus and cooperate with other agents. Currently, gpt3 takes 45 seconds per letter on an android, and gpt4 takes minutes.

[@himan12345678]

​@@thethiny??? ML is absolutely an optimization function. AI definitely optimizes. As someone who actually researches and writes AI models, I can only conclude you have no idea what you're talking about by how fundamental your misunderstandings are. And gpt3 (especially 4) cannot be run on a phone, let alone any level of pc workstation even. They are so large they have to be ran in a data center. Gpt3 is in the 100s of billions parameter size. That's 100s of GBs to just load the model. A mobile phone is not doing that. And information on gpt4 was recently leaked showing it is in the trillions size meaning it takes terabytes to even load. I suppose I could be charitable and say you misspoke and that you meant much much smaller, open source models can be ran on a mobile phone. Which have been shown to be possible, is highly misleading for many reasons. One trying to equate those models when they're barely comparable. And two, that is little more than a proof of concept tech demo because of how intensive it is to get that to work, there's no practical reason to other than to do it for that sake of doing it. And no, skynet will never happen. It's an apocalyptic fantasy. Thought experiments like paperclip maximizers and rokos basilisk, also are not things to seriously consider. Rokos basilisk least of all. That's not to say the underlying message of the paperclip maximizer isn't valid, it very much is. Tho it is most applicable to reinforcement learning and its weaknesses, ultimately we cannot know any entities true intrinsic motivations. And so we should spend time on looking into that problem and discovering what insights we can draw about such motivations without being able to directly observe them.

[@thethiny]

@@himan12345678 don't come at me with the AI BS by pretending to know AI just cuz you wrote a huge post. I know what I'm talking about and my reply was about the person's message, not AI in every scenario in the universe.

[@Mossie5965]

@@thethiny I think your internet is just slow

[@GivenKittens]

would the wave sim allow them to do the double slit experiment effectively?

[@OxyShmoxy]

Let me guess, this not gonna be available on old RTX GPUs?

[@Bloodlinedev]

Using Unity to show new graphics? Am I in the right timeline? _o.o_

[@tepidtuna7450]

Excellent news, next is Polarisation Modulation of simulated light and maybe that will lead to 3D without glasses. However I think display monitors will need to be reimagined too.

[@NotBirds]

I find it fascinating that using a wave function versus particles is faster... I think we'll start learning alot about our own universe pretty soon if this keeps up.

[@ElectRocnicOfficial]

Christian was my supervisor for my Bacc thesis :3 Happy to see his name here :3

[@Ashnurazg]

That small pun on your colleague's name. "Freude" means "joy" in German ... :)

[@picklypt]

Amazing video as always! Your channel is my single source to stay up to date with these topics, please don’t stop 😅

[@GAISENSE]

This is great, I'm missing the "how did they do it" part. What made their hand craft method work? What is this method?

[@alekdanser]

I found this video very hard to watch considering there's a weird artificial comma or pause every few words instead of having the whole sentence naturally flow. Either way good content.

[@no1Liikeglenn]

nice to see that it isn't just us swedes that string words together like that. this is our longest word with separations. " nord-väster-sjö-kust-artilleri-flyg-spanings-simulator-anläggnings-materiel-underhålls-uppföljnings-system-diskussions-inläggs-förberedelse-arbeten " translation: "north-west-sea-coastal-artillery-flight-reconnaissance-simulator-plant-equipment-maintenance-follow-up-system-discussion-post-preparation-works " want to work as one of those? cheers

[@atomictraveller]

how much weed can i smoke?

[@AlexTuduran]

But Karoly, that effect is called iridescence and it was tackled years ago by the folks at Unity - it's even available in the engine, runs in real-time, it's PBR and looks just like the ground truth. Also, regarding the glint, you presented us a paper from 2020 that implemented just that and it was extremely stable and realistic. What are the actual benefits of wave-based rendering? Does it do caustics at a fraction of time?

[@CharveL88]

As far as I can tell, wave-based rendering more accurately simulates the physics of light propagation in reality, whereas the technique Unity uses shortcuts which, I suspect, only work well for specific circumstancs and likely can't handle many natural edge-case scenarios. I don't know this for sure, but this new technique seems to be more general purpose so that it doesn't need to be massaged or compensated for in the scene design beforehand.

[@EMLtheViewer]

I am by no means an expert, but my understanding is that in addition to simulating iridescence on the surfaces of materials, this renderer is also capable of simulating other sorts of interference patterns and other optical effects that one would naturally get with wave light but not inherently with a typical ray-based renderer. Forgive me if I misunderstand, I know not much more than a layperson regarding this topic. This was just my takeaway from the video.

[@Cineenvenordquist]

Meanwhile in China: 'Hey you want a carbon fiber walker with an American fake iridescence on it for your Nan?'

[@marhensa]

can't wait to they capitalize it: "this brand new breaktrough will only works with RTX 5000 series". like what they did with DLSS3 that is locked only for RTX 4000 series.

[@ChimpanzeeUltra]

Is the video voice generated by AI?

[@cyber_robot889]

Can't believe this is happening! And finally will be in my 3d render workflow, without faking it in my workflow!

[@42ndMoose]

hi! im just a mere viewer of 2mp, i dont do any development myself, im just curious as to when this 'perfected light sim'? will reach into consumers' hands?

[@AntonioNoack]

@@42ndMoose We already have a perfect light sim, and it's called real life. On the other hand, simulations aren't really products, normal consumers are interested in. On a third point, "perfect" can be a really strong word, and for that it will probably need another 50 years (or way more/impossible). What could perfect be? All quantum effects, bending around mass, transfer objects from imagination into simulation, 500 fps on a toaster / smartphone, infinity-brightness HDR and contrast ratio, rendering with ~25 or so wavelengths instead of just three (makes the result look prettier, more realistic), ....

[@McRay2001]

This...... is..... a....... very......... informative......... video......! Thanks you....... for ........ up.......loading!........ ;-)

[@tirushone6446]

I just to say thankyou to all you light transport researchers out there. Although you don't get credit for it, it's you guys that enable the beautiful video games and real-time rendering of tomorrow. Bravo!

[@michaeltesterman8295]

Important question: What hardware was this tested on, and has it been tested on older hardware? If what I'm hoping is true, this could mean decently performant path-tracing simulations on at least GTX 1080s, and somewhat similar for graphics cards made around that time. If that's not the case, it's still a MASSIVE leap forward in rendering technology. Also good job on the new comparison paper! You guys crushed it.

[@jemandetwas1]

This was tested on an Rtx 3090 and it took very long to render. Not gonna happen on a 1080 I'm afraid

[@kyo666oyk]

And. Now. *Whispers*

[@_beel]

Is wave tracing the new ray tracing?

[@MonsterCreations662]

no, a light ray is a wave

[@carlosrivadulla8903]

@@MonsterCreations662 it's both, it depends of the viewer

[@Nightbanger89]

Awesome video! This is so interesting! Side note: light is actually both a particle and a wave at the same time, not just a wave 😃

[@JathraDH]

Side side note. Light is a wave until its looked at, then it becomes a particle. It is not both at the same time. It is one then instantly the other once observed.

[@Nightbanger89]

@@JathraDH You are absolutely right. What I should have written instead is "light _behaves_ like a particle and a wave"

[@valtarg1299]

Nice can't wait for the rtx 5090 with 60 fps at 1080p with path tracing

[@7kortos7]

the difference of 1080p and 1440p are almost negligible. just run at 1440p with the 5fps loss. i get you're making a joke but people often complain about low fps while not optimizing their graphics. all games are different, some games i spend an hour or more configuring the right settings.

[@GamesPlayer1337]

​@@7kortos7negligible... 78% more pixels. I wouldnt call a more than 50% increase in pixels to render 60+ times every second negligible lol

[@Knifewolf]

Wow that's really impressive. Good thing I didn't get the RTX. I can now get the next gpu with this technology

[@666Azmodan666]

this is not based on new hardware, it will work on older RTX.

[@dahahaka]

"Freude who was not Happy...." I see what you did there XD

[@buffaloSoldier519]

Kinda hard to see when all the examples were far away and heavily shadowed… Also what’s the cache size?

[@mohammedosman4902]

Congratulations on getting your work published. I downloaded and can't wait to read over the weekend

[@existentialselkath1264]

I don't see this being used in realtime. Path tracing is already 99.9999% there, so unless waves become just as efficient, I don't see it ever being worth the performance difference. Raytracing is already claimed to make a minimal difference, which is obsurdly incorrect, but this would be a case where I'd agree. UE5 is already working on a material system that can calculate thin film refraction and everything. I think emulating refraction with physically based materials would be better than doing it properly with brute force

[@test-rj2vl]

In video you showed just 1 car but how fast would it run if entire GTA 5 had some snow mod where all the cars were covered like that?

[@gdgd5194]

Haven't seen a single game with realistic effect using raytracing. It just like maxing out reflections in old games where everything becomes a mirror.

[@franklynotyourbussiness9401]

This is massive. This is as big as IBL, microfacet models, raytracing or metropolis light transport. It's gonna change the way we look at rendering techniques forever. I just watched the author's siggraph presentation as well as the paper and I cannot believe we're seeing rendering history absolute milestone being made

[@kenkioqqo]

So many good interesting things happening in so many technological fields, and so little time to delve deep into each one of them and build stuff. AI is changing life so fast, Unreal Engine and Blender are taking CGI to new heights with every new release, Web3 technologies are creating new types of blockchain-based businesses we never imagined before, Quantum Computers are promising to change how we compute the most complex mathematical problems humanity has ever faced, new programming frameworks popping up before you can learn to use the one you're currently learning, and so on. I would like to learn as much as possible about each field because I think they are fascinating, and yet, time is always limited.

[@tiavor]

wait a moment, what happened at 3:59 on the upper left corner of the car?

[@vasudevmenon2496]

I remember running luxmark and blender cycles render taking a long time to bake the effects and i had use to cache to speedup the render, now it's near real time. Nice find doctor

[@AlexanderFarley]

Ah yes, good old hand-rolled rendering layers like he had back in 2023

[@Sk-oh7rv]

This channel is incredible. Could I maybe suggest a less overacted voiceover? Thank you for your work!

[@Skyflairl2p]

Available only on the new 6000 series cards thanks to our new "wave optics accelerators"! -Nvidia marketing team (I bet).

[@joa1401]

2:56 If technology continues to progress at this rate, some day we might finally be able to understand what Hungarian people are saying! What a time to be alive. If computers can figure out what on earth “Örül, mint majom a farkának” is supposed to mean, they can do practically anything

[@colorado841]

Somebody should give this guy so O2.

[@msbealo]

I've missed this channel for 5 months because I wasn't getting recommended the videos, despite being subscribed. It looks like there's a lot to catch up on.

[@sonicalstudios]

Is there any demonstrations available that we can download

[@Cineenvenordquist]

What, besides the linked code and your deep desire to make ad agency specific ad jamming browser add-ons/extensions/WebAssembly contention?

[@sonicalstudios]

​@@Cineenvenordquisti was asking about a downloadable real-time copy that we can test. Dont know what you are talking about otherwise

[@rafaelraimer]

But why so much affectation on the part of the narrator?

[@itskittyme]

he. is. an. ai.

[@Lann91]

I can already see the -95% fps for a "minimum" settings in games for this ray tracing tech.

[@Mikelica69]

Cyberpunk overdrive be like

[@danielmilyutin9914]

That's really cool. I'd love to know how they achieved this. And I'd love to have it in games, of course.

[@jb9282]

It can be used for Keyshot rendering ? If yes , how can be used this please ?

[@qasimplays2531]

When can we expect this in 3D softwares like Blender and Cinema4D ?

[@7kortos7]

really soon. Q4 this year or Q1 next year.

[@GamesPlayer1337]

​@@7kortos7source?

[@sbnewsom]

@@GamesPlayer1337 previous experience with Blender's adaptability to tech trends.

[@alexmcleod4330]

The title of the video is terribly incorrect. It's making people think that ray-tracing can now be 10,000 times faster thanks to NVidia, when the papers are only discussing _very specific shading effects_ that can now be rendered _slightly_ faster (or more accurately) than with earlier methods. The figure of 10,000x faster comes from comparison with a method that _cannot_ converge, for that matter, so it's a completely bullshit figure, and yet there it is in the title of the video. Highlighting the "42ms" figure from the paper is making viewers think that the final frame took only 42ms, too, when the final frames actually required thousands or millions of samples and took hours to render. I admire the enthusiasm, sure - but maybe get your facts straight?

[@GraveUypo]

real time in what kind of hardware?

[@jaysonrees738]

To me, the real holy grail is full simulation of materials. Right now, cutting a simulated object in half will reveal that it's just a hollow shell. Pretty cool that they figured out some new tricks though.

[@JaviArte]

2:50 "megszentségteleníthetetlenségeskedéseitekért" :O what a word :O Is there a longer word? :O

[@Cineenvenordquist]

Yeah I still don't have it unpacked, it sounds like there's only one 'unprofane' and maybe it's not what expatriates could recommend. Longer words on hold kthxbi.

[@TheCrewExpendable]

2:07 Yes now we just need to simulate Fluorescence and Phosphorescence!

[@7kortos7]

I don't see any reason the new 'wave tracing' couldn't handle it. if you can add an emissive glow to something transparent or opaque. as far as computing goes those two are pretty similar.

[@fbnxo]

Is that going to be available for 4090 or only 5090s

[@markmuller7962]

I've been wondering when the real light simulation would have come and Moor's law has to surprise me every time, I was expecting this stage many years in the future

[@sebastianjost]

Jumping by a factor of 10000 in a single leap is very, very rare. This sounds like a huge breakthrough. There will be more like this in our near future, but probably not too many. 5 orders of magnitude really is a lot.

[@Luppoooo]

@@sebastianjost Moore's law has nothing to do with this, this is not an increase in comupting power

[@Cineenvenordquist]

​@@LuppooooWell then one hopes you're not the one picking bonuses for software engineers with 10,000x years. 🌯

[@Ragefist]

It's such a shame that this arbitrary and unnatural emphasis and the pointless pauses in speech are so annoying. I can't listen to this for 30 seconds.

[@shikome8111]

Bro sounds like ai voice

[@PrakashNaikade]

Is there any related NeRF paper already?

[@mz7315]

This is also a more tangible way for physics students to learn about light as a ray vs light as a wave. Use one technique, the other technique, then both!

[@tiavor]

I grew up in a snowy region, I love this!

[@InnerHacking]

I swear the voice of this guy is exactly like my favorite “medicine“ dealer just around the corner... With a trenchcoat full of hidden pockets and a Dick Tracy hat

[@hundvd_7]

3:00 I think it's more along the lines of: "For [plural]your acts of continuously doing things that are unable to be made unsacrilegious"

[@cunicelu]

Ray Tracing still greatly impact the performance of GPUs. You need to get a top tier GPU to use it and even then it has a big hit on FPS in some games.

[@lupintheiii3055]

This channel is just part of Nvidia marketing division at this point...

[@megapeiron]

When is this improvement going to be available for the general public?

[@xYamakaze]

I can't wait until this hits the consumer market in a few years. Do you think they would label their new lineup "WTX" for "wave traced lighting"?

[@someonewithsomename]

I don't understand why would you need to simulate light as waves for rendering. That all was not looking good. And you can make all those diffraction rainbow effects with a simple shader.

[@xl000]

Down the line we will get some kind of metallic super hero in the MCU with new effects. I think Ultron looked really nice in Age of Ultron but this will be something else

[@ipadize]

So its wave tracing then right?