NVIDIA Just Supercharged Ray Tracing! 

All PC gamers since 2001 bought nvidia parts and hyped it.. i got the Nvidia 2060 super from nvidia now in my pc, 1st gen ray tracing, and it is mindblowing graphics and reflections on glass and water.. these games already exist (mods for cyberpunk or even the famous 3dmark tests for PC benchmarks for years)... it is clearly a game changer for making realistic graphics, real world graphics.

Use the fan of gpu

@@dertythegrower You can run RT on a weak GPU like a 2060 Super?

@@dertythegrower usually video games have done ray tracing until recently, all the ray tracing mods were trickery to fake real-time gi and reflections because proper ray tracing with old techniques would take a long time to compute, which is why 3D renders take minutes to some time hours for one frame.

Uon. Didnt expected to see a comment by you. :D

Not a time to be alive

Noise one!

If we shear the wool, we can hold the static we saw in our televisions back when we were kids

Needs a visit from the laundromat guy.

Lol, this reference is golden, the animation is one of the best out there made in blender.

Absolutly! Making games isnt about reproducing real life things like physics or lightrays path to put lighting at the right place(good exemple for this vid :P) , Its about taking shortcut to give the illution something is simulated

Because hardware power has been steadily increasing throughout the years, we've been able to continually upgrade the graphics and capabilities of games just by waiting for the newest chips to come out. Unfortunately, it's not a sustainable system, and we're hitting plateaus in terms of price to computing power ratios. In the future, if games want to advance the field in any way or be considered 'next gen', they're going to have to put much more focus on algorithmic efficiency and clever programming. Which IMO is a good thing because as it stands now most games are very poorly optimized, and rarely make an impact in the field in terms of smart new algorithms.

Nobody will continue research, If hardware isn't also getting better over time. New ideas come possible with new hardware.

The shear amount of calculations that have to be made in rapid succession to emulate light's various characteristics is very demanding. Yes we have come up with clever techniques and shortcuts to speed thing up significantly, but robust hardware capable of handle these computations is still very much essential, just look at Unreal Engine 5. Hardware's importance cannot be overlooked. Hardware, software, algorithms, and techniques are all important parts to the puzzle. The fact is that many of the mathematical functions used in emulating light have been around for well over 100 years, yet it's only through modern hardware that we can utilize them to visualize light in modern graphics.

@@Hackanhacker That's not what they said 💀

What could you render with 1984 hardware? I can't even imagine what working on ray tracing back then could look like. I wasn't even alive back then. I remember working with 3d software and just rendering a single quality image from a scene in 3d too a good PC a couple of days (I don't think it had ray tracing, but I had no idea).

raytracing exists for that long? its older than me lmao

@@shadowlord0162 It's a very simple and intuitive method, not a surprise we came up with it in the early days.

@@shadowlord0162 Ray tracing has been around since the 60s, in fact it predates modern rasterisation techniques we see today. In essence ray-tracing 'really' simple, you can put together a renderer (albeit very basic) in a few hundred lines of code, is just naturally computationally expensive. Accelerating it though? Not so simple.

That _is_ true. Though it's also possible that the technique to denoise the data wouldn't take as much time, considering that it's working with far greater information. It could cancel it out, but with the image being more accurate. Either way, considering this is open-source, I'd imagine it won't be long before the frametime shrinks in a paper or two, it might not even become an issue.

It depends on how it scales. If the fidelity is 25x at 60% more time at 1spp, maybe you can go down to .5spp with the new method and still get better quality than 1spp in the old method. Then you're running both faster and at a higher quality.

​@link99912 0.5spp? What would half a sample look like in path tracing? I don't think that makes sense, you either take a sample for each pixel or you don't... Unless you take one sample for every two or more pixels in the image, but that's more like using DLSS for upsampling.

@@israelRaizer it is what dlss 3.5 ray reconstruction already using. one sample for every two or more pixels in the image.

@@wanhl2440 yeah, I get it, I just don't think that terminology makes much sense because in the "tracing step" of the pipeline (where the amount of samples is relevant) it's still a minimum of 1spp, you can't have a non-integer number of samples per pixel

😳

When she tracing your rays but you're gangster 🙀

maybe I’ll be tracer

Raaaage tracer!

Write that down, write that down

This!

I get the AI fatigue, but who do you think handcrafted the Ai learning algorithms? Not trying to throw shade, just find the distinction between two different tools /workflows to use GPUs to create images interesting, when in this context they're both massively impressive complex algorithms some very creative and intelligent people have made

Letting machines do the work for us IS the future. Wasn't that a good thing a couple decades ago?

​@@SibanamushI think it's a testament to the speed of progress when we are having breakthroughs every other day to the point where they get boring lol

Blender's currently focusing on EEVEE next for realtime and trying to keep Cycles as unbiased as possible without any sort of spatial or multi-frame information. However, just looking at the images on their project page, it seems a lot less lossy than re-stir so it seems feasible that they might implement it in the future. Of course, they still have to finish eevee next, spectral cycles (maybe), the animation rework, etc... It is open source but under an NVIDIA license, so IDK if it would be possible to use directly in blender

​@@theneonbop EEVEE and Cycles have completely separate devs, there is no "focus on eevee". This also extends to geometry nodes, ui, animation, etc.

@@ettiSurreal Yeah, you're right. Just from a glance at the blender repo, it looks like there is a little bit of overlap in the personnel between cycles and eevee but I guess it is mostly specialized. Anyways, one of the full time devs made a wip PR for re-stir about a month ago so hopefully that will get into a usable state relatively soon.

Has I understand it, games only simulate some reflections and shadows. Not the entire scene

TLDR; it's like having a better engine in a fast car; you can only drive an engine so fast, and at some point you're better off investing in a better designed engine. We can do raytracing with RTX realtime in such games only because it's used in *hugely* simplified [geometrywise] scenes with minimal light sources and bounces, compared to a fully rendered photorealistic scene with whoknows how many light sources and continual differences in texture, reflection, lensing, and complex edge geometry. And while it looks fantastic in games [usually] compared to non RTX games, it's still objectively terrible compared to true photographs of a similar scene. You _could_ crank existing tech up to some super high level of iteration (and this undoubtedly will happen), but without algorithmic optimisation, you're looking at rapidly diminishing returns.... and a huge electricity bill. Additionally; outside the realm of games and more just into non realtime CGI/VFX; imagine it taking you say 5 minutes to render a full 4K still image at present for a test render for a still for a movie, compared to, say half that time on the same hardware if some new algorithm like the above is introduced... that's a lot of cost savings in your work flow. More power/GPUs is useful. but novel algorithmic optimisation is better :)

Throwing more and more rays is _not_ what you want to do --- especially not for real-time applications. What you want is to develop assumptions and approximations that allow you to render a _perceptibly plausible_ image while generating only a fraction of the number of rays that would be required for an accurate render.

more rays = higher render time. for games tho, we dont really actually need realtime raytracing, since you can just bake all the lights in (except moving lights, but you can do them in realtime and noone will really notice), and use reflection probes for the reflections. yeah, raytraced reflections are way better, but probes and screenspace reflections still work fine. as graphics got better and better, i noticed the decline in creativity, graphics wise. take gta iv: the shadows and lighting are absolutely terrible, but my god is the design of the actual map beautiful. it still feels so realistic and lived in. then take the Matrix Demo for Unreal 5: yeah, looks kinda good, but given the lack of detail, it seems way less realistic. nowadays, people think, just because their game has accurate reflections and somewhat accurate lighting, the graphics are "good", when in reality the game looks like hot garbage with accurate lighting lol

@@carlosmarx2380 For me, GTA IV's game design was quite horrid. When you died during a mission that had several parts, you might had to start from complete opposite of the map and it took you like thirty minutes to get were you died, with the several mission stages. Also, the game felt buggy and I didn't love the controls (especially the shooting), and the map was the most boring in the series. But maybe that's just me. I think a better example is something like Ico that had a beautiful (game) design and style which we don't see much anymore in AAA games. Thankfully indie games are doing the heavy lifting when it comes to (graphical) innovation and (game) design that feels something different and new.

Older version of Blender pre 2.8 series @ 0:32 and then later on screenshot of Blender 2.8 @ 0:44

Yes, his voice is terrible to listen to, I always mute the video, I used subtitles 😅 (borzalmas hallgatni ,borzalmas hangsúly )

Tbh I can't bear his videos so much great content and I just can't listen his way of speaking, make louder on words in the middle of sentences

Looks like it uses a temporal technique which is likely incompatible with per frame offline rendering. Also notice there’s no volumetrics.

Yes!

​@@thornnorton5953 😢

​@@thornnorton5953 It is temporal, but I don't think there's any reason it couldn't theoretically be implemented into Blender, it would just probably take a lot of reworking and effort. If it was added, I would expect it to probably act as a 4th option in between the quality of EEVEE and Cycles. And yes, volumetrics don't seem to be showcased in their demos, but I would be shocked if it didn't support them - it seems like their technique relies on re-using ray paths but it looks like it keeps the logic of a singular ray hit identical. The one thing it doesn't appear to support is motion blur, but the authors of the paper say they have ideas to address this in the future. It looks like the blender devs are finally finishing eevee-next, so hopefully some other interesting rendering stuff will come soon. It looks like one of blender's full time developers is currently starting work on the original re-stir (with temporal elements) in a PR opened about a month ago (#121023)

@@thornnorton5953 Temporal is optional, it's still a big improvement even without it.

This guy makes me feel like I'm listening at 5fps

In a game with raytracing, the game is calculated normally, like any other game, with - usually - many light sources. And then some raytracing effects are added to shiny surfaces that are defined in the game as being able to receive ray tracing. It's very rudementary. In ray traced 3D software, like 3DS Max, you have all your object, but all light sources are not working, like in a game. They are ray traced calculated. It's a bit hard to explain, but in 3D, everything is built out of triangles. You have a light source, but it doesn't emit light. It "shines" on a triangle of a model and according to the camera position (your view), it simply calculate how much color should be on that triangle. Say the triangle is 100% red, but the triangle is in what's supposed to be a dark area, according to the light source and your view, a dark red is simply added. That's a very basic explanation. With ray tracing, it emulates real world, where a number of rays from the light source are hitting that red triangle and calculate how the triangle of the object should be lit. Not only that, but every ray is calculated how it bounces back and hit other triangles of other object. The calculations are even done on basis of how shiny the triangles in the scene are supposed to be. It is a lot more computational intensive, that how a game works.

And here's an example on how ray tracing works in 3DS Max. Note, that the ray tracing is sped up and not in real time, so every time he makes an adjustment, it's not in realtime. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-48YtLmJCPzE.htmlsi=Awt9f14bJhHo2boL

@@akyhne There are a few titles that offer full real time path tracing options though like Cyberpunk and Alan Wake 2. You really need the best hardware combined with upscaling to get playable frame rates and bounces are limited but it's still incredible that it's possible at all in these graphically intensive games.

@@mojojojo6292 You can't compare ray tracing in games with ray tracing in 3D software.

@@akyhne I agree. I was just pointing out that there are some games that use full scene path tracing for all light sources and not just some effects. The quality of the RT obviously does not measure up to 3d rendering software with only 1 to 2 bounces and a very limited number of rays but it is real time in very large scenes with reasonable frame rates.

yeah, that little "denoised" box in the video isn't really enough to see it lol. I tried the demo on their github - I had to turn it to half resolution or else it needs more than the 12gb of vram on my 3060 (it runs at about 25 fps at half resolution, maybe fine for a cinematic game if you combined it with dlss/xess), but it doesn't seem to have an option for denoising.

yeah it's amazing . with the help of AI we can change whole world

No? Currently the most popular GPU (do not look only at the steam hardware survey!) for home computers is the GTX 1650 (my current daily driver), soon to be dethroned by the 3060. If we can't do that high end consumer GPUs or server-grade GPUs now, no shot we'll be able to do that in 5 years. Also Nvidia is really greedy and will probably make their GPU prices explode again by then. Not to mention that many other computers don't even have a GPU at all. If people enable themselves to think Nvidia isn't the only company making GPUs, I could see this to become a bit possible on the X090 Ti, but of course this won't happen, like people will get Apple products even if they have clearly worst stats than Androids. I like the optimism though! If only everything wasn't ruined by capitalism...

@@theskyblockman I think the bad times of NVidia and GPUs in general are over. I think the crypto bubble bursted. RTX series is not that prohibitively expensive anymore and it delivers considerably more than the last generation. It's not TRUE (full blown) ray tracing yet, but we're getting there. One new feature I test is DLSS. It really works and it allows to play in 4K when otherwise only HD would work.

​@@01001000010101000100 When I was talking about prices going up I didn't really thought of crypto: Nvidia got their crypto infinite money glitch patched, so now they are interested in AI and have a new reason to explode the GPU prices again. Imagine a 5090 with NPU-like capabilities, this could totally make the already ludicrous price for the GPU go x2 (and people would buy it for sure), slowing the reach for the tech. Also I am pretty sure that because of AI Nvidia slowed down Ray Tracing because they would make more money on it than on RT. (You can skip this, this is just a big tangent on why the GPU market is bad rn) Personally when Nvidia started non-algorithmic processing on games it was the time to go to AMD for me (today I am on Linux so my next GPU is an Intel or an AMD). AMD here is totally the better option if people want better GPUs on the long run but nobody thinks like that with their money (Intel's first gen GPU was maybe bad but they arrived in a new market and actually did pretty well compared to Nvidia and AMD/AMI first gen GPUs, only they entered the market sucking but not being as near as the other GPUs, result: nobody took Intel GPUs and Battlemage is not sure to release at all) if people bought Intel GPUs, I can assure you we would have 32 gigs of VRAM with an acceptable bus size for a fraction of the current price of GPUs but nobody want to support the underdog with their money.

Portal RTX, Cyberpunk, etc all do path traced Re-STIR in real time on a home PC, and you only need a $300 GPU for it... We're not 'close to it' we already have it. Of course, it can be improved - I need to use heavy DLSS upscaling on my 3060 for Portal RTX, and I would expect cyberpunk to be even more demanding, but real time path tracing is very much currently possible.

@@theskyblockman "of course this won't happen, like people will get Apple products even if they have clearly worst stats than Androids" Have you even looked at benchmarks in the past 5 years? Apple's mobile CPUs have been at the top of the leaderboards for a long time. And I've yet to see an android phone with the 3D scanning features iPhones have, with both the structured light sensor and the lidar sensor. And yes, a high end GPU can currently do real time path tracing (with re-stir). Just look at some of the games NVIDIA have been showing off like Portal RTX and Cyberpunk. This video kind of gives the impression that they can't, because it only shows off frames before denoising, which is never what the user will actually see. But they can - Re-STIR just looses some details that can be recovered with something like this or NVIDIA's secretive 'DLSS 3.5'.

I don't know much, but I hope this'll help. Feel free to correct me if I'm wrong. This is pure path traced rendering. Which doesn't involve rasterization. Also Path Tracing is harder than Ray tracing as Ray tracing only does one bounce off a surface while Path Tracing might do multiple bounces for better realism. In games we use a combination of raster and ray traced rendering. As raster rendering is much easier, games will do most in raster and use ray tracing for specific areas. Even then, the ray traced areas will look noisy, but the noise filtering is easier as the raster layer below hides most of the artifacts. Please correct if I'm wrong, I'd love to learn more.

Path tracing is basically: shoot a ray from camera, if it hits anything bounce in a random direction (might not be that random, depends on material), continue bouncing until you're bored. Repeat the whole process 10-10000 times to clear up the noise. Ray tracing: shoot a ray from camera, if it hits anything shoot n rays towards each nearby light, If you don't hit anything on the path to light then this light affects the material. Shot a reflection ray and you're done. First step can be skipped if you use rasterization.

I appreciate the replies, thanks!

A lot of VR headsets are portable now and have weak gpus, so maybe not this decade haha

Already can. Just not worth the development cost, but I'm sure some indie studio with unreal will do it.

@@smoothbraindetainer Not path tracing. Lumen works, but path tracing at full res is still too slow.

@@theneonbop yes path tracing, just maybe not with first gen rt cores.

@@smoothbraindetainer What makes you think 2x4k 90fps path tracing is possible with current hardware?

10 years

"could be 100% raytracing" That would be a bad idea - there are way too many things that are way more effectively down with normal rendering. In general you want to use raytracing to get the extra data for the normal renderer - like getting the indirect light (not the direct light), volumetric materials etc.

how lmao

@@capitannerevar7792 no more glasses, just real time super high quality rendering

Control has very limited RT effects. Most of the rendering is still traditional methods with a basic layer of RT on top for reflections and shadows. There are more recent games that have full scene path tracing options though like Cyberpunk and Alan Wake 2 that use every trick in the book to give playable frame rates with a good image, 1-2 bounces, small number of rays, upscaling, frame generation and denoising. The result is still incredible. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-qvu4aru39SA.html