@Slome no, so that we can use more assets on the scene than the current tech allows us. The reason why you wont see hundreds of zombies, hundreds of cars, hundreds of spaceships or cities with amazing draw distance is simply because of this fact. Polygons. Any game, except 2D games, will benefit from it.
@@milenkopizdic9217 it is more a comedic cultural response the actual offense. (We really love puns but can do so publicly so we have to pretend to hate them.)
@@MrFrussel yes, this and foveated rendering together, which is already happening - make realism in VR go from just out of reach to accessible to indie devs, all within the next 18 months give or take. (mostly waiting on fov rendering on more headsets) and then if you set your sights just a little bit lower, standalone headsets could still maybe use nanite to squeeze out an acceptable vr gaming performance on a mobile gpu.
@@betterlifeexe Is there a foveated rendering solution like VRSS 2 that doesn't require DX11, forward rendering and multisampling? Because that's gonna hold this back. DX11 is on the brink of disappearing and rightfully so. Can't have Nanite (mesh shader) without DX12 Ultimate or Vulkan.
@@boredgunner You are absolutely right, at least I cannot find a solution with a cursory search. I'm sure someone will solve this problem, I have never worked on rendering tools so I can't do anything about it.
@Milen (splicer) Parvanov Maybe you should first go and learn about Nanite technology used in UE5. It's not about instantiating. And even so, try to instantiate milion of polygons at the same time of the same model even using the same material being GPU instantiated. Conclusion: you have no idea what you are talking about.
This is such a HUGE step into having very realistic games without needing a NASA computer. I'm amazed by this technology and what developers like you can potentially build 🤩
Asterisk: as long as they involve simpler assets that can be replicated, so it can benefit from them all being instances of the same thing. I think that's my main take away from this. It's cool as hell *but* you're gonna see a lot of replicated objects (which might not even be a problem, just something worth being aware of)
Photo realism is more a function of light calculations and less about the models themselves. And in models, the textures are way more important than the 3d-model. No matter how good your 3d-models are, they will always suck as much as the lightning sucks. Just by using correct lightning techniques you can make even Minecraft look photorealistic. The strengths of Nanite is in performance and development work flow and the realism comes as a secondary side effect.
@@RedTyrant Please, read my comment again. Take your time and read it as many times as you need until you finally understand what I'm talking about. Be sure to use external resources too just to study this subject as a whole. When ever you try to dispute something, you should at least try to understand the subject. Never try that on things you don't know.
I cannot begin to understand how this is possible, even with the marvel that is modern graphics hardware. Like, I see what it's doing but still, it just blows my mind that it all works so flawlessly and efficiently. Crazy times.
Nanite can render 20 mi triangles per screen if I remember correctly, so you can put trillions of polygons into one scene, that will not make a difference. The cluster system will reduce that to 20mi. The 4K image has about 8 to 9 millions pixels, so nanite can display 2x times triangles about the pixel count of the screen, so you will never going to see major difference compared to the scene rendered with the full polycount anyway.
I remember the whole euclideon fiasco from many years ago, they did a similar thing with point clouds by basically only drawing data which was relevant and required to populate each on screen pixel with geometry down to sub mm precision, however the idea of using this in point clouds is ridiculous due to the unfathomably huge file sizes, impossibility of physics calculations and complete lack of shader support. However on their demonstrations, they managed to display impressively large (although ridiculously repetitive and simple) environments from high resolution point cloud data. This is a much more intelligent approach. Essentially it's just intelligent self building variable LOD
this combined with Nvidia's resolution upscaling will certainly prepare the future of AAA games at fantastic resolutions running on affordable hardware (a circumstance created by the low supply of chips and an influx of millions of new and young gamers). I hope to be one of the people working on stuff like this in the near future.
@@joshpage4547 Definitely, engine improvements like this and others will allow developers to squeeze the most out of any given level of technology and you can almost guarantee that this framework will lay down the groundwork for lots of exciting things to come.
@@michaellvoltare I did watch the video. And I've been using Unreal engine for 20 years since the late 90's. What you're failing to understand is that this example does not use a million high poly meshes. It uses one single high-poly mesh which it then copies (as what's called an instance) 999,999 times. Using only one object plus its instances is not an effective test of the nanite tech; because one instance repeated does not stress the i/o throughput of the CPU or graphics hardware. In short: one instance repeated does not use as much memory as multiple different objects would. In fact - instances are designed specifically to enable the same mesh to be viewed multiple times without a big hit to memory usage. If this had been a demo with thousands of distinctly different objects my comments would have been very different. If you don't believe me, watch DF talk about the Unreal engine 5 Matrix tech-demo that was just released showcasing Lumen and Nanite in a AAA game format. That demo runs at 24fps. They're pretty clear that they don't believe the current console gen (PS5, X series) will be able to push AAA titles at higher than 30fps. Some less-complex games (or games on high-end PC) may exceed 30fps with Nanite and Lumen running but it's unlikely that the current console gen will get there for AAA titles. th-cam.com/video/ib6_c6uliLg/w-d-xo.html
Would be interesting to see Euclidean demo style particles of sand/dirt used to make up a path, and see how that differs in appearance from something with displacement and/or normal maps.
The speed at which it can cull clusters is incredible, but it's probably a side effect of modern day hardware. This wouldn't be possible with slower video memory. Incredible engineering.
This is absolutely mind-blowing. And it's just the beginning of this tech! Imagine what optimizations are coming and the crazy ways people will put this to use. Amazing!
Who ever conceptualized nanite, had the first white paper, and proof on concept, is a game changer. Thank you to that person. Then thank you to the dev team at epic for making it so easy to use.
Nanite is awesome. Shame it doesn't work on mobiles or older hardware. I feel like nanite would really benefit older or slower devices so I'm curious to know why it only works on more modern hardware only.
Because it requires the use of compute shaders. Most mobile targets OpenGL ES2.0 still which doesn't support them. They literally built a compute shader software rasterizer to do this.
To my understanding part of it also has to do with the nanite system requiring crazy huge memory bandwidth with low latency as it streams and decompresses a crap-ton of nanite compressed mesh data from storage to memory constantly, only holding in memory that which is necessary for rendering what's visible on the screen.
You know, when I worked at Epic in late 2003 to early 2005, working on UT2k4, James Golding once told me he predicted that, one day, rendering engines would probably be fast enough to use polygons (triangles) the size of pixels, and that map data might just be stored vertex info, per vertex. I of course didn't believe him, having been a 3d modeler and texture artist for years prior, but he was right on the first part of his prediction. Jesus...seeing this engine, it almost makes me want to get back into game dev.
@@Alphashow7 I already saw this video, we are truly blessed to live in these great times! I bet that 50 years from now, we probably going to laugh that we once considered these graphics realistic!
I remember watching a video about this kind of rendering technology back in the days of the first Crysis. Where some fella was talking about rendering dots instead of triangles to some classical music playing in the background. The implications were staggering to think about but now.... Like 14-15 years later, it's finally a reality and oh my gosh. Girls and boys, the future is looking bright. That much is certain.
@etethstrseh yeye exactly. I just looked it up. Looking at it now I'm not sure what I'm listening to just like last time. The tech described is very promising and makes sense but the guy behind it just sounds like a college student hyping up their "not so special" homework assignment.
I've only ever used Unity for game development, and I really like it. But seeing stuff like this makes my mouth water. Gonna have to give this a shot (as soon as I get a PC that can run UE5, lol)
they kinda did, but in a really crude way. depending on distance from the player/camera, models would instead display simpler hand-crafted versions of themselves to reduce workload on the hardware. but if it's not gradual enough you can notice the pop-in, from low detail to high detail and vice versa. nanite just does it on the fly without the need to manually produce lower detail models by hand & setting the transition distances. nanite is going to save a LOT of developer time, while looking a million times better and producing incredible performance gains. I will be shocked if other engine developers aren't furiously working on their own LOD solutions now.
@@TheObeyWeegee I thought it was really noticeable in half life alyx. Would be really cool if Nanite was open source for devs, even if it required a license/paid access. Now imagine Nanite and Dlss 3.5+ in vr in the future! That would be incredible!
Fantastic video! What a great demo, it's crazy how far this technology has come and the minds that made it possible. I wonder how this will optimize the mobile or VR experience. Or even how far Nanite can be optimized to see what FPS counts you can reach while achieving similar LOD. Gaming was due for some modern optimizations with intensive games are now.
All of this makes me wonder: Will other engines have the same feature or will everyone just start using UE5? Because even if UE5 isn't perfectly optimized for everything, just having nanite sounds like it should be an obvious choice.
Other companies will try to do something similar but it will take them a while and most companies don't have the budget for research and development that Epic does.
This engine is just such a mind boggling feat! Imagine the team showing this tech to their Epic execs for the first time and for the next two days, there are no questions because everyone is still completely speechless. Epic's own tech demo was already crazy but this video here shows how stupid fast this actually is. Essentially, this looks like the infinite detail demo we saw from this disappeared company a few years ago only this time, fully integrated with real time global illumination and physics. Absolute craziness. I wonder what this would look like with growing numbers of DIFFERENT models, not the same instance of the same thing. This could be done by creating the same ball under different files name and importing those.
Yeah and there are people saying nanite and lumen changes nothing and isn't revolutionary. This is basically a functional version of Unlimited Detail tech dudes but people can get their hands on and is less limited.
There's nothing particularly impressive about rendering instances of the same model as the data is already in memory but I understand how it looks. The truth is = if this were ten thousand different objects each with this geometric density - that would be impressive.
@@iconoclasttastic9258 It's impressive because almost no other graphics engine has done capitalized on this stuff. Also its doing more then copy pasting objects in memory, its doing DYNAMIC LOD... that is pretty awesome.
@@PRiMETECHAU Real time LOD has been around since the early 2000's. Again - this is impressive tech don't get me wrong, but when you understand the technology and code behind it, this isn't a particularly impressive demo.
@@iconoclasttastic9258 I don't think you understand the technology here. This requires only 1 high detail poly from the artist, not 3 or more to suit a LOD system that switches out models depending on distance.
damn i just saw you only having 20 subs. this channel is going places, great quality and good voice too. keep it up and people will come, ue5 is the big thing that everybody is looking for rn. Btw i found you through youtube search and filtered by "uploaded during last week". cheers!
@@LivelyGeekGames This video is indeed awesome :) Love these UE5 sneak peaks. I found it because youtube suggested it to me :D (btw, love your sense of humor)
Nanite kinda feels like the biggest revolution in graphics since PBR to me. My hope is that this technology will also make its way into offline rendering to an extent. I cannot even imagine how much faster movie frames would be rendered with this, since it's almost exclusively about quality and not render times there
Nanite is very impressive, and a super smart way to seamlessly optimise meshes for real-time. No doubt this will save a lot of inexperienced modellers from a world of lag
I've been getting into game development with Unity recently, but hot damn if Nanite doesn't make me jealous of Unreal 5. This is some seriously next-gen optimization technology.
@@snaileri That's an interesting limitation, that would mean you couldn't use this for trees and grass animated with the wind like in Valheim. Can you easily have a Nanite environment (dealing with stone, dirt, steel and glass) and then fill that environment with lower-detail "live environment"-props for greenery?
You can still use non-nanite meshes for characters, foliage, and other meshes that don't work with nanite, it doesn't need to be all or nothing. This is what they did in the UE5 demo project.
The idea is very similar to adaptive subdivision surface that blender and other engines implemented a while ago, but seeing it into a game engine is really cool.
Although you are correct, I think we need to understand how hard taking the latency out must have been, as blender does not need to provide this feature at 60fps
Yes. It is impresive but only few people talks about it's limitations. Nanite can be used ONLY for static objects. No moving parts, No changing geometry etc.
@@kvmairforce Well it depends of the game type. In games like "FarCry" almost everyting is moving: grass, trees, animals, bushes, water ... everything is constantly moving. But I can agree that in games like TombRaider in caves most things is stationary.
Excellent quality, informative, and straight to the point, i was suprised when i saw you only have 201 subs, i hope your channel succeeds and in the mean time, i'll be the 202nd subscriber hahaha
you can procedurally generate meshes instead of making them by hand. that way you can have as many LOD models as you want by adjusting a few params. if you want the final LOD can be handmade to show detail
also LOD models will always be faster then using nanite since nanite requires extra processing in real time to check the amount of detail each object needs instead of just using a draw distance LOD
I remember 2 year ago when starting my new York and importing it on unity, i was lagging the hell out no matters how hard i Try to optimize the 300 000 objects and most of them where not even detailled.... I just told to myself : darn it i wish there's was something to import such huge map without having problems and having 60 fps... And now it happened
@@Grumpy_old_Boot Nope my PC seems to hate UE5 and my gpu overheat when it start to open the engine i have weird issue with UE5 and it make me sad but one day i will try that again
@@Grumpy_old_Boot well it's beefy but old i guess that's the problem, i have 1070 I7 6700k the best of 2016 still hold up but i don't want to make it overheat too much and UE5 seems to hate my computer for some reasons even if i saw some peoples with the same spec running unreal engine
@@michelveraliot Hey, I have the 6700K CPU too ! 😁 I have it paired up with 32 GB of ram and a GTX 1660 super and a 1TB SSD I think the new UE5 engine likes graphics cards with RTX the best. So both of us are too old. 🤣
I think it will probably have issues if you place a fully textured model with many texture maps. It will probably reach a limit, I'm talking a model with transpadency, caustics, displacement. It would be logical for it to have a lower FPS.
So we have 8k Textures and more polygons we ever need for photo-realism in an game, now we need good performance for physics when it must show masses of it, than we need full destruction like in an voxel engine with nanite enabled and at last, we need real physics for water,fire,wind,gravity,gas and so on..... some physics today are just visual tricks
I'm new to UE and super late but maaan. I thought nanite was just a gimmick for high-end hardware, but that it's that easy to use... This is absolutely bonkers!!! O_O
Simple fact, nanite can render 20 mi triangles per screen if I remember correctly, so you can put trillions of polygons that will not make a difference. The cluster system will reduce that to 20mi. The 4K image has about 8 to 9 millions pixels, so nine can display 2x times triangles about the pixel count of the screen, so you never going to see major difference anyway.
Yeah, ideally it can cull down all possible geometry to have at most 1 triangle per pixel. It does an amazing job at quickly culling out unneeded clusters and triangles to render each frame. I even tested up to almost 1 trillion triangles in a world with another test, and it still performed quite well.
Ever since nanite was announced I’ve been flipping out about it. The capabilities this shows us are possible is insane. I’ve heard it may increase memory, which is a slight downside, but for real, even a trashy computer can work pretty well with nanites. I can’t wait to see how it develops further in the future 😁😁😁
Fear not, nanite uses a technology known as Mesh shaders or Task shaders as Nvidia calls them and were introduced with Directx 12 (this would be 2015 give or take) so it has existed for quite some time and I would imagine more than a few titles make use of them. Only reason there was no hype around it back then was because of the introduction of raytracing. But yes I agree this is really cool.
@@Gooseguy1000-1 It is revolutionary, mesh shaders take general purpose programming to rendering effectively making indexed triangle lists obsolete. I believe the microsoft has a talk were they go over the capabilities of this system titled 'reinventing the geometry pipeline'. As for the base it is compute shaders. Cheers
@@jonatanlind5408 I came across Mesh Shaders when sifting through the Vulkan API documentation. I read some of it and it seemed pretty powerful but I'm not ready to try using them.
Would you not have bigger file sizes due to heavy model detail ? It will always be king to have optimised mesh? This looks like a handy way to apply a basic re-mesher to a 3d object and use as a background object.
Yeah, you definitely have larger mesh assets from all the detail. The Nanite page covers this under "Data Size": docs.unrealengine.com/5.0/en-US/RenderingFeatures/Nanite/ But since you don't need to make custom LODs and if you end up not needing a high-detail normal map, the total size (mesh + maps) can be comparable.
@@LivelyGeekGames You didnt make as clear as you should have. Yeah the file sizes will not increase at all. Plus I heard them working on optimizing file sizes even more.
funny to think that on a very basic level, all nanite does is check if a triangle is actually larger than a pixel (ie visible) before spending time to render it. so simple in theory yet its taken ~25 years of 3d rendering technology for someone to think of it?
It's not really the size of the triangle but what to sample for any specific pixel. Think of it this way: there may be 600 triangles fighting who gets to decide the color for the pixel (or sample) so you still will sample from one of them and it should be deterministic enough that it won't alias too much. That decision is made by the rasterization rule, top-bottom, center, whatever that may. Even better if more than 1 triangle contributes to the color to get some anti-aliasing in there. It comes down to this: even if your mesh is made of 600 trillion very tiny triangles, you shouldn't have pixels missing the mesh in other words having pixels missing: Overlap is not acceptable either: if these 600 trillion triangles are drawn with additive blending, each pixel should only have exactly one time overdraw. For simplification the center-of-fragment sampling is easy to visualize in imagination.. the center of the pixel hits only one triangle, no matter how small they might be.. and that is The Chosen One. But then, we have finite precision.. if we have 10 bits of subpixel precision we only have 1024x1024 unique positions to sample, so we're screwed. There is a solution: maybe use a few triangles less.. nanite has LODing so use higher LOD, there will not be downside, it'll just be faster.
Where did you come up with that conclusion? Its not that simple at all. You forgot about determining which triangles are in front to be rendered. Thats a major issue that drags the performance down. I watched the Nanite presentation. If you use Quixel Megascans and instancing to model a scene you will have tons of overlapping geometry especially at glancing angles the engine struggles to determine which triangle is on top and this hurts performance a lot. They said its best to avoid it as much as you can. Maybe only a few layers of overlapping geometry but I suspect studios will be wanting every bit of performance so they wont use that workflow. Its easy but comes at cost. Unless they implement some sort of optimizer that will determine where do meshes intersect and cut the part of the mesh thats covered but I dont think thats coming down the pipeline.
@@outlander234 Of course it is not that simple but your comment is based on misunderstanding what overlap in scan conversion rule signifies. It means when fragment is considered to be inside the triangle. Two triangles sharing a common edge are not allowed to write into one fragment more than once. A completely different thing what you are raising a flag on.
you know, i think there is a way that infinite geometry can be drawn on screen but take no disk space, and that is when the geo is a 3D fractal generated by an algorithm. Would be great if a game could utilize such a fractal as a game world.
Something that’s crazier is if you read the siggraph presentation this isn’t even hardware rendered. That’s right. What you are seeing is software rendered. Pretty crazy math here. It’s a different way of doing a rendering pipeline.
I'm not a developer and if I ever made a game it would be 2d but I watched this all the way through with full attention. This is despite the fact that today I'm actually really exhausted and have had a hard time paying attention to anything. You make good content.
Infinity polygons. Every console generations (i.e. ps1, ps2 etc...) I look at the tyres of a car and notice how with each console they become less noticeable with ps1 being the most prominent and the ps4 the least but they're still there. I want to do the same with the ps5 games.
![Nanite: Everything You Should Know [Unreal Engine 5]](http://i.ytimg.com/vi/P65cADzsP8Q/mqdefault.jpg)
Be sure to check out my other Nanite video here: th-cam.com/video/v9kynURWW_I/w-d-xo.html
Wowowowowowowowowowowwow wow
So amazing 🤩🤩🤩🤩🤩🤩🤩
I love it so much 💖
What hardware is this on?
My respect to engineers and programmers who constantly optimize processes.
@Slome lmao
@Slome no, so that we can use more assets on the scene than the current tech allows us. The reason why you wont see hundreds of zombies, hundreds of cars, hundreds of spaceships or cities with amazing draw distance is simply because of this fact. Polygons. Any game, except 2D games, will benefit from it.
@plasticelephant1969 Its less about putting more on screen than it is allowing asset designers to more quickly create assets for their game.
You threw 1 million snowballs at an engine and it didn't freeze, impressive
get out!
Lol that's a good one
This is quite literally game changing
.......
@@cjmixmaster ?
@@milenkopizdic9217 we don't like puns where I'm from. They require punishment and ostracization.
@@cjmixmaster Oh, ok. I'm sorry he offended you that much.
@@milenkopizdic9217 it is more a comedic cultural response the actual offense. (We really love puns but can do so publicly so we have to pretend to hate them.)
next gen vr with nanite support is going to be insane!
Haven't even thought about that. But you're right, the increase in detail possible will be a gamechanger!
@@MrFrussel yes, this and foveated rendering together, which is already happening - make realism in VR go from just out of reach to accessible to indie devs, all within the next 18 months give or take. (mostly waiting on fov rendering on more headsets)
and then if you set your sights just a little bit lower, standalone headsets could still maybe use nanite to squeeze out an acceptable vr gaming performance on a mobile gpu.
Yeah VR needs Nanite BADLY.
@@betterlifeexe Is there a foveated rendering solution like VRSS 2 that doesn't require DX11, forward rendering and multisampling? Because that's gonna hold this back. DX11 is on the brink of disappearing and rightfully so. Can't have Nanite (mesh shader) without DX12 Ultimate or Vulkan.
@@boredgunner You are absolutely right, at least I cannot find a solution with a cursory search. I'm sure someone will solve this problem, I have never worked on rendering tools so I can't do anything about it.
Artist: How many triangles we can have on the screen?
UE5: Yes.
infinite triangles per pixel
Artist: How many triangles we can have on the screen?
Blender :
Blender.exe is not responding
Close this program
Wait for the application to respond
@Milen (splicer) Parvanov Maybe you should first go and learn about Nanite technology used in UE5. It's not about instantiating. And even so, try to instantiate milion of polygons at the same time of the same model even using the same material being GPU instantiated.
Conclusion: you have no idea what you are talking about.
@Milen (splicer) Parvanov have u even tried ue5
@@echofooman2702
Inventor user: "What the hell are triangles?"
This is such a HUGE step into having very realistic games without needing a NASA computer. I'm amazed by this technology and what developers like you can potentially build 🤩
Asterisk: as long as they involve simpler assets that can be replicated, so it can benefit from them all being instances of the same thing. I think that's my main take away from this. It's cool as hell *but* you're gonna see a lot of replicated objects (which might not even be a problem, just something worth being aware of)
Photo realism is more a function of light calculations and less about the models themselves. And in models, the textures are way more important than the 3d-model. No matter how good your 3d-models are, they will always suck as much as the lightning sucks. Just by using correct lightning techniques you can make even Minecraft look photorealistic.
The strengths of Nanite is in performance and development work flow and the realism comes as a secondary side effect.
@@RedTyrant Please, read my comment again. Take your time and read it as many times as you need until you finally understand what I'm talking about. Be sure to use external resources too just to study this subject as a whole.
When ever you try to dispute something, you should at least try to understand the subject. Never try that on things you don't know.
@@anteshell mate he was having a dig at you for typing lightning instead of lighting in your comment.
Photorealism doesn't make a game good tho.
I cannot begin to understand how this is possible, even with the marvel that is modern graphics hardware. Like, I see what it's doing but still, it just blows my mind that it all works so flawlessly and efficiently. Crazy times.
There's an hour-long video by one of the principal engineers about this. Very full of jargon but also quite thorough
@@schulace Do you have a link to that video?
Nanite can render 20 mi triangles per screen if I remember correctly, so you can put trillions of polygons into one scene, that will not make a difference. The cluster system will reduce that to 20mi. The 4K image has about 8 to 9 millions pixels, so nanite can display 2x times triangles about the pixel count of the screen, so you will never going to see major difference compared to the scene rendered with the full polycount anyway.
@Nagato is better than Punk Naruto guys, I know it's Nanites. But just dropping that name doesn't explain anything, does it?
@@BB_O_99, "major difference" compared to what? What are you talking about?
Never thought I'd be so satisfied seeing all these balls
That's what she said
@@lucasimonelli5038 dang it i was gonna say that
@@JizzEditzzz 😅🤝🏻
I remember the whole euclideon fiasco from many years ago, they did a similar thing with point clouds by basically only drawing data which was relevant and required to populate each on screen pixel with geometry down to sub mm precision, however the idea of using this in point clouds is ridiculous due to the unfathomably huge file sizes, impossibility of physics calculations and complete lack of shader support. However on their demonstrations, they managed to display impressively large (although ridiculously repetitive and simple) environments from high resolution point cloud data. This is a much more intelligent approach. Essentially it's just intelligent self building variable LOD
this combined with Nvidia's resolution upscaling will certainly prepare the future of AAA games at fantastic resolutions running on affordable hardware (a circumstance created by the low supply of chips and an influx of millions of new and young gamers). I hope to be one of the people working on stuff like this in the near future.
@@joshpage4547 Definitely, engine improvements like this and others will allow developers to squeeze the most out of any given level of technology and you can almost guarantee that this framework will lay down the groundwork for lots of exciting things to come.
Nanite is an incredible feature. Even if it has bugs at launch, this is clearly the way 3D engines are headed.
I hope not, I'd rather stick with smooth 60fps and lower detail personally.
@@iconoclasttastic9258 why though?
@@iconoclasttastic9258 did you watch the video? A million high poly meshes and he said he did not dip below 60.
@@michaellvoltare I did watch the video. And I've been using Unreal engine for 20 years since the late 90's.
What you're failing to understand is that this example does not use a million high poly meshes. It uses one single high-poly mesh which it then copies (as what's called an instance) 999,999 times. Using only one object plus its instances is not an effective test of the nanite tech; because one instance repeated does not stress the i/o throughput of the CPU or graphics hardware. In short: one instance repeated does not use as much memory as multiple different objects would. In fact - instances are designed specifically to enable the same mesh to be viewed multiple times without a big hit to memory usage.
If this had been a demo with thousands of distinctly different objects my comments would have been very different.
If you don't believe me, watch DF talk about the Unreal engine 5 Matrix tech-demo that was just released showcasing Lumen and Nanite in a AAA game format. That demo runs at 24fps. They're pretty clear that they don't believe the current console gen (PS5, X series) will be able to push AAA titles at higher than 30fps.
Some less-complex games (or games on high-end PC) may exceed 30fps with Nanite and Lumen running but it's unlikely that the current console gen will get there for AAA titles.
th-cam.com/video/ib6_c6uliLg/w-d-xo.html
@@mediumplayer1 Because to me, higher frame rates make for better gaming experiences than dense geometric detail.
I was actually pretty surprised you only had 50 subscribers with content like this. May the algorithm bless you
Thanks for the kind words!
Algoritm did bless him, thats how I got here!
🙌🏼
666 subscribers as of time of watching... Does that count as a blessing? ^_~
Hello world from the algorithm
Would be interesting to see Euclidean demo style particles of sand/dirt used to make up a path, and see how that differs in appearance from something with displacement and/or normal maps.
Yes yes yes yes. I've been waiting years for that tech. For they just disapear into something about VR Museums tech
The speed at which it can cull clusters is incredible, but it's probably a side effect of modern day hardware. This wouldn't be possible with slower video memory. Incredible engineering.
Can't wait now we will get actual optimization in UE games
You won't get any optimization for current day devs (PR, suits and sales people more precise). They will say - get better/more powerful machine.
@@mindaugasstankus5943 depression
@@mindaugasstankus5943 That's the thing, you can't
lol no
@@mindaugasstankus5943 "you aint using it right"
Managed to get on my recommended page! Going places bro!
Awesome! Thank you!
This is absolutely mind-blowing. And it's just the beginning of this tech! Imagine what optimizations are coming and the crazy ways people will put this to use. Amazing!
3:20 "infact my performance increased"
My coffee came out of my nose
Who ever conceptualized nanite, had the first white paper, and proof on concept, is a game changer. Thank you to that person. Then thank you to the dev team at epic for making it so easy to use.
Nanite is awesome. Shame it doesn't work on mobiles or older hardware. I feel like nanite would really benefit older or slower devices so I'm curious to know why it only works on more modern hardware only.
BS sales pitch (Nanite) for some hardware geometry accelerator, co-processor, core, what ever that don't exist in older hardware.
Yes but also: fuck mobile games and get a game boy color
Because it requires the use of compute shaders. Most mobile targets OpenGL ES2.0 still which doesn't support them. They literally built a compute shader software rasterizer to do this.
@@mindaugasstankus5943 you're just wrong
To my understanding part of it also has to do with the nanite system requiring crazy huge memory bandwidth with low latency as it streams and decompresses a crap-ton of nanite compressed mesh data from storage to memory constantly, only holding in memory that which is necessary for rendering what's visible on the screen.
I’m glad I work with Unreal, it seems to make a new technological break through almost every time.
It implements so many amazing features.
You know, when I worked at Epic in late 2003 to early 2005, working on UT2k4, James Golding once told me he predicted that, one day, rendering engines would probably be fast enough to use polygons (triangles) the size of pixels, and that map data might just be stored vertex info, per vertex. I of course didn't believe him, having been a 3d modeler and texture artist for years prior, but he was right on the first part of his prediction. Jesus...seeing this engine, it almost makes me want to get back into game dev.
I could literally feel the frames come back in my soul after hitting apply ☺️
Jaw Dropping. The quality of games this opens up is just... Unreal. :P
This is crazy stuff, i feel like matrix is just around the corner. I expect that in few years VR experience will be totally transformed.
Neo that was the wrong pill. Im literally dying over here!!!!
And you were right : th-cam.com/video/WU0gvPcc3jQ/w-d-xo.html
@@Alphashow7 I already saw this video, we are truly blessed to live in these great times! I bet that 50 years from now, we probably going to laugh that we once considered these graphics realistic!
I remember watching a video about this kind of rendering technology back in the days of the first Crysis. Where some fella was talking about rendering dots instead of triangles to some classical music playing in the background. The implications were staggering to think about but now....
Like 14-15 years later, it's finally a reality and oh my gosh. Girls and boys, the future is looking bright. That much is certain.
@etethstrseh yeye exactly. I just looked it up.
Looking at it now I'm not sure what I'm listening to just like last time. The tech described is very promising and makes sense but the guy behind it just sounds like a college student hyping up their "not so special" homework assignment.
I've only ever used Unity for game development, and I really like it. But seeing stuff like this makes my mouth water. Gonna have to give this a shot (as soon as I get a PC that can run UE5, lol)
So basically Nanite does what I thought game engines had been doing for years
Same, I didn't thought people would be manually making models for each LOD
they kinda did, but in a really crude way. depending on distance from the player/camera, models would instead display simpler hand-crafted versions of themselves to reduce workload on the hardware. but if it's not gradual enough you can notice the pop-in, from low detail to high detail and vice versa. nanite just does it on the fly without the need to manually produce lower detail models by hand & setting the transition distances. nanite is going to save a LOT of developer time, while looking a million times better and producing incredible performance gains. I will be shocked if other engine developers aren't furiously working on their own LOD solutions now.
@@zedsdeadbaby The biggest example of that is most of Valve's games especially Half-Life 2 and Team Fortress 2.
@@TheObeyWeegee I thought it was really noticeable in half life alyx. Would be really cool if Nanite was open source for devs, even if it required a license/paid access. Now imagine Nanite and Dlss 3.5+ in vr in the future! That would be incredible!
Fantastic video! What a great demo, it's crazy how far this technology has come and the minds that made it possible. I wonder how this will optimize the mobile or VR experience. Or even how far Nanite can be optimized to see what FPS counts you can reach while achieving similar LOD. Gaming was due for some modern optimizations with intensive games are now.
All of this makes me wonder: Will other engines have the same feature or will everyone just start using UE5? Because even if UE5 isn't perfectly optimized for everything, just having nanite sounds like it should be an obvious choice.
Other companies will try to do something similar but it will take them a while and most companies don't have the budget for research and development that Epic does.
This is very impressive stuff. I considerably underestimated the impact of Nanite.
Great example of how powerful Nanite is!
The fact that there are One million objects in the scene itself is already amazing (without ECS).
This engine is just such a mind boggling feat! Imagine the team showing this tech to their Epic execs for the first time and for the next two days, there are no questions because everyone is still completely speechless. Epic's own tech demo was already crazy but this video here shows how stupid fast this actually is. Essentially, this looks like the infinite detail demo we saw from this disappeared company a few years ago only this time, fully integrated with real time global illumination and physics. Absolute craziness.
I wonder what this would look like with growing numbers of DIFFERENT models, not the same instance of the same thing. This could be done by creating the same ball under different files name and importing those.
I made another video using different meshes here: th-cam.com/video/v9kynURWW_I/w-d-xo.html
Holy Crud man. This is impressive.
Yeah and there are people saying nanite and lumen changes nothing and isn't revolutionary. This is basically a functional version of Unlimited Detail tech dudes but people can get their hands on and is less limited.
There's nothing particularly impressive about rendering instances of the same model as the data is already in memory but I understand how it looks. The truth is = if this were ten thousand different objects each with this geometric density - that would be impressive.
@@iconoclasttastic9258 It's impressive because almost no other graphics engine has done capitalized on this stuff. Also its doing more then copy pasting objects in memory, its doing DYNAMIC LOD... that is pretty awesome.
@@PRiMETECHAU Real time LOD has been around since the early 2000's. Again - this is impressive tech don't get me wrong, but when you understand the technology and code behind it, this isn't a particularly impressive demo.
@@iconoclasttastic9258 I don't think you understand the technology here. This requires only 1 high detail poly from the artist, not 3 or more to suit a LOD system that switches out models depending on distance.
More videos like this please! So sayeth the panda.
And... The UE guys just solved all the problems related on comments and showed It on Matrix Tech demo, amazing
So basically it automatically does LOD's? And at 2:37 it's showing the level of detail changing?
Incredible overview and test of an incredible new tech. Thank you for this.
Reminds me of how the 2014 LittleBigPlanet 3 did infinite levels with their “dynamic thermometer” that renders what’s close and hides what’s further.
This is absolutely insane...
It truly is a great time to get into UE.
This will make games look insane. I couldn't believe it when I first tried it
unbelievable and yet so simple in it's idea
I remember Gran Turismo 6 did something like that on PS3
damn i just saw you only having 20 subs. this channel is going places, great quality and good voice too. keep it up and people will come, ue5 is the big thing that everybody is looking for rn. Btw i found you through youtube search and filtered by "uploaded during last week". cheers!
Thanks, I appreciate the feedback! I'm hoping folks find the videos useful, but either way I'm having fun making them while learning Unreal.
@@LivelyGeekGames This video is indeed awesome :) Love these UE5 sneak peaks. I found it because youtube suggested it to me :D (btw, love your sense of humor)
Nanite kinda feels like the biggest revolution in graphics since PBR to me. My hope is that this technology will also make its way into offline rendering to an extent. I cannot even imagine how much faster movie frames would be rendered with this, since it's almost exclusively about quality and not render times there
Nanite is very impressive, and a super smart way to seamlessly optimise meshes for real-time. No doubt this will save a lot of inexperienced modellers from a world of lag
This is super impressive, I'm excited to see what people do with this.
Nanite is truly next-gen in gaming. I can't wait to see it utilized by developers.
What a time to be alive!!
Hold on to your papers
This + fsr will make games run so well
I've been getting into game development with Unity recently, but hot damn if Nanite doesn't make me jealous of Unreal 5. This is some seriously next-gen optimization technology.
Ask them to start using mesh shaders for LODs, that's all what it takes
Stop being a pleb and learn UE. Unity is not even on the same league anymore if you want to do 3D
This looks absolutely revolutionary.
Is it really true that tri limits are gone for good?
Definitely not. You can't build moving stretching meshes (i.e. character animations) with Nanite. Only static environments.
@@snaileri That's an interesting limitation, that would mean you couldn't use this for trees and grass animated with the wind like in Valheim.
Can you easily have a Nanite environment (dealing with stone, dirt, steel and glass) and then fill that environment with lower-detail "live environment"-props for greenery?
You can still use non-nanite meshes for characters, foliage, and other meshes that don't work with nanite, it doesn't need to be all or nothing. This is what they did in the UE5 demo project.
@@LivelyGeekGames couldnt you just throw a vertex shader on the triangles after nanite processed them(if it allows it) to make animated grass/trees?
@@id_NaN I don't think Nanite supports vertex shading yet, but I have not tried it.
The idea is very similar to adaptive subdivision surface that blender and other engines implemented a while ago, but seeing it into a game engine is really cool.
Although you are correct, I think we need to understand how hard taking the latency out must have been, as blender does not need to provide this feature at 60fps
@@gwentarinokripperinolkjdsf683 yeah nanite does this but far better
nanite reminds me of Euclideon's unlimited detail tech demo from 2011
Yes. It is impresive but only few people talks about it's limitations. Nanite can be used ONLY for static objects. No moving parts, No changing geometry etc.
Most things in game don't move, environment is a major part in games, but that leave more resources for things that do move.
@@kvmairforce Well it depends of the game type. In games like "FarCry" almost everyting is moving: grass, trees, animals, bushes, water ... everything is constantly moving. But I can agree that in games like TombRaider in caves most things is stationary.
Wow!!! 50,000,000 triangles all in one frame. Unfortunately I can't see them all. My crappy 1080p monitor only has about 2 million pixels.
Haha, yeah. Definitely a bit of overdraw happening there.
I can't wait till Epic cracks how to get Nanite functioning on all meshes including skeletal mesh actors.
Excellent quality, informative, and straight to the point, i was suprised when i saw you only have 201 subs, i hope your channel succeeds and in the mean time, i'll be the 202nd subscriber hahaha
Thanks for the kind words! And for the sub!
Nanite is incredible. I love that I'm alive to witness this leap in video game tech.
you can procedurally generate meshes instead of making them by hand. that way you can have as many LOD models as you want by adjusting a few params. if you want the final LOD can be handmade to show detail
also LOD models will always be faster then using nanite since nanite requires extra processing in real time to check the amount of detail each object needs instead of just using a draw distance LOD
really high quality video
imagine showing this to someone in 2019
THIS IS AWESOME MAN!😍😍❤❤❤❤
Honestly I’m amazed at the performance it was getting before Nanite. Afterwards, it’s just…. unreal
I'm a Unity guys, I have no experience with Unreal... Now I want to. God, it seems unbelievable, really.
Looking forward to your Xtreme Snowball Warfare game.
That's hella dope. Finally we can get next-gen graphics despite the hardware companies slacking for the past decade.
I dont know what rock you’ve been living under, but the advances in discrete computer parts have been astounding. Especially in the pc space
I remember 2 year ago when starting my new York and importing it on unity, i was lagging the hell out no matters how hard i Try to optimize the 300 000 objects and most of them where not even detailled....
I just told to myself : darn it i wish there's was something to import such huge map without having problems and having 60 fps...
And now it happened
Have you tried it ?
@@Grumpy_old_Boot Nope my PC seems to hate UE5 and my gpu overheat when it start to open the engine i have weird issue with UE5 and it make me sad but one day i will try that again
@@michelveraliot
Oh, that's sad.
I take it your PC is beefy enough to run it ?
@@Grumpy_old_Boot well it's beefy but old i guess that's the problem, i have 1070 I7 6700k the best of 2016 still hold up but i don't want to make it overheat too much and UE5 seems to hate my computer for some reasons even if i saw some peoples with the same spec running unreal engine
@@michelveraliot
Hey, I have the 6700K CPU too ! 😁
I have it paired up with 32 GB of ram and a GTX 1660 super and a 1TB SSD
I think the new UE5 engine likes graphics cards with RTX the best.
So both of us are too old. 🤣
Can you imagine big companies with machinery they have to build games with this ? it's going to be insane..
thats pure magic for me... to the genious guy who came up with that: thank you bro! I just wish that we could use that with raytracing...
I think it will probably have issues if you place a fully textured model with many texture maps. It will probably reach a limit, I'm talking a model with transpadency, caustics, displacement. It would be logical for it to have a lower FPS.
Keep up the good work man, great and fun video! :D
Thanks, will do!
So we have 8k Textures and more polygons we ever need for photo-realism in an game, now we need good performance for physics when it must show masses of it, than we need full destruction like in an voxel engine with nanite enabled and at last, we need real physics for water,fire,wind,gravity,gas and so on..... some physics today are just visual tricks
"1 million snowballs. It got better..."
I guess Epic just thought: "let's make it render faster with more tris. Why didn't we think of that before?"
This video is beautiful @LivelyGeekGames. Absolutely beautiful. Keep Doing more videos please. 10/10
I'm new to UE and super late but maaan. I thought nanite was just a gimmick for high-end hardware, but that it's that easy to use... This is absolutely bonkers!!! O_O
Unity profusely sweating in the background
Well AAA isn't unity's priority so i doubt it
@@ciixo8510 le 🧂
This is content I need in my life. Subscribed.
Thanks!
Thanks for teaching me how nanite works
Simple fact, nanite can render 20 mi triangles per screen if I remember correctly, so you can put trillions of polygons that will not make a difference. The cluster system will reduce that to 20mi. The 4K image has about 8 to 9 millions pixels, so nine can display 2x times triangles about the pixel count of the screen, so you never going to see major difference anyway.
Yeah, ideally it can cull down all possible geometry to have at most 1 triangle per pixel. It does an amazing job at quickly culling out unneeded clusters and triangles to render each frame. I even tested up to almost 1 trillion triangles in a world with another test, and it still performed quite well.
@@LivelyGeekGames Yes, also how is done to avoiding mesh distortions, different than old decimation systems.
I actually really like this channel dawg, I gave you that subscribe, unreal engine 5 is somethin anit it
Thanks, I appreciate it! Yeah, UE5 is pretty amazing.
Can't wait too. This will really go well with my next project.
I believe that this tech was what John Carmack had in mind for IDTech 6.
Next Gen games are about to become insane.
That’s amazing man! Thank you for this👍🏽👍🏽👍🏽
Interesting how the unwraping and texturing pipelines will changes.
Ever since nanite was announced I’ve been flipping out about it. The capabilities this shows us are possible is insane. I’ve heard it may increase memory, which is a slight downside, but for real, even a trashy computer can work pretty well with nanites. I can’t wait to see how it develops further in the future 😁😁😁
Fear not, nanite uses a technology known as Mesh shaders or Task shaders as Nvidia calls them and were introduced with Directx 12 (this would be 2015 give or take) so it has existed for quite some time and I would imagine more than a few titles make use of them. Only reason there was no hype around it back then was because of the introduction of raytracing. But yes I agree this is really cool.
I imagine they must have had a base for what to go off of, but this is a revolutionary improvement in my eyes 😁
@@Gooseguy1000-1 It is revolutionary, mesh shaders take general purpose programming to rendering effectively making indexed triangle lists obsolete. I believe the microsoft has a talk were they go over the capabilities of this system titled 'reinventing the geometry pipeline'. As for the base it is compute shaders. Cheers
@@jonatanlind5408 I came across Mesh Shaders when sifting through the Vulkan API documentation. I read some of it and it seemed pretty powerful but I'm not ready to try using them.
Best explanation so far.
Wonderful content! I'm glad to have stumbled on this.
Thanks!
Great video mate! Subbed.
Much appreciated!
The 1 million player battles will be epic!
Hah, now we just need game servers that could handle that many!
Would you not have bigger file sizes due to heavy model detail ? It will always be king to have optimised mesh? This looks like a handy way to apply a basic re-mesher to a 3d object and use as a background object.
Yeah, you definitely have larger mesh assets from all the detail. The Nanite page covers this under "Data Size": docs.unrealengine.com/5.0/en-US/RenderingFeatures/Nanite/
But since you don't need to make custom LODs and if you end up not needing a high-detail normal map, the total size (mesh + maps) can be comparable.
@@LivelyGeekGames You didnt make as clear as you should have. Yeah the file sizes will not increase at all. Plus I heard them working on optimizing file sizes even more.
Finally, I no longer need to collect my golf balls from the game
"It just works."
Except that it really does.
funny to think that on a very basic level, all nanite does is check if a triangle is actually larger than a pixel (ie visible) before spending time to render it. so simple in theory yet its taken ~25 years of 3d rendering technology for someone to think of it?
It's not really the size of the triangle but what to sample for any specific pixel. Think of it this way: there may be 600 triangles fighting who gets to decide the color for the pixel (or sample) so you still will sample from one of them and it should be deterministic enough that it won't alias too much. That decision is made by the rasterization rule, top-bottom, center, whatever that may. Even better if more than 1 triangle contributes to the color to get some anti-aliasing in there.
It comes down to this: even if your mesh is made of 600 trillion very tiny triangles, you shouldn't have pixels missing the mesh in other words having pixels missing: Overlap is not acceptable either: if these 600 trillion triangles are drawn with additive blending, each pixel should only have exactly one time overdraw.
For simplification the center-of-fragment sampling is easy to visualize in imagination.. the center of the pixel hits only one triangle, no matter how small they might be.. and that is The Chosen One. But then, we have finite precision.. if we have 10 bits of subpixel precision we only have 1024x1024 unique positions to sample, so we're screwed. There is a solution: maybe use a few triangles less.. nanite has LODing so use higher LOD, there will not be downside, it'll just be faster.
Where did you come up with that conclusion? Its not that simple at all. You forgot about determining which triangles are in front to be rendered. Thats a major issue that drags the performance down. I watched the Nanite presentation. If you use Quixel Megascans and instancing to model a scene you will have tons of overlapping geometry especially at glancing angles the engine struggles to determine which triangle is on top and this hurts performance a lot. They said its best to avoid it as much as you can. Maybe only a few layers of overlapping geometry but I suspect studios will be wanting every bit of performance so they wont use that workflow. Its easy but comes at cost. Unless they implement some sort of optimizer that will determine where do meshes intersect and cut the part of the mesh thats covered but I dont think thats coming down the pipeline.
@@outlander234 Of course it is not that simple but your comment is based on misunderstanding what overlap in scan conversion rule signifies. It means when fragment is considered to be inside the triangle. Two triangles sharing a common edge are not allowed to write into one fragment more than once. A completely different thing what you are raising a flag on.
you know, i think there is a way that infinite geometry can be drawn on screen but take no disk space, and that is when the geo is a 3D fractal generated by an algorithm. Would be great if a game could utilize such a fractal as a game world.
Something that’s crazier is if you read the siggraph presentation this isn’t even hardware rendered. That’s right. What you are seeing is software rendered. Pretty crazy math here. It’s a different way of doing a rendering pipeline.
Well, that’s it. We’re officially in the future 🤯😱❤️❤️
This will actually be a quantum leap Forward in battle sim games like total war.
I'm not a developer and if I ever made a game it would be 2d but I watched this all the way through with full attention. This is despite the fact that today I'm actually really exhausted and have had a hard time paying attention to anything. You make good content.
Thank you!
Infinity polygons. Every console generations (i.e. ps1, ps2 etc...) I look at the tyres of a car and notice how with each console they become less noticeable with ps1 being the most prominent and the ps4 the least but they're still there. I want to do the same with the ps5 games.