Sparse Voxel Octree demo.
HTML-код
- Опубликовано: 6 окт 2024
- This is a demo of an out-of-core sparse-voxel-octree realtime raytracer. This demo usually runs at more than 60 fps on my GeForce GTX 480. The maximum resolution of a single octree is 131,072 x 131,072 x 131,072.
![Sparse voxel octree modification and benchmarking [Voxel Devlog #3]](http://i.ytimg.com/vi/yjOLx4O634I/mqdefault.jpg)
![Sparse voxel octree modification and benchmarking [Voxel Devlog #3]](/img/tr.png)
![Implementing sparse voxel octrees and the ray caster [Voxel Devlog #1]](http://i.ytimg.com/vi/zbmYzugnEC0/mqdefault.jpg)
![KSI - Thick Of It (feat. Trippie Redd) [Official Music Video]](http://i.ytimg.com/vi/At8v_Yc044Y/mqdefault.jpg)
10 yeats after, this still looks very impressive, I wonder I it would perform using modern hardware and coding techiniques
I mean, 1024x768 is like the quarter of 1920x1080, and we probably need 2x-4x the fps to be able to render other stuff that doesn't rely on an octree. But yeah, I'm interested too. Todays raytracing hardware supports Sparse Voxel Sets (use BVH raytracing, instead of an octree), which could speed things up too.
Frozein, Xima, Douglas, John Lin are all youtubers working on impressive voxel engines
@@Nerthexx you know, computers have gotten a bit faster than 2-4x since 10 years ago
You have to love Bruce's excuse 2012: “I suppose our big problem for a while was that we had so much power, but no content. Perhaps you noticed that Euclideon Island had to repeat the same objects over and over because one artist can’t make everything different. That problem has been fixed now because we have been doing a lot more work with laser scanners. You can make a lot more content when you start importing real environments from the real world.”
10 years ago 😭
Voxels are the future of gaming and 4D simulations. Keep on the good work!
Here we are 10 years later into gaming and the only popular voxel game is minecraft. While octrees became standard usage in data load/unloading, voxels only remained feasable in the real of computational methods. SOR, boltzman etc etc
@@Roxas99Yami I think a problem is content creation. Vectors graphics are so much easier to animate with the tools we have now than pixel or voxel graphics where we're generally forced to hand-draw each frame of animation or accept degradation and artifacts to what we drew. Discretized data is so much easier typically for programmers to work with, but sometimes much harder for artists when they want to do things like facial morphs or extreme bone deformations (ex: lifting a shoulder all the way up on a character).
@@Roxas99Yami I would say that Minecraft will sooner or later not be the only popular one.
The first Competitors are coming in.
Vintage Story, a Released Indie one
and
Hytale, a unreleased one which got acquired by Riot
have you seen Teardown? We were right, voxels were the future
Gtx 480 was a beast of a gpu back then. Nice demo
I've been waiting for more from you for a long time. Hope you release at least a very short demo video soon. :D
F**k me diagonally. This is impressive.
Welcome in my recommended, I hope this blows up soon!
Basically how eucledions unlimited detail works
I think somewhat to extremely different depending on your perspective. Euclideon renders point clouds and they're limited to orthographic rendering to my knowledge. The real test to me of both is how they can handle dynamic content like dynamic lights and dynamic modifications to the environment (ex: destructible environments). Euclideon seems to be at a significant disadvantage there to voxels since the discretized nature of voxels makes them easier to dynamically update efficiently as well as re-rasterize from sources like meshes.
I think voxel rendering could benefit greatly from procedural generated elements for detail. Instead of repeating the same texture as is done currently, you would use procedural materials that vary and can be fully volumetric. This would remove the need to stream lots of data over the internet.
This benefits gameplay. Voxels make in-game detailed environment destruction possible, and better graphics isn't automatically a bad thing.
This is pretty impressive.
But if you think in procedural materials the size on disk problem is nearly solved from theras to kbytes , really nice demo and runs fast .
Very damn nice.
Woah that man could spit facts like hes an old wester spitoon pro god damn!
Really impressive!
i found a neat octree where each node was not more then 1 byte, and color was stored as a hash table; i could've had a map as large as that decompressed in less than 80 mb/s using RLE to inflate compressed data quickly, and then deflate it when done using it.
This is not a demo for the masses, it's aimed at other graphics programmers. Artist created assets would look prettier but is not needed yet since this is a technology demo, not a trailer for a game.
Definitively not a waste of time, I agree.
I personaly find voxels really interesting for large amount of things. It could be a future technology, it may be not, but since id software are interested about it, we can suppose that it's not completly useless :)
A GTX 480???? HOLY FUCK WHAT THR FUCK
nop, is the other way arround. triangles are deform and collide easly and fast. the thing with triangles is that a scene query has O(n) complexity, n being the numbers of triangles, and a SVO has O(log8(n)) complexity. global ilumination, caustics, etc, are perfect for SVO, cause they need to take into account the surrounding geometry. I'm working in a indie game, and my engine uses a hibrid SVO/Triangles scheme, raytracing SVO to get global effects(cant show anything right now, sorry ;) )
Great demo!
From the details released in the video, it's obvious that the engine is custom.
You could handle the hard drive bottleneck fairly easily, by having on-the-fly decompression using an algorithm that operates quickly, and works well for this kind of data. Given that it compresses down to 80 MB, it sounds like it has a lot of duplicated sections in the file.
That said, RAM usage would still be a problem.
Euclideon has given the public enough proof enough. Directly and indirectly.
They did an interview with a live demo, as direct proof.
They submitted their technology for review to an Australian technology company that ended up investing in it, so it's clearly not a hoax.
Not to mention that Microsoft is interested in using the tech in Bing maps.
And with the lack of public information I will not take your beliefs on the lighting system seriously, because it's all just speculation at this stage.
Amazing!
really cool!
Great Job!
This technology would only be applicable to artists for a while, allowing them to sculpt very highly detailed levels. The memory size of this data is going to be HUGE, and would be best offset by first making a low-poly for this super-high-poly (either by hand or by procedural generation), and then baking the detail from the high to the low.
Trying for destructible environments like this in games won't be feasible (for the avg machine) for a long while still. I'd jump on it when it is though.
The day has come
Atomontage (To my knowledge) is unlimited.
But what's the difference? both are using SVO.
If you ever get around to it, creating a large-scale demo of this would probably prove a lot people out there wrong in regards to this technology.
@NewSoundGames Any source code available on this?
nice work there :)
how about a new kind of render?? So.. if we will get all screen and subdivide him on 4. And trace from the centers of this 4 squares Ray that bouncing from the walls To The Light Source, and get the information about the color, And return that Color to the Square.. So if it was done faster than 1Second/60Frames - subdivide this 4 squares on 4, so we will get an 16 that squares which will calculate ray from them To Light.. May be we will get 60 FPS everytime, and get graphics that your card can produce, but without lagging
come one we know that voxels or atom engines are future but no computer can actuali run them moving in real time.... so this is just prove of dream
Check out Atomontage.
This is NICE.
Do you think there could be some way to reuse data for repeating patterns? I'm thinking something like blocks. Like, instead of repeating a pattern over and over, there would be spots at so-and-so position around the playfield which would point to a single block of data at so-and-so angle, like we put textures on polys. Then generate the relevant data on-the-fly when a particular instance of that block is messed with in some way. I'm also an idiot who knows jack shit about coding.
You can run this at real-time but without any animation. As soon as any animation computing is involved, the hardware would get overwhelmed with all the data it has to process. And also another problem obviously would be computer space for such an incredible amount of detail.
***** i think everquest did exactly that
(im not suire what they did though so dont take my word for it)
Yep I second Brooklynns comment. Very well spoken.
Darko Bakula m.ruclips.net/video/nr5JqYYye3w/видео.html
@@Nightmare1066 Voxel animation comes almost free of charge actually assuming a raytracer, since you can do it on the fly during traversal of the SVO/SVDAG. The problem with animating voxels is not the computational aspect but if you imagine voxels that are supposed to stay connected together without gaps in between, they will begin to show gaps if you animate them as soft bodies unless we start trying to compensate for that somehow. That's what makes them ill-suited for things like character animation -- not because it's expensive -- but because it quickly leads to artifacts.
Exact same problem with animating pixels on a per-pixel basis at render time. Vector graphics are much more well-suited for animations that require deformations than pixel/voxel graphics.
I'm currently in the planning stage of creating a 3d voxel engine with a form of ray-tracing. Couldn't you create individual instances for each repeating object, but have a special model for each instance that stores each voxel as either there or not there in a single bit? It would be far more efficient.
Can you interact with/change the environment in real time? Even though it's being read from disk, could you do it to the in memory parts, even though they may be reset when you look back?
One word... HOW
what about those people like me with not so SUPER ULTIMATE FAST computers?
Euclideon claim to not even use the graphics card for pixel calculations! Saying they use a search engine algorithm on one core of the CPU. Or maybe it's all a lie, I don't know.
While voxels dont have a future for main surface rendering, global ilumination, diffuse reflections, and all the other soft global techniques are perfect for voxels.
7 years later, do you still think voxels have no future for main surface rendering?
I am a member of a team working on a game project that could benefit from making use of your system. Any chance of being able to negotiate an agreement on the use of your technology?
480 MY GOD
And its just gone. just gone.
As if the basic game developer in this day and age does "NOT SPECIALIZE IN 3D GRAPHICS".
the year is 2021, actually 2D games are still developed, sometimes
1 gigabyte decompressed? clearly he learned more about the GPU and spatial hierarchies then he did formatting and memory scaling;
Is it dynamic though ? Can you have physics for example ?
Bit late, but there's papers out there on modifying voxel octrees in realtime without having to decompress and recompresse the whole thing
is there any way i can get the source code to this or at least some references to work off of, I have attempted my own SVO but it runs at a measly 7fps at 1920x1080 with a tree resolution of 65536x65536x65536
are you trying to render mandelbulb?
release a demo? please?
There is something similar near 2 years old stuff. Maybe it is intersting for you as a developer:
Search in youtube for "Atomontage Engine - The Future Is Volumetric"
Triangles are still the best way to represent hard surfaces, taking into account quality/memory/rastertime ratio-
So many modern youtubers making voxel octrees and saying that its a brand new concept and they're on the first to do it, meanwhile this guy did it 10 years ago
What kind of card do you have? GTX 480 is 3 generations behind the current generation VCs. If yours is that old, perhaps an upgrade is in order?
Download?
What engine is this?
Can I path Trace it?
How do you handle non-axis normals?
Assign a normal to each voxel.
Why bother? this technology would be best with procedural generated data rather then storing it.
Would you be releasing the code for this demo, so I could learn from it?
1 gigabyte
It looks like bacon
Euclideon hasn't shown anything relevant, Atomontage is doing everything better.
This technology isn't/wasn't new.
To be honest it smells like snake oil, so just make some REAL reference, e.g. write an Unlimited Detail Renderer for Minecraft...
Cool demo, but hearing you slurp drool is disgusting.
Well Euclideons technology is far more impressive :p
I see cube's not voxels. A voxel is a 3D pixel. A voxel is similar to a sprite.
Pixels are squares, voxels are cubes
I think maybe you're conflating the data with rendering methods. A pixel can even be rendered as a circle, triangle, diamond... it's typically rasterized as a square but it doesn't have to be, and if you ever played an emulator of Super Mario, some of them offer rendering options to render the pixels in that game as something other than a square. Same with voxels. They might be rendered as billboard squares, or circles, or 3D cubes, or spheres, or tetrahedrons, or anything else. Also with the 3rd dimension, they can be anisotropic or isotropic.
you guys are NUTS
I've got cheaper voxel engine go out side ;)
if I want a game I want game play not graphics
dang where is world going to, ...