Unlimited 3d geometry, no polygons!1??

I saw a video about this technology a few years ago and it was incredible with just basic shapes. This actually shows detail now and it's even better than I could imagine.

*edit* Starting at 2:46 in, with the pyramids, is that video I saw a couple years back. Another interesting thing I remember now about this is that even though it's unlimited detail because of using points, it only draws an amount of pixels which is necessary to be viewed on the screen (the amount your resolution H*W is.). Absolutely fascinating technology.
 
Last edited:
Cool -- wonder if it will be fast enough. The video seems like they are running at a really low framerate, but the higher level concept seems awesome. I still am unsure of how creating content on this system would work -- not really textures and wireframe like the tech of today.
 
Welcome to 1995 and Voxels, this is nothing new. This is some tiny little company playing with voxels and the video is basically speaking to people who know nothing about computers at all, probably to drum up investment or something. I'll wait until we can see what kind of hardware this implementation needs.
 
Last edited:
Spare-Flair said:
Welcome to 1995 and Voxels, this is nothing new. This is some tiny little company playing with voxels and the video is basically speaking to people who know nothing about computers at all, probably to drum up investment or something. I'll wait until we can see what kind of hardware this implementation needs.
Click to expand...
Yeah ... and if this is legit they are going to do something with it by turning it into a modern graphic engine rather than letting the technology rot. Oh by the way, if you watch the rest of their videos they say "we're working with Voxel editors"

What the end up saying is, you make your game with polygon counts of that of CGI movies, or even more, then scan them into the engine so you get that level of detail but you don't have to render that amount of polygons. One of the big things about this engine though that I like is there is absolutely ZERO pop-up/pop-in of graphics since there is no model swapping.

This video discusses the points I've brought up. Looks very good.

Also a comment by this Mr. Dell on another site "There was always a problem when you use point cloud data ( we say that because voxels have such a bad name , poor little things )" ... he continues "Polygons runs fast but has poor geometry, Ray-trace and voxels have perfect geometry but run very slowly. Unlimited Detail is a fourth system, which is more like a search algorithm than a 3D engine. It is best explained like this: if you had a word document and you went to the SEARCH tool and typed in a word like MONEY the search tool quickly searches for every place that word appeared in the document. Google and Yahoo are also search engines that go looking for things very quickly. Unlimited Detail is basically a point cloud search algorithm. We can build enormous worlds with huge numbers of points, then compress them down to be very small. The Unlimited Detail engine works out which direction the camera is facing and then searches the data to find only the points it needs to put on the screen it doesnt touch any unneeded points, all it wants is 1024*768 (if that is our resolution) points, one for each pixel of the screen. ... The result is a perfect pure bug free 3D engine that gives Unlimited Geometry running super fast, and it's all done in software."

If this engine works out, I can see those new 6 and 12 core CPU's finding a home in every single computer since you will not need a 3D card anymore... unless of course they can get this running on OpenCL/Cuda.
 
Last edited:
Rendering a point for only every pixel needed is a logical and clever idea IMO, look forward to hearing more about this.
 
bigdogchris said:
Yeah ... and if this is legit they are going to do something with it by turning it into a modern graphic engine rather than letting the technology rot. Oh by the way, if you watch the rest of their videos they say "we're working with Voxel editors"

What the end up saying is, you make your game with polygon counts of that of CGI movies, or even more, then scan them into the engine so you get that level of detail but you don't have to render that amount of polygons. One of the big things about this engine though that I like is there is absolutely ZERO pop-up/pop-in of graphics since there is no model swapping.

This video discusses the points I've brought up. Looks very good.
Click to expand...

I prefer ray tracing to voxels as the next step in graphics. I really don't know how realistic this can look, it's not easy to coordinate a voxel cloud to have the right appearance and texture as opposed to throwing a photo realistic texture with bump mapping and lighting and occlusion onto a polygon model.
 
Spare-Flair said:
I prefer ray tracing to voxels as the next step in graphics. I really don't know how realistic this can look, it's not easy to coordinate a voxel cloud to have the right appearance and texture as opposed to throwing a photo realistic texture with bump mapping and lighting and occlusion onto a polygon model.
Click to expand...
-Sn1PeR- said:
Cool -- wonder if it will be fast enough. The video seems like they are running at a really low framerate
Click to expand...
As for the speed, Mr. Dell commented "we are busy optimizing and are up to 25-65 fps depending on whats on screen. We want to be able to bring the system to the point were we can say we are as fast as a graphics card but much more powerful, once speed is out the way we will begin the import of laser scanned point cloud data, at that stage Unlimited Detail should look completely real."

I am wondering though how this system would work with moving objects. Sure, they can design an engine to display real images and make it look nice moving around in a 3d environment but that's not a game engine with moving characters. Also, what about ambient effects or things like shadows/smoke etc. This engine seems totally like a middleware type where it would render your game world but that's it ... sorta like Speedtree or something.
 
Last edited:
This technology sounds a lot like the concept used for the PowerVR chipset:

* PowerVR used infinite planes instead of polygons to render scenes. This allowed for much more complex objects using less scene data.

* PowerVR used a sort algorithm to determine which planes were visible within a tile. This is as-opposed to traditional renderers, which do a Z-buffer check for every triangle overlapping a pixel. This meant that the overhead processing geometry was much lower, and had the side-benefit of adding hidden surface removal for free.

While this technology is certainly not "unlimited," the quantum leap in efficiency of processing could make this worthwhile. But they will have a tough time making headway, as this is simply not feasible on a desktop single-socket machine (their demo was done on a server). So, either Nvidia or ATI will have to sign-on, because this tech still craves dedicated hardware.
 
I just hope I'm still young enough to appreciate it if/when something like this becomes reality.
 
ValeX said:
So it's almost as if they have to PRE-render all of the possible screens and then it uses this "unlimited-detail search engine" to pick out the screens that actually occured based on your user inputs and then displays them in order?

Wouldn't that take like...an unlimited amount of storage?

I must not understand how this works...
Click to expand...

No, it's basically just "searching" for the points that are in the viewport of your screen (1 point for each pixel) and then displaying those points. So as you get closer more points are displayed, as you get farther away less points are displayed, but it looks like the same level of detail to you because its always 1 point per pixel. :)

Personally, I think it's mostly BS. This technology is not new. Crysis uses voxels for terrain, although it probably converts them to polygons before rendering. I'm guessing the secret here is their proprietary "It's like Google!" rendering engine that converts the voxels into pixels on your screen. The fact that it's only getting 25-26 FPS on what is likely extremely high end hardware is discouraging. How will animating work? How will texturing work? What about collision detection? What about transparency and effects? Imagine the time required to model all these things with millions of points. It just seems like there are so many problems with creating an actual game with this thing. It might be decent for real-time visualization, but I doubt it will work with anything interactive.

I'll believe it when I see an actual game prototype being played at over 40 FPS.

Also, it looks pretty damn ugly.
 
Brent_Justice said:
Rendering a point for only every pixel needed is a logical and clever idea IMO, look forward to hearing more about this.
Click to expand...

voxels already do this... infact this was done years ago by the guy that made the duke nukem 3d engine. this stuff was being done on pentium 1's and shit years ago, fast enough for real time.


there's nothing done here that voxels haven't already done, i suppose this is just more fine grained to the point where you're down to pixel-level detail vs little cubes.


http://voxels.blogspot.com/
here's a CUDA implementation of voxels.

and yes, i'd think voxels / point cloud stuff could certainly be done very, very fast will proper hardware support.


one thing they don't mention though is that animation is your worst enemy, as recalculating the data structures to handle dynamic geometry will probably annihilate crush you
 
Last edited:
Brent_Justice said:
Rendering a point for only every pixel needed is a logical and clever idea IMO, look forward to hearing more about this.
Click to expand...

Ditto. I hope something comes of this. It'll take time, but I don't plan on dying tomorrow so......:D
 
Voxels powered the snes commanche game, using the snes 3.58 mhz cpu and built in super fx 2.

Here are a few scans:

comanchesnes3.jpg


comanchesnes04.jpg


Full link to scans:

http://www.unseen64.net/2008/04/14/comanche-snes-unreleased/

This is a very blurry video of commanche, ces 1995:

http://www.youtube.com/watch?v=4uZeF4AvxM0

Voxels have always been far more advanced then polygons, but developers are lazy and as such, we use polygons.
 
LstOfTheBrunnenG said:
brb, posting my perpetual motion machine in fs/t
Click to expand...

Way to contribute to the discussion. Your just so dam witty. *cookie for you my boy *

Anyway I did a little more digging and it appears there is something to this technology however it really has never been pushed properly or had the correct or big money backers ir needed. I don't think team green or red would like to see this ever evolve into anything.
 
There was an interesting discussion about these on Reddit a few weeks ago: http://www.reddit.com/r/gaming/comments/bbg9c/unlimited_detail_the_end_of_poligon_based/

Here's the top comment:
Ahh, sparse voxel octrees. Carmack intends to use them in the id Tech 6 engine (post-Rage). Here's what the video didn't mention: you can't animate these. They are as utterly inflexible as sprites. The closest you can come is to define every frame, can still have amazing object detail, but can't be procedurally animated (e.g. by ragdoll physics) and will move at a fixed framerate with no obvious tweening method. You can have this voxel world and it will look awesome, but it will be almost completely static. By the sound of things this company is using a binary search like the PULS 256-byte raytracing demo, so at least bounding volumes will work and moving static objects around won't be a huge performance issue.
Five-sided lettuce, on the other hand, can be tessellated into thousand-sided lettuce and molded in accordance with its depth & normal maps, which can in turn use virtual texture quadtrees to have infinite two-dimensional detail (see id Tech 5). The underlying model of dozens or hundreds of polygons can be warped, skewed, or reshaped at an arbitrary framerate from sparse keyframes or procedural instructions.
I like the technology presented, but the presentation itself is vapid and disingenuous. This is not a cure-all. Even Carmack is only using it in conjunction with polygonal actors.
Click to expand...
 
Zellio2010 said:
Voxels have always been far more advanced then polygons, but developers are lazy and as such, we use polygons.
Click to expand...

No, its because Voxels have serious problems. Animation being a major one. You can't just rebuild your voxel tree every frame, its just too slow. Polygons scaled, voxels didn't.

As for the Unlimited Detail demo, they don't show of some major features. Does it do dynamic lighting? Global illumination? Ambient occlusion? Soft shadows? Polygon count isn't super important anymore, its the lighting and textures that really make you sit back and say "wow". Also, did you notice that they only actually had a smallish number of points and then instance the crap out of them? 8 billion points with a 32-bit float for x, y, and z is already 96GB of data. Then you also have the color data for each point as well as the data for the structure storing and sorting the points and you're well over 200-300 GB of data. Hence why they did boatloads of instancing. I don't know about you guys, but I kind of expect my games nowadays to NOT reuse the same model over and over and over again.
 
Savoy said:
Way to contribute to the discussion. Your just so dam witty. *cookie for you my boy *
Click to expand...
Posters before me had already enumerated the problems with this technology. Yeah, it looks damn pretty, but good luck getting any kind of animation or effects working without absurd effort.

You really can't get something for nothing. If using a search algorithm was the only thing standing between voxels and world domination, don't you think the PhDs at Intel, nVidia, or ATi would have caught on by now? Most of these engineers would have learned about binary search et al in the first or second year of their undergraduate work.

We're just now getting really awesome destructible environments (Red Faction notwithstanding) in games. If all of Bad Company 2's buildings were represented as infinite clouds of little dots? Good luck trying to compute physics interactions in any reasonable time frame.
 
LstOfTheBrunnenG said:
We're just now getting really awesome destructible environments (Red Faction notwithstanding) in games. If all of Bad Company 2's buildings were represented as infinite clouds of little dots? Good luck trying to compute physics interactions in any reasonable time frame.
Click to expand...

actually, this is something that voxels excel at, destruction is simple and easy.

http://advsys.net/ken/voxlap.htm
 
I can't be the only one that thought it looked incredibly ugly? Something just seems off about that technology.
 
The art is ugly as hell but technically (from a polygon perspective) it is impressive.

That quote that legoman666 already posted explained why the video sucks ass and why voxels aint' all that they're cracked up to be better than I ever could. They're cool to play with but there still isn't a way to use them effectively in a manner that is better than currently existing systems so no one uses voxels, just that simple. Same thing goes for ray tracing and REYES systems. They all have their technical advantages but they also have lots of drawbacks that prevent them from being practical in gaming so they go nowhere and will probably continue to go nowhere until the hardware and software changes in a fundamental way, which may never happen.

The '08 Carmack interview talks about that stuff in more detail, but the gist of it was that hybrid (blending traditional style rendering with ray tracing and voxels and other custom systems) rendering engines are the future, no one system alone will be enough to improve things significantly from here on out.
 
The problem with voxels is that although they can render very complex scenes and props the way they are processed makes them look VERY rough around the edges...and even though I guess its still basically a "3D" scene, if you arent moving everything looks 2D...you can see this for yourself if you check out the demo they have here:

http://www.gpstraces.com/sven/tmp/VoxelDemos.v2.zip (from http://voxels.blogspot.com/ ...if link doesnt work go to the site, its at the bottoms of the page)

It says "CUDA" but it can also be done on CPU (I know, Im running ATi)...you will see that from afar it looks very good as though its using millions (perhaps billions) of polygons...but when you move in closer you will see how bad voxels REALLY look...quite crappy really...I feel that this is whats holding this technology back. I mean, only thing I could really see this being used in is what its already been used in...and thats flight sims where the ground isnt something youre going to see super close up...in a first person shoot or what not, it would really look quite bad.

However, I feel that what Unlimited Detail has made far surpassed voxels...some say thats all they are doing...but you can tell from the video and screen shots on their website that their method is 1000x more detail and has gotten rid of all the 2D looking scenes and rough square edged voxels...Unlimited Detail has something here that could easily make the most realistic game ever, easily. As far as characters and other things like smoke and shadows...these can still be done using traditional methods overlaid onto the Unlimited Detail technology. At least, it should be...the shadows in one video they did the narrator admits that the shadows are done traditionally and has been intergrated with their technology...

In either case, if this shit takes off...well...Matrix, here we come!
 
Spare-Flair said:
Welcome to 1995 and Voxels, this is nothing new. This is some tiny little company playing with voxels and the video is basically speaking to people who know nothing about computers at all, probably to drum up investment or something. I'll wait until we can see what kind of hardware this implementation needs.
Click to expand...

A voxel has volume (that's the whole definition). It's basically a cube. The points in point clouds don't really have volume, because each point is as small as it can possibly be - never bigger than 1x1x1 pixels no matter how close you get (though they will have to implement interpolation at some point because you can't store unlimited data unless you have infinite storage space).

So building a world out of voxels is like building with Lego or stacking sugar cubes, while building a world using point cloud data is like building it with atoms. That is, *if* the engine actually works like they claim it does.
 
Last edited:
AndroidVageta said:
However, I feel that what Unlimited Detail has made far surpassed voxels...some say thats all they are doing...but you can tell from the video and screen shots on their website that their method is 1000x more detail and has gotten rid of all the 2D looking scenes and rough square edged voxels..
Click to expand...
They're just doing voxels at a higher "resolution" (smaller boxes) than the nV demo. The hardware has improved over the years so this isn't so surprising, of course they can do more now than years ago.
 
Eriksrocks said:
No, it's basically just "searching" for the points that are in the viewport of your screen (1 point for each pixel) and then displaying those points. So as you get closer more points are displayed, as you get farther away less points are displayed, but it looks like the same level of detail to you because its always 1 point per pixel. :)

Personally, I think it's mostly BS. This technology is not new. Crysis uses voxels for terrain, although it probably converts them to polygons before rendering. I'm guessing the secret here is their proprietary "It's like Google!" rendering engine that converts the voxels into pixels on your screen. The fact that it's only getting 25-26 FPS on what is likely extremely high end hardware is discouraging. How will animating work? How will texturing work? What about collision detection? What about transparency and effects? Imagine the time required to model all these things with millions of points. It just seems like there are so many problems with creating an actual game with this thing. It might be decent for real-time visualization, but I doubt it will work with anything interactive.

I'll believe it when I see an actual game prototype being played at over 40 FPS.

Also, it looks pretty damn ugly.
Click to expand...

the demo in the video was running on CPU only no GPU hardware assist so 25fps with that kind of detail on CPU alone :eek:
thats not to say there isnt some smoke and mirrors to all this there is some kind of magic going on here id like to see more of how it really works befor i make up my mind
 
this is just wishful thinking

a game is made from more than unlimited detail (static) terrain.

From the video: "...you can create models using laser scanner..." this along should give you a hint of what it takes just to create models for the system

Now imagine rigging those models
 
EvilNando said:
this is just wishful thinking

a game is made from more than unlimited detail (static) terrain.

From the video: "...you can create models using laser scanner..." this along should give you a hint of what it takes just to create models for the system

Now imagine rigging those models
Click to expand...

They also say that you can import normal polygonal models.
 
I would love to hear what carmack or sweeney would have to say about this particular technology. Someone email them the link. :)
 
ahh the days of voxels and Delta Force..

voxels are nice, but they it seems they died out in gaming land because they just couldn't be accelerated with the graphics hardware.

this was a big deal with Delta force 2 as they tried to get around this by having the characters polygonal and the terrain voxels. it was an attempt to mix the 2 but just didn't work (didn't look right).


at the time though , delta force was awesome for its view distance. you could see over, what seemed like, an infinite distance. although the only downside is you where shooting at a literal brown single pixel that was an enemy in that infinite distance.

this has peaked some interested though, i think i'll dig out the delta force cd and just see what hte performance is like on more modern hardware (i have an AMD3400+ running win98se, see how that handles it)
 
You must log in or register to reply here.
Back
Top