Welcome to Fractal Forums

http://www.youtube.com/watch?v=00gAbgBu8R4

So what do you all think, is this the future of 3D rendering or is it bogus? And what do they mean by "unlimited" ?

There are even some claiming the examples in this video are fake, apparently...

unlimited detail requires unlimited memory, that's what they are not telling you.

I've seen examples of point cloud or voxel rendering that merely consisted of a few heaviy detailed objects being repeated many (somtimes even a gazillion) times over.

This technique profoundly fails at animation (which is essential for games), and it fails at large scenes containing many different objects.

By the way, this is what I like about 3D fractals. Infinite detail with minimal memory requirement. And they are easily animated by changing a few parameters... I'd love to see a space shooter happening inside the bowels of a Mandelbox... on the iphone ;)

It's fake.

I don't think it is fake, but neither is it revolutionary. The big question is how they will store the information, without heavy reuse/instancing of objects.

There is a good post here, comparing to other Voxel visualizers:
http://procworld.blogspot.com/2011/08/unlimited-detail.html

And if the tech really was that innovative, how come they didn't patent it?

I don't think that it is done in real time! Even they say it's done on software. It has to be a supercomputer (and a mega-hard drive for all that information)

http://www.youtube.com/watch?v=JVB1ayT6Fdc

I watched a little bit of it and apart from "taking one atom per pixel on the screen" (which is just ray tracing isn't it) he doesn't say how they do it (as I mentioned I haven't seen everything so it might actually be in there), which is essential if they want to convince the critics.

+1 for concise summary and clear thinking  O0

I think the key thing here is "unlimited detail" but no mention of "unlimited variety" ;)
It's quite obvious that many identical, or almost identical objects are repeated many times in the demo videos.

Personally since he mentions "search" many times I guess the tech is hierarchical object instances with each object pointing to *all* those adjacent as well as subsidiary objects with a search algorithm that effectively ray-traces, just not in straight lines but rather following the closest hierarchical links to the ray's path until an object on the ray is found at which point sub-objects are checked in similar manner.
Essentially irregular voxels -- i.e. atoms.
This method would lend itself well to being quick to render but also highly compact as long as the variety of different objects is not too large - though for most applications adding a single random value to distort/modify individual items would perhaps be enough to introduce sufficient variety at least at maybe the scale of leaves/grass etc.
The great thing is the method could quite easily be combined with almost any other conventional method - at almost any level within a combined hierarchy - including adding true ray-traced fractal objects ;)

exactly right; what they're showing here is commonly called "instancing", and frankly a very poor instance of it ;)

just recently these fine examples of instancing were brought to my attention (http://www.indigorenderer.com/forum/viewtopic.php?p=109502#p109502):

(http://www.indigorenderer.com/forum/download/file.php?id=13719&t=1) (http://www.indigorenderer.com/forum/download/file.php?id=13719&mode=view)
(http://www.indigorenderer.com/forum/download/file.php?id=13718&t=1) (http://www.indigorenderer.com/forum/download/file.php?id=13718&mode=view)
(http://www.indigorenderer.com/forum/download/file.php?id=13716&t=1) (http://www.indigorenderer.com/forum/download/file.php?id=13716&mode=view)

fractal iteration is essentially a procedural form of instancing, and programs which render 3d julia sets, mandelbulbs etc are all concerned with intersecting these procedurally-defined bounding volumes against light rays. this is truly infinite (modulo finite precision, storage and computation etc...) detail, and uses even less storage than their approach  - near zero - for this particular application.

Thanks for those Thomas - I edited as you posted ;)

i forgot to mention that all the scene geometry (which includes an incredibly detailed house interior) takes 680mb of memory, which fits comfortably on a geforce gtx 260 - released more than 3 years ago (june 2008).

i wonder how many little voxels you'd need to get this kind of detail... bear in mind that the leaves are all partially opaque, and unless you assign an opacity for each voxel that's literally going to need "atoms"...

I don't think the "unlimited detail" they are talking about requires unlimited memory.
I think they simply mean every pixel on the screen will be unique.

I was excited about this, but not sure what to think now
http://www.youtube.com/watch?v=8vGXqXL7zrI

If you're basing your opinions on YouTube videos, instead of either studying Computer Graphics or reading what people who do study it say, then you're almost always going to get someone's "armchair expert" opinion (i.e. nonsense).

Agreed.
I don't know what to think after reading the posts on this forum, not after watching the video I posted.
I just wanted to see if anyone had anything to say about what the guy in the video says.

The atomontage engine look much more promising :)