Whats your guys take on this?
http://unlimiteddetailtechnology.com/home.html
Looks impressive.
My brother was reading something about bit/point clouds in a New Scientist magazine.
Will look deeper into this…
It was talked about a bit here.
http://forum.unity3d.com/viewtopic.php?t=46071
I am very skeptical. They leave a lot of holes in their explanations and examples, and have not made any progress at all in months. I say its an investor trap.
It actually looks pretty neat. If this is true and as awesome as it says, I think 3D Graphics in games will take a huge leap forward.
I was initially very excited about this…and yet, nothing else has come of it. Many people asked some very specific questions regarding the static nature of environments, animation…nothing. They should at least have released simple, playable demos or something, even of simple environments.
So…
The trouble is their insistence on unlimited detail. Nothing is unlimited in computing.
In addition, their web site looks awful and their tech demos look like they’re made of Styrofoam. They really don’t have much to show, and haven’t seemed to make any progress since I first heard about them many months ago.
That said, if this is valid technology, it would be really awesome.
It would need to be Hybridized with current Rendering tech I’d think, as Animation would likely cause a system like that to scream to a halt.
It also seems to use a lot of Hyperbole and comes off as an over inflated ad rather than an effective tech demo.
Unlimited Detail seems interesting, but I think I will try to hold off on too many more judgments until it is more demonstrable.
Too little too late.
With DX11 and hardware tessellation, we’re already starting to see polygons that are smaller than pixels when some scenes are running on newer graphics cards.
This voxel+ technology would require a whole pipeline revamp and new tools. It’s not worth the investment considering it’s still quite a few years away from looking decent. And with polygon crunching being so easy now, there really is no point.
The website makes the whole thing look very suspicious. They go to great lengths to explain what polygons are and barely mention voxels. And there is no word about animation.
Ask a random gamer if they know how those shapes are made…
Anyhow, I would believe it when I see it. It sure has a pretty crappy website to be some new company with a system of “Unlimited Detail”. If it can run unlimited, I’m going to go into my old habits of SUPER-SMOOTHING*!
*: mesh → smooth, repeat x 5.
I wish this would work out, voxels would have many advantages if only they found a way to process them fast enough for animations and physics.
With polygons you need weird ways to work around their true nature, pretty much everything with polygon is a way to work around it, since it’s all flat polygons trying to play as volumetric mass.
With voxels, if we had the technology, everything would work straight away, for example fluids and melting things, destructible environments and characters, terrains. With voxels you got real volume, you can cut something by half and see what’s inside. That’s just plain awesome, we need that ASAP
this technology is nothing else than an evolution of what Comanche etc had years ago already, brought up to date with current generations gpu to utilize their raw power to overcome the limitations a cpu poses in parallelization of linear equation systems like physics within such a point mass etc
its interesting but years away from actual usage in an end user environment like a game.
Right now its primarily a research and science interest field that someone brought to the public in a pretty much non-mature state.
I’ve seen more powerfull and usefull usages of point based systems years ago already in university and not even those have found their way into real world usage cause there are too many restrictions on their usage at the time and the amount of data required to store such clouds is beyond any realistic measure (keep in mind, each point has to store various data, at worst each of them needs to store more data than a vertex with multiple texture pixels on it would have to store … thats a horrible amount of data, causing gigs of information in no time and without any significant gain - thats the nature of all voxel data where its not end processed due to assumptions like “its now static” in which case you cut out the data from the whole volume within etc)
With the release of fermi, completely altering the traditional opengl or dx rendering pipeline is completely possible. Fermi is completely programmable, there is no hard wired pipelines in it. This isnt too late, the possibility to do radically different rendering pipelines on GPUs actually just became possible.
Tim sweeney was talking about how he sees dx and opengl fading out in the future now because there largely unnecessary with fermi. A game engine coder can write the engine to render in any way he wants it to on fermi, much how programmers would write software rendering pipelines for old games,that can be done now, but the GPU can run it. Sweeney was talking about how the future unreal engine rendering pipelines may not be ogl or dx, but they will use some new method entirely that they write.
I was actually referring to the workflow pipeline at studios with a complete set of new tools, new connections, new problems, etc…
If you don’t believe me watch the video explaining it. The server must be located in the middle of a feedlot.
Obvious reasons as to why this most likely will not work : lighting, shadows and animation.
-
Imagine a GPU calculating how light reflects off a 10-billion-vertex-scene and/or how it portrays a shadow onto other 10-billion-vertex-objects.
-
Imagine how a CPU will compute all of the billions of vertices that will have to move whenever a bone in a character is moved. (This is the basics for animation.)
Current GPUs and CPUs simply can’t make this applicable for any other purpose than static images. But hey, maybe these guys made some ground breaking algorithms that also interpolate light and animation.
Possible. Very unlikely.
Here’s an idea I just had.
One way to look at the difference between rasterization and raytracing is that rasterization is a forward process (take all these verts, transform/project, touch pixels) while raytracing is sort if backwards (start at the pixel and figure out what the thing is that affects it). But the rasterization step itself (and texture mapping) is backwards, i.e. in image space. So how does texture mapping work? The u, v coords are linearly interpolated across the poly, but so is the homogenous coordinate “w”. The texel lookup requires dividing u and v by w which gives the perspective corrected mapping back into texture space. The same thing happens with 3D textures. So what if the mapping weren’t so simple? What if you had a way to inverse map from a screen-space x,y,w back to the proper point cloud entity? Of course the pixel is really the bottom of a cone, and you have to find the nearest point in that cone. What kind of data structure could do that quickly in a view-independent way? One obvious thing that comes to mind is a k-d tree, maybe with some kind of DAG (directed acyclic graph) format. Or maybe just a DG since that would make it truly “unlimited” (of course it would be repetitive, but they aren’t claiming their system isn’t repetitive).
So what I’m describing is an acceleration data-structure for raytracing, but they say they aren’t doing raytracing, so perhaps I’m wrong. But then again, maybe I’ll try coding this up and see if it works 