Unlimited Detail Technology
category: offtopic [glöplog]
http://www.youtube.com/watch?v=Q-ATtrImCx4
what do you think about it?
it seems like voxels.. but what about rendering method..
i mean there is no information on their site..
http://unlimiteddetailtechnology.com/
what do you think about it?
it seems like voxels.. but what about rendering method..
i mean there is no information on their site..
http://unlimiteddetailtechnology.com/
I say raymarching. The "wireframe" view looks like iq's snapshots showing the number of iterations in the edges of the equation. As they say, it's not voxel, it's not triangles, it's not raytracing ...
Quote:
i mean there is no information on their site..
well, there's enough buzzwords in there to swindle a bunch of technologically incompetent investors out of decent but not spectacular amounts of money, what else do you need?
hm interesting. But what's *that*
all about?
all about?
We watched their video during lunch today, and found it ridiculous how bad their tech demo looked. Maybe the technology behind is solid, but they should at least hire a decent artist.. After all, they want to sell something.
Quote:
If you have a background in the industry you know the above pictures are impossible. A computer can’t have unlimited power and it can’t process unlimited point cloud data because every time you process a point it must take up some processor time. But I assure you, it's real and it all works.
That's believable.
free energy for everyone
That's crap. Anyway if you look at the videos it's not that great. You see some gaps from time to time. Polygons with subdivision look a lot better, on today's hardware.
love that logo!
now that picture answers the question, if you had unlimited graphics power... what would you do with it?
play Rob Is Jarig in HD
unlimited ugliness.
The videos look promising, the description sounds like bollocks. Well, crazy shit comes true once in a while, no harm in wishing.
I think, based on the point they make about 'conventional polygon technology uses model swapping, which looks crap, but we don't have to do that', it's probably something along the lines of traversing a massive octree and bailing out of the recursion as soon as you reach pixel resolution. Possibly with a directed graph rather than an actual tree, so that the 'sierpinski pyramids of camels' scene (or whatever they were) doesn't have a load of redundancy.
But then I'm somewhat distanced from state-of-the-art 3D rendering, so for all I know I may be describing a well-known 20 year old algorithm there.
But then I'm somewhat distanced from state-of-the-art 3D rendering, so for all I know I may be describing a well-known 20 year old algorithm there.
I love the concept of "unlimited point cloud data", especially the "unlimited ... data"-part. Who are they trying to kid? Most of their "massive geometry"-samples looks like axis-aligned volumetric data-structures which can share leaf-nodes:
- All their data seems strangely repetitive and axis-aligned
- Everything looks pretty damn static
- They surely don't like complex shading models .. or any of the other problems that are considered HARD these day. They only seem to focus on geometry, which is NOT the biggest problem of today's rendering-model.
- Their "green x-rayish image" looks like normal kd/quad-tree traversal-depth plots.
They also does not mention anything about how these "unlimited point clouds" are created. See, that's really the biggest value of polygons today - they are grok-able for artists. And that's why "limited" data-models is going to prevail; it's well understood how they work - we have a massive work-force that knows well how to make data for these data-models.
I also like how they state "absolute truths" like claiming that 24bit color-depth is the "end of the road".
Ok, I'm going on an Optimus-style rant here, but I'd like to state the following: the reason why they didn't reach the board-rooms of nvidia and ATI might *not* have to do with the rendering-technology, but with the lack explanation of the other (and much, much more interresting) aspects of rendering.
It doesn't matter if you try to sound like Statix or not. You're not Alex, and his point about the cardboard-cutouts wasn't as much about polygons being planar as it was about the lack of an unique look and feel. At least that's how I understood it. Optikusma over and out.
- All their data seems strangely repetitive and axis-aligned
- Everything looks pretty damn static
- They surely don't like complex shading models .. or any of the other problems that are considered HARD these day. They only seem to focus on geometry, which is NOT the biggest problem of today's rendering-model.
- Their "green x-rayish image" looks like normal kd/quad-tree traversal-depth plots.
They also does not mention anything about how these "unlimited point clouds" are created. See, that's really the biggest value of polygons today - they are grok-able for artists. And that's why "limited" data-models is going to prevail; it's well understood how they work - we have a massive work-force that knows well how to make data for these data-models.
I also like how they state "absolute truths" like claiming that 24bit color-depth is the "end of the road".
Ok, I'm going on an Optimus-style rant here, but I'd like to state the following: the reason why they didn't reach the board-rooms of nvidia and ATI might *not* have to do with the rendering-technology, but with the lack explanation of the other (and much, much more interresting) aspects of rendering.
It doesn't matter if you try to sound like Statix or not. You're not Alex, and his point about the cardboard-cutouts wasn't as much about polygons being planar as it was about the lack of an unique look and feel. At least that's how I understood it. Optikusma over and out.
maybe they've got something ; I long ago saw one other site with no info and a "groundbreaking" new tech...
but i really, really cant stand the attitude and tone of the speaker. instant hate.
i'd say, it's raytracing a bsp of voxels, or something like that.
linear to the resolution, log to the number of voxels.
dynamic scenes : impossible.
Then again i might be wrong.
but i really, really cant stand the attitude and tone of the speaker. instant hate.
i'd say, it's raytracing a bsp of voxels, or something like that.
linear to the resolution, log to the number of voxels.
dynamic scenes : impossible.
Then again i might be wrong.
It probably is only good for static geometry in limited cases, but still.. what's raymarched distance fields good for? Demos? Maybe it's time for the return of the env mapped flyby.
the plot thickens :)
http://www.tkarena.com/Articles/tabid/59/ctl/ArticleView/mid/382/articleId/38/Death-of-the-GPU-as-we-Know-It.aspx
At the end there are q&a
http://www.tkarena.com/Articles/tabid/59/ctl/ArticleView/mid/382/articleId/38/Death-of-the-GPU-as-we-Know-It.aspx
At the end there are q&a
Quote:
Questions and Answers with Bruce Dell, Lead Programmer/Inventor
Q: How long have you been working on Unlimited Detail?
A: Unlimited Detail has been around for four years but in fact taken thirteen years to develop. There are many problems to overcome when you are going to completely rewrite the 3D system from the ground up! It’s only now about to move in to the commercial scene because it’s finally found a way to run fast enough.
Q: How Finished is Unlimited Detail?
A: All the R&D is finished and it’s all working, we’re just optimizing it and getting a few visual bugs out of it. Then I’m off to Silicon Valley to see who wants it! It’s running 30 fps at 1024x768 with no hardware graphics accelerator being used, but it has not yet been written for dual or quad core systems. Theoretically, if the system reaches dual core 50 fps in software alone then the industry will have the biggest change in graphics since we went from 2D to 3D.
Q: How Does It Work?
A: It works like this: If I gave you a billion books and said “be a dear and fetch Little Red Riding Hood for me” then you would have to read the title of a billion books until you found it. But If I sent you to a library that had a billion books and asked for Little Red Riding Hood, you would go to the children’s story section then the “R” section then the “RE” section then the “RED” section etc until found. Looking for things in this way takes only a few jumps, as opposed to reading a billion titles.
Unlimited Detail files all its data in a strange format so that it can jump through the data really quickly and bring back only 1 atom for every one pixel on the screen. Sounds easy, until you realize that it all has to be retrieved from a 3D camera angle and that’s the tricky part. Unlimited Detail has found an algorithm that retrieves points from 3D aspects and jumps over all the points that aren’t going to appear on the screen. This was always theoretically possible, but no one knew how to do it until recently.
Q: Does it gobble up a ton of memory?
A: No it uses about 40% less memory then pixel for pixel polygon models.
Q: Can you have moving sprites and animation?
A: Yes, it was tricky but Unlimited Detail found a way to make the data dynamic and it bends and stretches and pretty much does everything polygons can do.
Q: Can you have deformable environments?
A: Yes, not on the polygon object level but rather down to the 3D atom level. So footprints in the sand won’t be stuck on flat textures, they will be real indented geometry and it will stay that way rather then disappearing after 30 seconds.
Q: Does it have real time lighting?
A: Yes, though only one light source is being used at present. There is a way to reform the Unlimited Detail system to give unlimited light sourcing, but it is only theoretical at present.
So...
-it's fucking voxels :) , unlike he previously told , even if they are in an accel struct.
-1024@30fps...even RTRT can do better if i'm not mistaken. Of course it depends on the scene and the quality.
-appears to be log(voxels) :"the “R” section then the “RE” section" ... well, how much is log(inf), bitch? :)
-"bring back only 1 atom for every one pixel on the screen": still not clear if there are precomputed "voxel mipmaps" to find a voxel of size equal to 1 pixel...then again dynamic things become tricky at best.
-"Unlimited Detail has found an algorithm that retrieves points from 3D aspects and jumps over all the points that aren’t going to appear on the screen." : maybe after all the guy has something quite new, maybe it means he's got a structure where tracing rays from one point is natural.
-it's fully DYNAMIC! well done. "it was tricky" :) ... "pretty much does everything", mmh? pretty much or everything?
-one light for now, lolwut.
-cockiness :) "the biggest change in graphics since we went from 2D to 3D" :)
-it's fucking voxels :) , unlike he previously told , even if they are in an accel struct.
-1024@30fps...even RTRT can do better if i'm not mistaken. Of course it depends on the scene and the quality.
-appears to be log(voxels) :"the “R” section then the “RE” section" ... well, how much is log(inf), bitch? :)
-"bring back only 1 atom for every one pixel on the screen": still not clear if there are precomputed "voxel mipmaps" to find a voxel of size equal to 1 pixel...then again dynamic things become tricky at best.
-"Unlimited Detail has found an algorithm that retrieves points from 3D aspects and jumps over all the points that aren’t going to appear on the screen." : maybe after all the guy has something quite new, maybe it means he's got a structure where tracing rays from one point is natural.
-it's fully DYNAMIC! well done. "it was tricky" :) ... "pretty much does everything", mmh? pretty much or everything?
-one light for now, lolwut.
-cockiness :) "the biggest change in graphics since we went from 2D to 3D" :)
If this were for real, the guy would get himself a software patent, provide a free downloadable demo for everyone to sample, and within days he'd no doubt have plenty of offers from various companies wanting to licence his technology. But alas, no demo, just some vague descriptions. He's simply trolling for venture capital.
It's also suspicious that he's been publicly claiming a breakthrough for more than a couple of years, but no company has yet come to his financial rescue.
Colour me skeptical.
It's also suspicious that he's been publicly claiming a breakthrough for more than a couple of years, but no company has yet come to his financial rescue.
Colour me skeptical.
of course it' real, but it's unimpressive (don't want bore you guys with links to the good old gigavoxels, modern raytracing, etc...), useless (static geometry blah blah) and pretentious. And very ugly too.
Is this all a joke?
Is this all a joke?
If that's ever made into something useful, I'm definitely interested. I fucking hate Ati & NV hw difference-hell and would really like a solid software rendering lib without the headaches of incompatible hw shit (and polygon-based rendering is just stupid and a timewaster to me anyway).
