Comments on: Unlimited Detail Wants To Kill 3D Cards

By: JustaName

JustaName — Thu, 10 Jun 2010 18:29:19 +0000

@ Seol,

I don’t think it’s rubbish. When it comes to efficiency, a properly distributed database is ridiculously fast. I mean, hell- forget databases. Think filesystems for a second. ZFS, ResierFS, Ext3, NTFS, etc. Btrees.

The way that efficiency has evolved with those (outside of obvious things such as journalling) are the pointers and methodology of inodes and similar technologies. If you are compare ext2 to ext4 there is a huge leap in performance.

This is no different. DX11 is heavy on resources to begin with, even though it provides a lot of great ‘functionality’ and the like. If meta data can be accumulated through a sparse distributed database and be directly relevant to 3d information that is thrown about in application, then unlimited detail is indeed viable.

As per the grid comment, it also looks like a test to prove instancing is possible. I haven’t seen the video in some time mind you, so I can’t recall what the author (or voice at least) was trying to point out.

By: Mike Talyor

Mike Talyor — Sat, 24 Apr 2010 17:58:16 +0000

I highly doubt this would eliminate the need for graphics cards. The process they are describing would perform best with a massively parallel processor. Ideally one “core” to search for and process the data for each pixel. Fortunately we have massively parallel processors – they are called GPUs or graphics cards. Just instead of processing lots of polygons, they would be used for lots of pixels. While this process may be possible to run in software like they are claiming in their tech demo, once you add proper shadows, reflects, etc. to make the game look half way decent you are going to still need a powerful multiprocessor GPU to pull it off well.

By: VoxelKing

VoxelKing — Tue, 13 Apr 2010 17:45:15 +0000

unlimited does not mean “unfinite” thats a wrong interpretation. They just mean the engine dont put limits on the grahic-designer / worldbuilder, like polygon games where you need to rebuild the same gameobject in different polycounts depending on viewing distance.

They also claim to have been working on this for over 15 years and to have been started this as an hobby – so its not just out of the blue, but still it needs a demo to be prooven right.

Interview on Facebook:
http://www.facebook.com/topic.php?uid=105528666147262&topic=15&_fb_noscript=1

By: VoxelKing

VoxelKing — Tue, 13 Apr 2010 17:24:20 +0000

The underlying technology is related to something called “sparse voxel octrees”, which aren’t anything incredibly amazing. What the video doesn’t tell you is that these are not at all suited for things that need to be animated, so they’re of limited use for anything that uses procedural animation (e.g., all ragdoll physics, etc.). So they’re very inflexible. You can get great detail, but you get it in a completely static world.

A rough summary of where things stand with this technology in mainstream games is here:
http://mainroach.blogspot.com/2009/01/sparse-voxel-octrees.html.

You will also want to check out Samuli Laine’s work; he’s a Finnish researcher who is focusing a great deal of his attention on this subject and is unlocking some of the secrets to implementing it well.
http://www.tml.tkk.fi/~samuli/

By: Johnny

Johnny — Tue, 13 Apr 2010 08:47:18 +0000

If it had any plausibility, Intel would be all over this like a rash. Ipso facto I call BS.

By: Plusfours

Plusfours — Mon, 29 Mar 2010 14:06:49 +0000

Gulag (mar 10)
Virtual dirt can be modelled effectively in finite element simulation packages. Either by a deformable network of balls connected by springs (implicit modelling), or by a pile of objects that have some affinity for each other (smooth particle hydrodynamics) The first type is used effectively for things that bend but don’t smash, the second for things that smash or slide or flow.
http://en.wikipedia.org/wiki/Finite_element_method
http://en.wikipedia.org/wiki/Smoothed-particle_hydrodynamics

By: Doggy

Doggy — Fri, 19 Mar 2010 14:31:49 +0000

I did a little background research, because as many people have noticed there is something off about the whole thing.

This is what I found:
1) Bruce Dell didn’t know what a CPU cache was.
2) The website is registered to an address that sells caravans in Australia.
3) Lack of any sort of specific information or realtime demonstrations
(I left the links at home but I can find them again if necessary).

My thoughts on how it could work:
1) There’s a huge grid of points containing all graphical details (no, it’s not unlimited it’s just big)
2) Datastructures are built from the ‘point cloud’ that allow it to be rendered from arbitrary angles and distances more efficiently then raytracing through a huge cube of data.

My problems with this are memory consumption and CPU usage.
The screenshots with the thousands of blue animals were said to be in a single pointcloud.
The resolution of this would have to be more than tens of thousands.
10000x10000x10000 x (size of a single point) would be the uncompressed amount of data to store this, which is not feasible on modern computers. Add to this the datastructure that would optimise drawing speed and it’s taking up ridiculous amounts of memory.

By: Flying dutchman

Flying dutchman — Thu, 18 Mar 2010 06:41:48 +0000

This sounds nice, but…

It’s no problem for a 3D engine to show round objects in great detail. The flopped game Republic had a novel graphics engine that allowed you to zoom in on anything without losing detail. Well, up to an impressive point. The trees did have individual leaves for instance and were not bumpmaps or low poly models. A vase didn’t look like a hexagon. The videos immediately reminded me of Republik.

There must be a reason why manufacturers did not adopt technologies like this. I bet it has to do with gaming speed, interaction and the speed of building worlds.

As for the claimed 20% increase in polygon counts, that’s not the whole story. Memory becomes better, processing power becomes better, DirectX/OpenGL becomes better. GPU’s follow Moore’s Law and double in power every 16-19 months. Proof of concept demos already show very realistic graphics.

By: Kyle Klouzal

Kyle Klouzal — Tue, 16 Mar 2010 22:12:48 +0000

If you had an algorythm that computed an unlimited amount of points, then the algorythm would run for an unlimited amount of time(lol) Then if you had another algorythm to search through those unlimited amount of points you would be searching forever because the points are an unlimited number! THUS they would have to set a finite limit on the amount of points therefor not unlimited detail. :O And a cellphone to compute such a feat? hah! yeah right.

By: Zecc

Zecc — Tue, 16 Mar 2010 21:15:47 +0000

It should be at least a year old now. See this forum post:
http://www.somedude.net/gamemonkey/forum/viewtopic.php?f=12&t=419&sid=17111f9b43feb101fd9e7ca59ec1ad6a

By: namuol

namuol — Sat, 13 Mar 2010 01:16:51 +0000

Precisely what I was thinking, KillahMate. This sort of tech has been demonstrated before… if they’ve been trying to sell this since 2008, it’s no surprise they haven’t gotten anywhere; most people who sleep on a mound of moneybags have a consultant or two.

By: namuol

namuol — Fri, 12 Mar 2010 22:56:39 +0000

@Mike Arthur:
Lightstage 5 (or any version, for that matter) is very impressive indeed, but it’s not really related to the rendering techniques presented by “Unlimited Detail”… the fact that they use voxels is kinda unrelated to the quality you see in that video.

That being said… if you want your mind blown, watch some of the videos on Paul Debevec’s (Lighstage researcher) website: http://www.debevec.org/

This is LightStage version 1.0: http://www.debevec.org/Research/LS/
Watch the video, it explains how it works pretty well. Also, play with the (windows only) demo!