unlimited detail software renderer

[erikd]

http://unlimiteddetailtechnology.com/videos.html
http://unlimiteddetailtechnology.com/description.html

Looks interesting

[princec]

Aye, old news though. Wonder why they've not managed to sort anything out yet? On the other hand listening to the voiceover on the demo reel they did way back I think I can guess why.

Cas

[erikd]

They mentioned that they planned to have the SDK released in about 16 months (that should be somewhere late spring next year)

Games published by our own members! Check 'em out!

[JL235]

About 9 months ago I remembered this and did lots of Google around to see if anything had happened. Generally I read really bad stuff about it. Lots of experts I came across claimed that there are research methods already which can do this (which they predicted this was based), and that it's only efficient for fully static data (because of how it's indexed). Once you want to update or alter the scene (like adding, removing, moving or animating objects) it suddenly becomes really slow which would make it unusable for games.

But those 'experts' were just stating their own theories and conjecture. Even with a fully static scene, if it runs like it does in the videos on a standard PC then it would clearly look very impressive if it had some decent models.

[Orangy Tang]

We've done this before: http://www.java-gaming.org/topics/unlimited-gfx-detail/22102/view.html

I stand by my previous comments that it's too static to be practical for use in a game, and iD have already done far more interesting experiments with sparse voxel trees.

[princec]

I'm not sure I see why you couldn't in theory blend a whole bunch of trees together using the depth buffer, and have some of them animated and some of them not. Or even mix the voxel stuff with ordinary 3D rendering.

Cas

[erikd]

I stand by my previous comments that it's too static to be practical for use in a game

They did post a video on youtube that shows off animation:
http://www.youtube.com/watch?v=cF8A4bsfKH8

It may not be new and there may be other similar efforts (I haven't followed this kind of thing for a long time), but I hope things will eventually move in this direction and away from current (IMHO short-term) solutions of having separate video cards.
A separate polygon pumping GPU with it's own separate memory seems a bit like a dead end to me on the long run, so I hope things will move towards having a more unified architecture where you do your rendering completely in software. A bit like how the PS3 was originally envisioned.

[DzzD]

I remember that we already talk about this (unlimited detail) in another thread http://www.java-gaming.org/topics/unlimited-gfx-detail/22102/view.html

[princec]

I don't see how having separate GPUs is a dead end. In fact it's probably going to be the future for a long time - all it is is bolt on purpose multiprocessors. Maybe when a home computer contains a 128-core fully configurable grid computer and appropriate O/S it might be.. but for now it's definitely the best solution. Pipelines ahoy!

Cas

[erikd]

I might well be wrong, but the way I see it is that having separate bolt-on GPUs was mostly a practical solution, but not an ideal one. I'd say that it's quite limited by bandwidth constraints if you want to use all that processing power for any other way than how it was specifically intended.
For things like this 'unlimited detail' thing (or other software rendering approaches), it's probably more efficient to have a more unified architecture?

Games published by our own members! Check 'em out!

[DzzD]

what I would wish for the futur would be something as a 256 core CPU and no more GPU, it is not what I think wil happen but IMO that would be the more practical for everybody

[princec]

1080 cores Then you could raytrace every line in a frame, hehe.

Cas

[appel]

I wonder if you could mix together this static environment and dynamic environment with traditional 3D.

 

[erikd]

I wonder if you could mix together this static environment and dynamic environment with traditional 3D.

 

Yes, I can't see why not?
I suppose this algorithm could write a depth map, and it could be possible to dynamically (or even statically) generate some sort of low-poly collision model of your surroundings.

[Riven]

This engine mixes voxels and rasterization:

<a href="http://www.youtube.com/v/Gshc8GMTa1Y?fs=1" target="_blank">http://www.youtube.com/v/Gshc8GMTa1Y?fs=1</a>

<a href="http://www.youtube.com/v/1sfWYUgxGBE?fs=1" target="_blank">http://www.youtube.com/v/1sfWYUgxGBE?fs=1</a>

[appel]

Nice terrain.

Now, we gotta have some game. I was a bit surprised he never zoomed too close to the terrain :\