Hey,

I noticed this interesting video on a new technology called 'unlimited detail' in gaming.

http://www.youtube.com/watch?v=Q-ATtrImCx4&feature=channel and http://www.youtube.com/watch?v=THaam5mwIR8&annotation_id=annotation_115233&feature=iv

Every time a new 3d graphics card or games console (Nintendo, Xbox, Playstation) is built, they make it more powerful in order to run more geometry (Put more objects on the screen and make them look more rounded). Games today still have a long way to go before they look like 3d movies (Shrek, Nemo, Toy story); billions are spent every year on R&D to make hardware more powerful. Unlimited Detail is a software algorithm that gives unlimited geometry. When we say “unlimited geometry” we really do mean it. It really is Unlimited, Infinite, endless power, for 3D graphics.

Sounds pretty neat, interesting to see if this ever comes about.

sounds cool, do you need unlimited hardware / cycles to get unlimited details?

Interesting concept, but the guys voice and the background music weirds me out.
I wonder how this works with effects like fog and reflections.

Unllliimited.

sounds cool, do you need unlimited hardware / cycles to get unlimited details?

I've just watched both videos; he lets us know its a software based solution - you don't need a GPU to use it, and that they want to put it on Nintendo Wii, your mobile phone, etc.

Sounds awesome. Be nice to see WoW using this tech.

Wasn't Carmack talking about stuff like this for id Tech 6?

I saw something about this a little while ago, but I hadn't seen these videos before. It's a cool sounding concept, and if they truly are rendering that amount of detail in real-time (using software, no less), then yeah, that's pretty amazing. There are some other things I'm still skeptical of, though.

Having it work on static meshes is one thing, but animating all of that seems nightmarish. Every frame, it would have to be calculating the updated positions on potentially millions or billions of points. That sounds heavy.

Related to that is physics. Having a few billion points colliding with one another sounds like it'd be far too much for any CPU to handle.

But, assuming they have clever solutions for optimising both of the above situations, that still leaves the problem of storage. Surely storing an entire game where every level, character and prop is made up of a few billion points would take up a stupid amount of memory. I noticed that a lot of the stuff in the videos appeared to be instanced, which is partly why the art looked kind of crap. Sure, that'd save on storage, but a quite a big expense. You're kind of just swapping one limitation for another.

But, I suppose we'll see. The theory behind it sounded plausible to me (in my limited programming knowledge), so I'll give them benefit of the doubt for now.

I also heard that this was being developed by some Australian dude, but the voice in the videos sounded British.

Shadows clearly don't work well.

So its basically just a search algorithm on top of voxels? It sounds interesting, but I can't see it scaling to commercial games. All the demos use a high level of replication of objects in the scene, which seems dodgy. Show me a final fantasy cutscene quality model in realtime plz. I would assume the two big limitations will be SDK dev time (the guy seems to imply that you simply don't use polys for modeling) and disk space. But its true that doing search on a voxel engine would reduce the primary issue of huge memory requirements (ie: tiberium sun).

Its that whole argument of "if we created a digital world which accurately mapped every element in the universe, using the ultimate compression techniques and algorithms, we would end up with something the size of the universe".

actually.........good luck to em......heaps more work to go imo.
i wanna see a car blow up using ud, that will be interseting

last edited by lewd at 15:33:33 14/Mar/10

Related to that is physics. Having a few billion points colliding with one another sounds like it'd be far too much for any CPU to handle.

I agree, it was nice seeing what they could do it with it as far as object detail, but (maybe it was the s***ty youtube quality) things didn't look that pretty, and with out animation everything appeared lifeless, id far rather keep the tech we have and work on developing more realistic animations and object interaction then focus on being able to render unlimited points via software.

Having a few billion points colliding with one another sounds like it'd be far too much for any CPU to handle.

I doubt they'd be applying physics algorithms to individual points. They'll probably group the points into a bigger, simpler object (like polygons, lol) and apply the physics to that.

Surely storing an entire game where every level, character and prop is made up of a few billion points would take up a stupid amount of memory

I'm assuming they have an algorithm that generates the geometry on the fly based on some parameters. So that they can generate/delete the points dynamically. Plus the search algorithm would stop most of the scene from being inspected for each frame.

So its basically just a search algorithm on top of voxels?

Yeah seems like it. They haven't really demonstrated how it deviates from voxels or raytracing - in fact I thought it was a combination of the two at first.

What I don't like is them using some very old games for most of their comparisons (sure it highlights low polycount, but still!). If a developer was after detail, they could easily bump up the polygon count on a tree and make something as realistic as what they are showing. But they don't really care about minute details like that. Especially now that most games seem to focus on multi-platform console support, resulting in simpler geometry, art and gameplay for easy porting.

To be honest I'm looking forward to proper raytracing. You get uber-realistic lighting and colour effects. Combine that with a good space-division algorithm and parameterised surfaces and you can have your infinite detail right there.

wow - that's so cool.

I hope we see more of this in the next couple of years. I'd imagine some pretty spiffy remakes of games might pop up in the years that follow that

http://www.reddit.com/r/gaming/comments/bbg9c/

Haha exemplars link. There is a town in the UK called Sc***horpe. http://en.wikipedia.org/wiki/Sc***horpe_problem And they have all kinds of problems due to word filters. LOLc***.

Ahhh yes my day is complete.

Animations would be tricky unless somehow they pre-recorded animations and knew what the relative positions where then it would be easier.

Things like virtual physics however... would be tricky.

This clearly disobeys the second law of thermodynamics.

Sounds awesome. Be nice to see WoW using this tech.

it would still look like my ass

but (maybe it was the s***ty youtube quality) things didn't look that pretty

That's simply because it was art done by coders. It didn't look very pretty at all, but that wasn't the point of the video. When they get some actual artists making environments, it'll be a different story (provided the tools allow them to work in a reasonable manner).

I doubt they'd be applying physics algorithms to individual points. They'll probably group the points into a bigger, simpler object (like polygons, lol) and apply the physics to that.

Yeah, that's what I was considering. As limited as polygons are, they would be fine as collision objects, I think.

I'm assuming they have an algorithm that generates the geometry on the fly based on some parameters. So that they can generate/delete the points dynamically. Plus the search algorithm would stop most of the scene from being inspected for each frame.

I should have been more specific. I wasn't talking about RAM, though that's obviously a factor. I was more talking about simply storing the data on a HDD. That to me seems like it would take up a HUGE amount of disk space. There's only so much that can be created procedurally. Level layouts, props and characters would all have to be created by the artists. Maybe they have some clever way of compressing the data. But, if they don't, I can see storage space becoming an issue.

I can't see this replacing polygons, maybe working alongside polygons but certainly not replacing them. Maybe use this stuff for static parts of the environment, but anything that needs to animate at all, and certainly stuff like characters, just wouldn't work. And yeah, on the collision side of things, stuff like bounding boxes and convex hulls would still be in use, but that doesn't have much bearing on this stuff, cos collision geomtry is not rendered anyway and they're not so much polygons as mathematical constructs.

I Watched the vid last week, I call BS on all of it. I didnt see anything in the demo video's that was amazing, actually it looked pretty s***.

Surely if its capable of doing all these amazing things, they would be able a decent demo video that didnt look like ass.

Yeah, pretty much what Khel said. And, I guess, if this really does work, and it's that efficient, then the GPU would be left with nothing to do other than render characters, so we could give them massive polygon counts and texture resolutions anyway.

I'm not sure about the viability of this tech, but one thing is certain from that video:
ATI and nVidia both need to actively head hunt BBC Channel 4 documentary presenters to spruik their future GPUs.

Edit:
Also - DirectX 10...DirectX 11...10...11...10...11...AMAZING!

The only part I don't get is that the models themselves had to be created with polygons... Unless it's some kind of funky spline system, but their rock models shows the exact same polygon issues. If they had been generative it may have been different.