Polygons to be out-dated?

[jdude]

 

An Australian company called Euclideon has released a video of their new technology that is supposedly ready for game developers to start using quite soon. The tech removed polygons and replaces them with a 'atom' based system. It will be a titanic leap forward in graphic technology, it can render 100,000X more than existing 3d technology. If this technology is as good as it sounds, so long to the millions of dollars many companies have invested into polygon based engines.

 

I thought this was very cool.

[nbohr1more]

Several companies including id Software (Id Tech 6) are working on Voxel or Point-Cloud render methods. It's a natural extension of scaling triangles to sub-pixel precision. Though the needed data-set is huge... Thus far, I've only seen Carmack remark on ways to keep that in-check... (Mega-Voxel etc)

[jdude]

By large data set I'm assuming you mean they take up loads of hard drive space, right? Funny how Carmak is already working on ID6 when he just finished ID5

[rich_is_bored]

I'll quote myself from the relevant topic on D3W...

 

These guys are parading around old tech like they invented it. The word your looking for is voxels. Punch that into google and you'll be much more impressed.

 

The reason this hasn't become the standard is because it has it's share of problems. How do you handle dynamic lighting and shadows? How do you handle animation? How do you handle collisions? How do you handle the bandwidth constraints imposed by data sets that measure in terrabytes? I love how this video and the one that was released a year ago are all about how detailed things could be while ignoring all the technical problems.

 

Until these guys upload a video demonstrating their tech powering a game or at least something game-like, there's nothing to showcase. The lighting is static. The world is static. Who cares how detailed it is?

[jaxa]

I heard a lot of skepticism about it 2 hours ago but now that I'm seeing the demo vid I'm impressed. Like you've said though, I have no idea what's on the other end, maybe 1 PB of data and a supercomputer? Better hurry up with those holographic discs, HP. I read the transcript and I don't really appreciate the term "atoms" heheh. So these aren't voxels but are "point cloud data"? What the hell does that mean? Anyway, thanks to this topic I'll be tracking this story a lot more closely than I was planning to.

 

Here's the transcript: http://www.euclideon.com/transcripts.html

Edited August 2, 2011 by jaxa

[demagogue]

Voxels are a virtualized data, so "point cloud data" doesn't even sound that much different in principle. I think they'd be good for the right kind of game, where the added atmosphere adds a lot to the game but you can get away with a hacky environment until they work out the kinks.

[nbohr1more]

By large data set I'm assuming you mean they take up loads of hard drive space, right? Funny how Carmak is already working on ID6 when he just finished ID5

 

Without something like Megatexture streaming for Voxels BOTH video ram usage and HDD usage would be enormous. Carmack had earlier remarked that unless something changes enough in the flexibility of graphic hardware that the Voxel portion of his engine would only be used for static background objects.

 

He has recently made a statement about how well his id Tech 6 stuff is working so... [speculation] he probably has some good next-gen AMD and nVidia GPU's to play with and they are probably powerful enough to move those voxels. [/speculation]

[Baddcog]

How does this fit into art asset creation (can't watch any vids right now)?

 

Will you be able to export standard formats/models... Or will this also require specialized programs?

[jaxa]

I was thinking about the imported rock. The way this thing imports objects into voxel / point cloud form, does it mean that textures as we know them are not used? AKA is it the "death" of textures?

Edited August 2, 2011 by jaxa

[nbohr1more]

The video seems to indicate that you can create standard models and textures then convert them to point-cloud data.

 

If we are looking at this from the practical perspective of id Tech 6 (which we know will be a real product) then textures are already gone to some degree... Megatexture is not really "textures as we know them"... though you can bake tiled textures into a Megatexture and that is still a practical thing to do unless you have billions of man-hours to hand-draw every nook and cranny. Voxel Octree should take that a step further and merge the voxels with the mega-texture data.

 

I can see, however, that modular "mini point clouds" that can be deployed via procedural methods (like SEED) might be more practical than a large monolithic render and the video shown here has a lot of asset re-use so that might be their method. If id doesn't offer smaller scale voxel-set management they may be at a disadvantage to engines that do.

[Serpentine]

Point clouds are always 'the future', as the guy explains - they've been used in scanning for years and are well understood, however the concept while rather powerful is one of those academic areas where try as people might like, there's very little traction in getting things that make games, closer. The points are pretty much like a vertex, if you cluster them close enough, or can represent them in a clever manner; once you get a dense enough collection you'll start to describe whatever shape you want. They also have advantages as you can attach color and various other abstract qualities, for example you could add a density to each point and use them to do dynamic physics, tempreature transfer from forces and such.

 

The problem however is that as said - they're academic and used widely in scanning... scanning... non dynamic, non animated, non abstract things. Now, when you think about animating polygons, even bones which are relatively 'new' HL1 era things, you can understand how it all fits together; your mental model of a stretchy bag with some rods and such deforming it - bam we're done. Now, how do you animate point clouds? well umm, yeah it's a bit like animating a bag of sand - plausible but much harder to think about. Notice how there's no animation in those videos? Just that lovely moving camera - it's a very tricky problem. The less procedural the system becomes, the harder it is to support, colouring the particles to form texture, lerping distances to fill gaps, it's all fairly deterministic. You also see an incredible amount of duplication, allowing lot of instancing to take place, this once again is fine for a tech demo, but offering 'infinite' details without 'infinite' variation is a bit self defeating.

 

Games also use a lot of constraints to fight open-world problems, the more freedom given in regards to rendering is all well and good, however you will find that the games will end up requiring more levels of abstraction to handle everything from AI to sound and animation, at least in the early days. These are the real deal breakers, and it's also the strength of voxels, which are a midpoint and far easier concept to play with in your head.

 

Polygon -> true voxels -> points, sure. But I cant see a fancy static dataset being proof that we can skip that middle-'point'.

[Komag]

The bandwidth/processing department could be off-loaded to a super-server, sort of like how OnLive gaming works (and it works pretty well). So just the image is transmitted to your computer, all the processing happens in the cloud

[jaxa]

Speaking of OnLive, Steve Perlman just released his white paper on DIDO (distributed input, output). Here's some juicy excerpts:

 

"When OnLive servers are co-located with DIDO servers, there is unprecedented low latency, to the point where remote-hosted OnLive gaming and computing is lower latency than the same games and applications running on a local device," Perlman wrote - in other words, DIDO is faster than a game running on a native console or PC. DIDO should have a latency of just a millisecond, he wrote.

 

"We believe DIDO wireless will completely transform the world," Perlman added.

 

...

 

This particular tactic has Perlman claiming that a single transmission can travel 250 miles, in part due to the fact that the DIDO access point transmits at a higher power because of the reduced risk of interference because of the individualized channels.

 

If clear, uninterrupted wireless signals weren't enough, Perlman has also stated that DIDO has a latency of less than 1 millisecond, in comparison to a couple of milliseconds with Wi-Fi and 150 milliseconds for 3G connections.

 

DIDO technology is a lot to swallow, if you feel like delving deeper into the inner workings, feel free to check out his white paper.

 

Personally I'm a lot more excited for DIDO than these point cloud graphics. That's an interesting idea there to combine the two though.

 

http://gear.ign.com/.../1184791p1.html

http://www.pcmag.com/article2/0,2817,2389374,00.asp

Edited August 2, 2011 by jaxa

[Serpentine]

Procedural World - Unlimited Detail

 

Pretty much covers it in simple terms and connects back to the voxels nicely. And yeah, if you havent had a look at the "Atomontage Engine" vids, there's one linked in there, and even avoiding that it's made by a single guy, it seems a far bit more worthy of attention.

[Serpentine]

Personally I'm a lot more excited for DIDO than these point cloud graphics. That's an interesting idea there to combine the two though.

I suggest reading the PDF. A bit like Fox News, except with a bit more denial and marketing bullshit.

[jaxa]

I suggest reading the PDF. A bit like Fox News, except with a bit more denial and marketing bullshit.

 

Yeah I suppose its just a longer press release, but I'm still not convinced that it's bullshit. It is not a technical specification and it alludes to the "complex mathematics" behind giving each user a "channel". I'd like to see them publish that paper. I'm also waiting for the inevitable patent troll lawsuits.

 

The paper also mentions a "Contour live character". Looks like a bullshit tie-in but I guess it can be counted as an actual example of what can be done with onlive/dido: centralize processing and deliver graphics with low latency.

 

We do not know the theoretical limitation to how many users we can add to a DIDO system without degradation in data rate per user.

 

Hmm that might be important.

Edited August 2, 2011 by jaxa

[jdude]

Procedural World - Unlimited Detail

 

Pretty much covers it in simple terms and connects back to the voxels nicely. And yeah, if you havent had a look at the "Atomontage Engine" vids, there's one linked in there, and even avoiding that it's made by a single guy, it seems a far bit more worthy of attention.

 

The guy says there's no real world application in gaming for this tech, but say you have something like grass or tons of plants, rocks, bricks, signs, lamps, leaves, etc which would be the same thing repeating over and over. You could technically combine polygon tech and this tech, allowing much more polygons to go into characters and animated objects while the static and repeating things could be voxel tech.

 

Also is this the same tech used in Delta Force 2? http://www.activewin.com/reviews/software/games/d/images/delta_force_2_3.jpg

[Serpentine]

We do not know the theoretical limitation to how many users we can add to a DIDO system without degradation in data rate per user.

 

It seems like they're baiting people to think that the problem, the catch will lie there. That stops people from thinking about the underlying logic of the whole thing, for example : I can share a limited resource equally, without degrading services to other people if I divide it into small enough fractions that its unlikely to ever require me allocating more than the max slices. You'll just always have a small slice. It also makes a sales point only in the place where there is radio contention, which for me, in a reasonably well populated area is... wow, pretty much nothing at all on the contended wifi bands (and even if it were, would it matter?). His testing required large antennas and cabling? Umm ok right. But I guess when you notice that he's advertising the speed of an em signal as the defining factor (cos we've never used them before, right? I mean nothing else could possibly have the same speed? oh) - well this all makes a little bit more sense.

 

Also is this the same tech used in Delta Force 2?

If you're talking about voxels : Yeah the early DF and Comanche games (iirc) used voxels quite a bit, as did Outcast - which was rather nice looking.

[nbohr1more]

Yep.

 

id Tech 6 will use a hybrid approach.

 

If things go as poorly for John Carmack with id Tech 6 as with id Tech 4 it'll be like:

 

1) Yay a working demo!

2) Crap the drivers are as unstable as hell

3) Revise the render method for stability

4) The new stable version is SLOW

5) Waits for eternity for drivers to improve

6) Drivers never improve or barely improve

7) Release SLOW version since Moore's Law has now caught up

8) Brief moment in the spotlight (oohs, and ahhs)

9) Overshadowed by a better looking game on an "old fashioned" engine with polygons and tricks like tessellation and instancing and good art...

 

OpenGL is back to having a good API for now so the above sequence is unlikely. You can never count on the future though...

[demagogue]

That ProcWorld blog is fun to read.

 

The concept of an atom-based next-gen Minecraft-styled game is going way beyond what my little imagination would have thought of, looking at how he's generating terrain and realistic looking cities on them... If you could actually deform the atoms in-game & the geometry would respond properly, so you could add & subtract atoms in set clusters like materials (bricks, wood planks, gravel, stones, etc, then maybe "pouring" "liquid" masses like sand and water), then holy shit. It would be a full on virtual planet... People toss around the term groundbreaking lightly these days, but I think that kind of sandbox world would really make a big splash and change how people can really "be in" a gaming world.

[Serpentine]

That ProcWorld blog is fun to read.

Yeah, he's rather damn good at being well versed and his approach to things, looking to not just stick purely to one strength but to accept that a hybrid of voxels and traditional polygonal methods is nice to see. Nothing is perfect in it, but he hits so many 80%'s that its really interesting to watch over time. There's no realtime physics or anything, but take a look at the Atomontage Engine, which is a more voxel/game specific development, also made by a single (or two?) guys; The physics videos are very interesting to watch, with some polish it'll be interesting to see what can be made.

[Mortem Desino]

I find it hard to think that voxel/point cloud twill be useful for anything but visual rendering yet. The best bet so far is to combine technologies. I could think of a system of point-cloud data for rendering, and seperate polygonal collision models (physics interactions), but that doesn't solve the animation problem.

 

The only extensive work I've seen done almost completely in voxels is Voxelstein 3D. They're pretty dang inefficient, but static objects (like those seen in the video) can be made to look very believable. Until genius engineers flesh this out, I would only use voxels for destructible terrain & destructible static objects.

[STiFU]

I could think of a system of point-cloud data for rendering, and seperate polygonal collision models (physics interactions), but that doesn't solve the animation problem.

I guess you could attach a set of points to a polygonal bounding box, you can animate the bounding boxes as we already do today and use some soft selection algorithm in the neighbourhood of bone joints to animate the whole point-cloud.

 

But still, I guess it would be a lot more efficient to extend the Concept of parallax occlusion mapping to "real 3d textures", so that you won't get the stretched-texture effect when you look at a surface from a very steep angle.

[stumpy]

you wouldn't need a large data set, if they have a program for turning polygon based models into voxels based models, then the model could exist as a polygon model in memory and and the side visible to the viewer would then be changed to voxels on the fly. So the voxels only exist when being displayed, in memory they're still a polygon model. And therefore to animate it you would animate the polygon model but view it as a voxel model in the display.