Notgames Forum

http://www.youtube.com/watch?v=00gAbgBu8R4 (http://www.youtube.com/watch?v=00gAbgBu8R4)

I want this and I want it NOW!

Makes me think of this guy:

http://www.youtube.com/watch?v=_CCZIBDt1uM (http://www.youtube.com/watch?v=_CCZIBDt1uM)

Voxels seem like a very interesting technology. Who know what we´ll see next?

I remember this, exactly like the narrator described at the beginning - a fantastic claim that was followed up by nothing.  Glad to see they are still working, though.  I will remain skeptical until I see it working in a real project.  Voxels have always been way to heavy to run in real time, so how they are giving you an "unlimited" number seems almost like magic.  I hope they can back it up, because modeling in voxels would eliminate some of the irritations you get with polygons.

This is nothing new.

http://www.tml.tkk.fi/~samuli/publications/laine2010tr1_paper.pdf

Realtime 3D will always be an illusion. I personally don't mind the artifacts of lower-poly models too much. As long as the overall effect is good. Which has more to do with art direction than technology.

What I would really be interested in is a new way of creating these 3D models. And this technology doesn't seem to have a solution for this. The movie just talks about exporting from stone-age Max or scanning things in.

I'm also wondering if their claims extend to the inside of objects. I've always felt the hollow-ness of 3D models to be very unnatural. I would love to be able to create models that are filled with atoms and that are not just shells. This would feel a lot more intuitive when creating them.

I do like how this technology might take away some of the head-aches when running up against polygon count restrictions. But given that we already have trouble creating a smooth-running game with small amounts of low-poly objects, I'll join the sceptics here and wait and see.

It's not all that complicated. A demo ( including source ) from NVIDIA is available here (http://code.google.com/p/efficient-sparse-voxel-octrees/), or you can simply check out a video (http://www.youtube.com/watch?v=Mi-mNGz0YMk) of the demo.

There are some serious downsides to this approach though.


Agreed

I don't mind low-poly models either, in some cases. But I think it would be awesome to take a super detailed model from Zbrush or Sculptris and put it directly into a game and not have to screw around with normal maps, modelling a low-poly version and optimizing textures and all the other bullshit you have to do now.

I agree. It would be great if this technology could take away some of the hurdles towards being creative in this medium. I'm pretty sure I spend more than half of my time on finding workarounds for performance limitations. It would be nice to be able to be a bit more productive.

But I'm afraid that now that the engineers have figured out how to make games run at 60 FPS on a normal computer (which means that our games run at 30 -still decent), this new technology might mean that we're back to 30 FPS for engineers (meaning 15 for us -unacceptable).

I use 3D Coat, which allows you to sculpt in voxels. That means objects actually have volume. It's quite nice to work in, but you can still mess things up; this is sort of the first commercial software of its kind. :P

http://www.3d-coat.com/

One day we'll be using this, I think that's a given. However, I don't think these guys (who have been around a while) are going to give us a magical breakthrough that will allow us to utilise this on near-future systems. Hardware needs to move on too much for this to become a reality soon, even with clever software technology. Indeed, hardware needs to be developed around this and that doesn't look like a route that the big hardware players are aggressivley chasing.

The possible physics application of this technology gets me more excited than anything. Saying that, it's also another major drawback when competing with current tech.

The strong push towards mobile computing is somewhat of a setback for increases in hardware performance. These days technology seems to evolve by taking steps back before going forward (a little bit). Maybe we're too nostalgic. Or maybe we're not ambitious enough. Or maybe we don't really want all this stuff.

The engineer in me screams when anyone uses the word 'unlimited'.

HardOCP did an interview and shows a real-time demo: http://www.youtube.com/watch?v=JVB1ayT6Fdc

I would love to see this used to bring natural complexity (plants, dirt, etc.) into game environments... It's something I appreciate so much in the real world, which games currently cannot even come close to.

Everything they claim in the first video (and in the second, for that matter), is pure unsubstiated bullshit. What they claim to have done is absolutely impossible. What they have done is nothing new. It's perhaps more efficient than previous voxel-codes, but no revolutionary breakthrough.

What they actually show you in the demo is a very small amount of voxels (compared to how many they say there are) in chunks repeated all over the place. Notice how all of the world is flat and structured in the same way, constantly repeated? Notice how all big structures face the same direction? They're just the same chunks repeated.

What they've done is very impressive but they keep lying about it for reasons I can't exactly understand. If you've got the funding, if this isn't a scam, why not be honest about it?

(Always beware of claims of anything unlimited, they're always lies.)

The reason to be dishonest is if you're looking to create hype to get bought out, which is most likely what those funding would want to happen.

I don't care if it new or revolutionary. It looks amazing. Also, he said quite plainly, it's not voxels. I think you're being overly cynical.

And as he also said: it is voxels. ;)

I'm not being overly cynical at all. What they CLAIM to have done is something UNLIMITED. No one can ever do anything unlimited. It's not unlimited and it never will be. Not unless they break the laws of physics, it's not.

You should probably actually take and read Notch's analysis: http://notch.tumblr.com/post/8386977075/its-a-scam . It's the one mentioned in the video but never actually adressed (he also goes on to completely misrepresent what Notch said).

What this system might be good for is video games that take place in a small enclosed environment (like a couple of rooms or something) or environments that constantly repeat. It's not going to work for open-world environments or any game that includes a lot of non-repetitive enviroments, e.g. terrain.

Why don't we just wait and see and hope for the best? I think we're all experienced enough with computer hypes that we know that things don't exist before they exist. And even then it takes some times for the likes of us to get their hands on them.

You're right, they should be saying "virtually unlimited". Even when you use 64-bit indices in your look-up code, there's still a range .. even though it's huge.


Notch calculates the data under the assumption it's stored as a constant volume, while all you really need is the hull of geometry ( skipping anything inside & outside ). Something that isn't rectified in his http://notch.tumblr.com/post/8423008802/but-notch-its-not-a-scam post either.

I do agree their "marketing" is awkward at best.

It's not nearly "virtually unlimited" either, though. Virtually unlimited would mean that it would suit all regular needs without you having to compromise (you could of course not make an infinitely huge world, but you could make a regular sized game world as they are now without problems). It won't do that; to achieve good results you're going to either have to limit the play area (or at least the number of unique objects) significantly or combine it with other technologies to achieve certain things. E,g. it won't be able to do terrain with smooth height-curves (since you can't store that much unique data); however, you could use it almost as you use a texture now by applying it to a (polygon) heightmap and let the polygons do the curves whereas the point-cloud data stands for grass and texture of it. You could use it like "tiles" (like they've done themselves, though without a heightmap like I'm talking about). Granted, I'm not that great of a programmer so I don't know if there are problems with this that I haven't thought of but it seems like it could work to me.

So let's assume it's just 1 mm of ground (less than they showed in the video) and no other objects exist. Just one mm of plain ground (a very moderate guess considering we are shown ground with geometry more complex than that). That's still 64 000 000 000 000 atoms. One byte per atom (which is very low for voxels, let alone impossible for point-cloud data) means 58,2 TB of data. And those are very generous approximations for the ground alone (so, no grass, rocks, etc.).


Now, what they have done (as far as I can see) is amazing. That's not the point I'm making. The problem I have is that I know they're lying. If they simply would have told the truth I'd be amazed, but now I don't know what other statements I can trust and what statements I can't. (Can I trust that it's now running on software only and not GPU considering I know they're dishonest about other stuff?)

The "unlimited" aspect they refer to is that their engine can render geometry at whatever level-of-detail without significant performance differences. Whereas with polygon based renderers the polycount has a direct influence on performance regardless of the area it occupies on the screen.


I doubt they store their data at a constant resolution ( at least in compressed form ). Regions of molecules that have the same content ( air, soil etc. ) don't have to be stored individually. A 2D representation ..

(http://me-lrt.de/img/num-406-quadtree-octree-netzverfeinerung-kante.png)


Current GPUs aren't of much use for what they are doing actually.

64 atoms per cubic millimeter = 4x4x4. So if the ground is merely one atom thick (which we've seen it isn't*), that's still 54/4 = 13.5 TB.

* In case it is, then they lied about it when comparing it to bumpmaps.

You mean 15.3 MB ;)

Regardless, this way of calculating the "filesize" is irrelevant.

No? 10^12 * 4^2 / 1024^4 = ~14,55 TB (yeah, different from 13.5 but that's because 54TB was rounded)


Why?