next »

[raquo]

A bit new here, let me know if I'm not in the right place.

Has anyone seen any 3D games that are voxel-based but not cube-based? i.e. there's some unit of destructing/building that fits nicely together to form surfaces like the ground and built into structures like houses but is triangular/pentagonal/whatever and not just cubes.

Voxels are just so hip right now but I love the concept for its simplicity in design and simulation. I'm curious if anyone has gone in a slightly different direction to stand out a bit against the crowd.

[Schrompf]

You can use surface extraction algorithms such as Marching Cubes or Dual Contouring. A million games is using one of these technologies, Rust for example.

You can control the smoothness of the surface in any of these algorithms to get specific parts to be hard-egded. The first game the comes to my mind is http://stonequest.de/ but there sure are many others doing this.

And finally: my own take on voxels. We aimed for an old school voxel look like Commanche, Outcast and such, so I went for splatting literally millions of billboards.



Still very early. And I'm not sure if it will ever fly, because we're now at 20 trillion voxels. Even the simpliest task gets difficult when handling such amounts of data.

[Sik]

Those billboards look like circles, right? Maybe rendering quads and using a shader to give them a circle shape (instead of textures) could be better, not accessing VRAM may end up improving performance a lot (would need to actually test it, but could be worth a try).

But yeah, the data seems really huge to handle even remotely sanely Looks good though, maybe coming up with compression-like methods (clustering?) could help.

[Schrompf]

It is already compressed with various methods. Otherwise it wouldn't fit a hard disc, wouldn't it? Let alone the main memory. Depending on the amount of detail involved the 20 tera voxels fit into 3GB of RAM. Generating and displaying the data is still slow, but reliable. Transmitting changes over the network is my current challenge.

Haven't tried to generate a circle in the pixel shader, that's a nice idea. But it would just exchange one texture sample for a bit of math. And latest profiling suggested we spend 80% of the time in the vertex shader, so I guess it won't yield anything.

[Sik]

Yeah, should have guessed *facepalm* Was thinking more about it for improving performance (compression could let you skip most voxels when handling them).

In the worst case you could use it for demoscene =P Looks like a painting, seriously.

[gimymblert]

Yeah actually having random (brush) shape per material instead of circle would go a long way to explode the rendering visual impact

[Schrompf]

We experimented with various shapes - rectangles, for example, work great for brick walls. We also tried real textures to add small details, but it proved to be difficult to LOD this smoothly. I also tried to overlay a large modulating texture to break up the repetitions, but it's deactivated in the image for the LOD again.

[Sik]

Eh, honestly I prefer the circles, if you ask me.