Skip to main content
GameDev.net gamedev.net
🔒 Locked

Voxel Challenge...

Started by Zeidolon Jun 13, 2007 at 10:41 PM 8 replies 4.3k views
Original Post
Zeidolon
Zeidolon
[Apologies if this post isn't appropriate for this category] Is anyone working on any active voxel research, simulations, or game projects? I have recently had some strong debates with several friends and colleagues about voxels and I seem to be the only one that thinks this technology still has promise. No, I don't mean voxel based heighfields for landscapes...I mean full 6dof with destructable environments. Unfortunately, the only examples and demos I can find are rather dated and still look blocky or feature "programmer art". It is hard making an argument when you have to compare Outcast, or even Ken Silverman's Voxlap (still the most impressive stuff I can find so far) to any polygon based game or engine. Polys have the unfair advantage of almost a decade of advancement and of course the proliferation of hardware accelerator cards. I am hoping to find some smart engineers that would be interested in talking about voxels and exploring some ideas. Even better, I'd love to see a tech demo or proof of concept to make my point. Who knows, if you can blow my socks off, there might be a job offer down the road. Then again, maybe voxels are dead and buried and there is no room for any innovation or startling new advancements. If you think I'm wasting my time, I'd like to hear that too (as well as some good technical reasons why). Thank you for taking the time to read this and reply. Robert
Adam Luchjenbroers
Adam Luchjenbroers
Sounds interesting, can't say I've had much experience with Voxels but it can't hurt to evaluate the technique.

With newer graphics cards heading towards programmable pipelines, it might even be possible to do the Voxel processing on the card. Maybe look at what the General Purpose GPU (http://www.gpgpu.org/) guys have come up with?

Zeidolon
Zeidolon
Thank you Adam, that is a fantastic link, I wasn't aware of them before.
stephanh
stephanh
According to the gdc07 paper, Nvidia let the cubes march on the geometry shader in their cascades demo.

http://developer.download.nvidia.com/presentations/2007/gdc/CascadesDemoSecrets.zip


And well, even for terrain voxels are still usefull, just remember how everybody was so excited about cryteks sandbox demo showing caves and overhangs. So at least for editing voxels are still alive in terrain rendering.
Zeidolon
Zeidolon
Thanks Stephanh, I missed GDC this year.
Abominacion
Abominacion
It seems that at least NVIDIA is still quite fond of voxels. Besides the already mentioned procedural voxel mesh used in the Cascades demo, they also released at GDC07 a fairly insane paper where they voxelize meshes on the fly into a volume texture so they can collide against smoke in a GPU fluid simulation.
YSlang
YSlang
Last year, I implemented a simple SM3.0-GPU-based 6DOF voxel raycaster which samples at grid intersections. (Although it casts a ray per pixel, not per column like Ken Silverman's Voxlap. And I tried to do a vertical-column voxel raycaster but SM3.0 shaders were too inflexible to do that.)



Real-time footage video:
http://www.quakewars.jp/slang/files/videos/flx_vrc_01.m1v

Then I soon realized that SM3.0 GPUs were totally unpractical for voxel raycasting and further voxel-based applications. On GeForce 6800 GT, the raycaster performed 20fps at a minimum with only 320x240 resolution (I employed 4x4x4 uniform blocks for empty skipping). This could be attributed to the VRAM latencies by a lot of 3D texture fetches in the fragment shader and the NV40's low dynamic-branching performance due to its high granularity.

So far, I have not tried to do a vertical-column voxel raycaster using SM4.0 shaders or CUDA (well, I don't have a GeForce 8800 card [wink]), but I think that Intel's upcoming homogeneous many-core processors would be more suitable for voxel applications both hardware-wise and market-wise.

[Edited by - YSlang on June 16, 2007 4:31:34 AM]
taby
taby
Quote:
Original post by stephanh
According to the gdc07 paper, Nvidia let the cubes march on the geometry shader in their cascades demo.

http://developer.download.nvidia.com/presentations/2007/gdc/CascadesDemoSecrets.zip


And well, even for terrain voxels are still usefull, just remember how everybody was so excited about cryteks sandbox demo showing caves and overhangs. So at least for editing voxels are still alive in terrain rendering.


My sentiments exactly.

Fluid dynamics and General Relativity both use fields for their data. There's no way that this type of visualization system is going away anytime soon.
Zeidolon
Zeidolon
Thanks guys, this is some great info you are providing and linking me to.

It is beginning to seem that there is a fair amount going on, at least using the voxel concepts to other areas, particularly for volumetric effects.

Honestly though, I'd like to see what something like Outcast (image below) would look like if the engine and technology was updated from 1999 desktop standards to 2007.



I would also like to see a real-time version of Tom Dobrowolski's Cave Demo based on Voxlap. I get the shivers everytime I watch that demo, even if it was rendered at .001 fps (a few years ago)

Some pics:









See what I mean? I know we can do a lot of awesome stuff with polygons, but I cant shake the notion that we are missing out on a lot of neat mechanics and options that would open up for gameplay if a few companies focused on voxels for a few years and advanced some of the tech.

Again, if anyone has the time and inclination to experiment and try some stuff, please contact me.

Keep the links and info coming please, this is vastly appreciated and informative.

Robert

[Edited by - Zeidolon on June 14, 2007 12:23:54 PM]
PolyVox
PolyVox
Hi,

Plese take a look at my project here:

http://www.david-williams.info/index.php?option=com_content&task=view&id=24&Itemid=37

And also here:

http://www.ogre3d.org/phpBB2/viewtopic.php?t=27394

It's implemented as an extension to the Ogre3D graphics rendering engine. My approach uses a volumetric representation of the world and then uses the marching cubes algorithm to generate the polygon mesh so that it can be rendered quickly.

It's in very early stages but let me know what you think :-D And if there might be a job done the road then I'm definatly interested... I believe this technology has a lot of potential. Especially when you start considering physiscs, which I believe has a lot of advantages in a voxel environment because the collision detection becomes much easier.

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.

Sign in to reply to this topic.