Game Development Stack Exchange is a question and answer site for professional and independent game developers. It only takes a minute to sign up.
Sign up to join this community
I stumbled across the following article on Wikipedia regarding voxels that says:
One such problem cited by Carmack is the lack of graphics cards designed specifically for such rendering requiring them to be software rendered, which still remains an issue with the technology to this day.
Does anyone know why this is the case and why voxels are not supported or optimized on a hardware level by graphics cards?
Primarily because, as of this writing, hardware vendors likely don't have a reason to implement such a thing. Mainly because there isn't any benefit to doing so.
The majority of games and 3D software applications are based heavily around the traditional scan-line rasterization model, and would see no benefit (and in fact, likely significant drawbacks) to adopting a voxel-based model.
Further, there are no emerging standards for software-side APIs for voxel rendering that are backed by significant industry powerhouses (like already exist with D3D and OpenGL).
With ATI starting to explore the direction of moving away from industry-standard APIs towards something proprietary (Mantle), and the possibility of VR headset technology actually becoming a realizable commodity technology via Oculus Rift, there is a slightly greater possibility of seeing something like this in the next few decades.
But up until now, and likely for quite some time yet, the cost/benefit ratio for all parties concerned is just not favorable.
It's a chicken/egg problem. There are no serious voxel-based graphic engines, so there is no need to develop a voxel-optimized GPU, and as long as there are no voxel-optimized GPUs, nobody will create voxel-based engines.
However, the quote by Carmack was specific to Quake III which is 15 years old by now. While it is true that there is still no standardized abstraction layer for voxel-based graphics which could be compared to OpenGL or Direct3d, GPUs have become a lot more flexible in the past years. Today, a hardware-accelerated compute-shader can be useful to speed up various voxel-operations.
One might wonder why we still see so few serious voxel-engines outside of the demoscene then. The reason is that while the development of voxel-based rendering essentially stopped in the 90s, polygon-based graphics made a trendemous development since then. The voxel technology first needs to get over this huge research gap before it can again compete visually.
There's a company that claims to be seriously developing a point-data engine, and that it somehow gets by on software rendering. They had previously stated that graphics hardware developers were uninterested in non-polygon methods of rendering, so they've more recently begun pursuing geospatial applications.
This technology description gives no technical details, though various pitch videos claim that they "only load the point data required for the render", "can work with a total pool including terabytes of data", and "have overcome the problem of frequent, tiny disk-reads".
That last claim is the most suspect to me. If they've overcome file I/O latency, that is a bigger accomplishment than their point-data render. Their demo may be a complete fabrication, but they seem to be very persistent if only perpetrating a hoax. On the other hand, they seem to be rendering not "unlimited data", but rather "unlimited instances of flat-shaded algorithmic geometry".
Interesting, nonetheless. https://www.youtube.com/channel/UCI9bdH0LDRCVxeeAFC1Sk3w
