Sunday, April 11, 2010

Skipping Moores Law with unlimited detail

Moore's law predicts that the number of transistors per CPU chip will increase with resulting processing power doubling every two years or so. This process is driven by game software. There's no good reason for a computer doing accounting, administration or email to go so fast. Web browsing and other fast software has been riding the wave but its gaming that pushes everything along. 3d Graphics is the great challenge of computer game, graphics lag the biggest effect of any failure to meet that challenge.

Now a couple of very smart software writers in Brisbane Australia have come up with a piece of software that side steps most of the problems giving unlimited detail.

Its a compression process and search based decompression process that puts pixels on the screen without creating in memory somewhere the whole virtual world including all the unseen bits. It works with pure point cloud data and not polygons. This means that polygon counts are irrelevant. The graphics are pure points of colour in memory.

Some have said that unlimited detail is impossible because it would require unlimited random access memory. I'm always amazed at the ability of the media to find some academic that is eager to say If I can't figure out how to do it nobody can. It rebinds me of the statement."Those who can, do; those who can't, teach."

I think they have something very powerful. By its self it will allow even the most basic game to surpass the high resolution images seen in movies and ads.
However it may not end there if their software works with any point cloud data input then not only can the laser scan objects into games but they could use ladar and bifocal cameras to map in landscapes, trees, whole forests. Live animals, people and buildings.
That's because a ladar produces point cloud data and the bifocal camera can use an Boolean AND algorithm that creates point cloud data from a scene. Any point that has the same colour at a given focal length from both cameras defines a point. If it remains as the camera orbits the target a point is archived. Its a little more complex but not much. Point cloud data and voxels are the same thing. Voxel means a volumetric pixel. 3D pixel.
We may see within a year or two photo-realistic game worlds. That wont eliminate Nvidia and ATI though some may fear that. Once the polygon problem is solved the next great challenge is game AI and more realistic animations. We will want our virtual worlds to move right not just look right. While Unlimited Detail Technology can do awesome animation I expect such motions must be integrated to other parts of the game physics, lighting etc.

The other challenge is creating fast low memory point cloud editing systems that creates convincing objects quickly and easily. Such editors exist and are powerful but its easy to get things wrong. If Unlimited Detail Technology is adopted by a object editor then even the editing of the objects could be done on smaller computers. Also don't forget that it will very quickly go 3d on the screen.

Another break through may follow. Very fast Voxel physics. If they can use a search algorithm to find the colour for each screen pixel then a second 'camera' may be used to look at the data in another way. What's the closest voxel doing? Is it solid, stationary, moving, fluid, flexible, gas? Which are my bounding box, avatar, clothing. Is it the ground. A voxel that can move could be water, fog, rain, bullets, magic fire balls, or just grass crushing under foot. By adding a byte to the points in the cloud you could add 255 options.

The world of computer gaming is about to spike. Hang on to your hats it will be a wild ride.

No comments: