Slashdot Mirror
Refresh your Memory: Advanced Graphics Algorithms
subtle writes "DevMaster.net has posted an interesting article about advanced graphics algorithms. The article discusses six widely used algorithms in graphics rendering of indoor and outdoor environments, namely: quad-based static terrain, Roettger's approach to continuous levels-of-detail in terrain, real-time optimally adapting meshes, portals, BSPs and PVSs. In each case the algorithm is discussed and some aspects of implementation are considered, as well as analyize each algorithm for its application in modern graphics systems."
I hate when something is called a refresher course when it's something I never learned to begin with...
Matt Fahrenbacher
James Tiberius Kirk: "Spock, the women on your planet are logical. No other planet in the galaxy can make that claim."
(I am a Game Developer) and I'm trying to work out why this article was posted. Its too advanced for beginners, its not detailed enough for professionals. Its basically a list of the names & very basic approaches of a few graphics algorithms. I suppose people vaguely interested who know the basics but haven't tried some of these out are the target audience.
Anyone there fit the bill? Did you like this article? Was it helpful and informative?
Game dev and music blog
The type of problems solved by Octrees are also solved by BSP algorithms. So to put both in the article would have been a little redundant.
Tubal-Cain smokes the white owl.
The content of the article aside, its "smart" format is appealing on several levels.
Among other things, it is a format that is immediately usable. That isn't to say that internet-magazine articles are generally difficult to read, but some are easier than others due to advertising, having the article split into several pages, etc.
Also, this article is a more permanent statement: it can be referred to in the future, all at once, easily. There really is no reference value to most internet-magazine articles - they are so cumbersome and ephemeral that they quickly become part of the "forgotten internet."
The archival policy of the hosting site aside, this article could, by virtue of the atomiticity of the information, have greater staying power than a traditional, graphically infused pop-article. It can be transferred to other locations trivially, and its potential use in more traditional contexts figures in favorably.
Whether or not the content is of value is, again, a different question entirely.
?/o
It's easy to tell when slashdotters know nothing about a particular subject.
We get to read lame joke after lame joke.
This article is like the 10,000,000-foot view of these things.
For the socially retarded among you, the amusing part of that is that anyone actually cares enough to bother!
All current graphics accelerators that I am aware of don't do voxels. They start with polygons, apply transforms to those, apply textures, lighting and such, and then rasterize that to pixels. Well since they don't support voxels, a voxel based game would probably just suck. Graphics cards way outstrip CPUs in terms of crunching power because they are specialised for what they do. However that means that if you don't do things the way they want, they aren't useful and you are back to pure CPU processing.
No one is going to be able to market a blocky looking voxel game next to stuff like FarCry and Half Life 2. Even if the voxel engine is "technically" superior, it's not going to matter because lacking hardware acceleration, the end result will be far less impressive.
Lame joke after lame joke, modded higher than you think they should be. Welcome to Slashdot, newbie!
Don't blame Durga. I voted for Centauri.
Today, the issue is significantly different. Consider an old 'gigapixel' card. 1 billion pixels per second at 60 frames per second == 16.6 million pixels per frame. Even at 1600x1280 there's only about 2 million pixels onscreen at once. What does this tell you? You can now afford to sink some time into overdraw because it just doesn't matter.
Also, consider the fact that 1600x1280 == 2.04 million pixels onscreen. Even if you draw a single polygon per pixel @ 60 hz, you're still only looking at 120 million polys/second. That's going to be a pretty reasonable number for next-gen hardware (Xbox2). What does this say? Well, unless you really need 1 poly per pixel, you can afford to draw some extra polygons.
Now, let's say you've got a scene with 10,000 discrete objects. This is a pretty reasonable number these days. Even on a 3ghz processor, doing distance-to-plane checks for plain jane frustum culling is pretty darn expensive when done 10,000x per frame. Multiply that number by some constant C to do N world space occlusion checks. All of a sudden you're sinking multiple milliseconds into just doing scene graph traversal. Don't forget, you only get 16ms per frame @ 60hz. PVS on triangle strips or individual triangles? Teehee - you'll spend forever trying to process all that crap. Throw a simply quadtree or octree onto a scene, and cull at the object level (where object == say, 1000 to 10,000 polys). Let the video card deal with the little bit of extra overdraw and wasted polys. Point is, -you- have saved, say, 4+ milliseconds of raw CPU time. 4ms is a lot. Send that extra time to Havok or Karma. Send it to a fancy effects system. Don't waste it doing needless old-school culling. Don't forget, if you're doing things right, the 'rendering' is really just DMA slinging verts and textures in the background to the GPU. Paralellism with the GPU. These mega hoopdeedoo X800+ cards can deal with a little excess load.
The point is, these algorithms are most decidedly not advanced. They're from 2-3 years ago! Quite literally, that's ancient history in the games world. Ridiculous, but true. The real brilliance of these new generation GPU's is not wacky implementation of bizarre obscure culling algorithms. The brilliance is that they are so powerful, they allow you to spend your programming time implementing beautiful shaders and effects. In 1-2 years, graphics programming will have truly morphed from glorified bookkeeping (managing and organizing data has been the hallmark of the 'hardcore' graphics programmer for several years now) into actual effects and shader programming.
I cannot wait until dicking with exporters and preprocessors, and goofy custom renderers are a thing of the past. With some clever planning, graphics peeps will be able to really sink more and more time into making things beautiful, instead of being forced to be excessively clever to make things happen. There will always be room for 'real' advanced stuff (like modern GPU + CPU shadow techniques, spherical-harmonic lighting, and other esoterics), but the power of these new cards lies in freeing up graphics programmers to actually write graphics code instead of being accountants. That's the real practical results of this next generation of hardware :)