pc999 said:
Can anyone explain ( in a "easy" way) to me what are this techniques
I'll give it a shot.
The new 3DMark engine is a next generation shader engine, with a wide variety of shaders dynamically built and runtime compiled
One reason there's been controversies over 3dmark in the past is that the shaders that were written for it ran sub-optimally on some pieces of hardware due to differences in basic GPU design between different vendors' products. (Comparison: compiling Half-Life without 3dnow support made it run crappy on the AMD K6-2, while it ran well on Intel processors because they had better FPUs).
Instead, by writing shaders in a high-level language, and then have each driver for the different GPUs will allow the driver to tailor the shader as best it can after the actual hardware design; IE, it would be like getting Half-Life to compile at run-time WITH 3dnow support when it's started on a K6-2 system. The output would be the same in all cases - the game being played is Half-Life, but shaders on different hardware would contain different instructions to best take advantage of THAT particular platform's strength.
I've seen demos of terrain and worlds, with no textures in them whatsoever and no geometry - it's just a program that's creating a scene for you,"
This might be a time- and money-saving technique IN A WAY, but not neccessarily. What it really does allow however is greater variety.
For example, instead of having an artist painstakingly build a tree out of polygons, you write a piece of software that runs a (fractal) algorithm that builds the tree for you. By tweaking certain parameters, like how far apart branches should be on the trunk, wether branches sprout out in concentric rings or not, the angle they point outwards from the trunk etc etc, different trees can be created, and they'd all look different, even trees of the same species. Generating a tree with millions of polygons in it would only require a few dozen bytes of data to describe the tree to the algorithmic generator, but storage space would of course be traded for computations when the tree has to be built in order to be displayed on-screen. This would likely happen during level load.
The same basic technique can be applied to textures; a general fractal algorithm that when parameters are set a certain way creates patterns that looks like stone or wood grain or brushed metal or skin/leather/scales or whatever. The program could even generate bump/parallax maps to create fake 3D depth at the same time, but lots of these maps at high resolutions would require lots of computations. They would also need to be compressed afterwards, and doing automatic realtime texture compression and get good results is also computation-heavy.
So in part it would perhaps reduce workload on artists in some regards, but only for certain features that can be generated algorithmically - mostly plantlife really, and also larger geographical features such as hills, valleys, mountain ranges and such, though I feel level designers would rather retain control over this themselves than trust it upon some random generator.
. Also, the tools would have to be made first (and that is probably not a small task), and they have to be used correctly too. And houses and man-made objects, creatures etc still have to be made by hand...
