SIGGRAPH 2003: Day one


Day one of SIGGRAPH is on a Sunday this year, allowing for the travel benefits of staying over a Saturday night without having to find something to do that isn't related to the conference on Sunday.

However, none of the papers start until Monday, so you will only miss the courses if you skip- out on Sunday for travel purposes or because you want to go play at the San Diego Wild Animal Park.

So far, the most interesting piece has been a GPU ray tracer from the guys at the U of I (later overcome by a Stanford implementation). The idea is to use the SIMD capabilities of the GPU to perform the basic ray tracing and triangle intersection, which is then later fed back into the renderer.

The afternoon included presentations from ATI, nVidia, and 3DLabs about their existing implementations. They were, as one would expect, advertisements for their implementations with good demos. However, the guy from nVidia was probably more on message than the others were. He had two important things to say:

  • nVidia will continue to fully support OpenGL and DirectX, including OpenGL Shading Language and HLSL
  • Cg will be about to output OGLSL at some time in the near future, so you will be able to use Cg to create optimized NV shaders under GL and DX as well as compatible shaders using OGLSL and native GL commands.

After the advertising portion, an ex-nVidia, now-UT Austin researcher gave a good presentation on the GPU vs. CPU vs. Stream CPUs (vector processors). After the general differences, the talk turned to potential future of GPUs and models such as MIMD and multiple-core designs that may be the future of GPUs.

Next up was an interesting pair of discussions of layering shading tools on top of the existing programmable shaders that will allow the creation of shaders that are too complex to be executed in today's pipelines. First, an MIT grad student discussed RDS, a shader compiler that automatically splits shading into multiple passes and combines them appropriately to achieve complex shaders that execute quickly. Then, a member of the ATI technical staff provided information about ASHLI, which is a high level texture programming environment which provides segmented shading that similarly gets around the problems endemic to the relatively small execution environments of today's graphics cards.

If I can stay awake, I'll be hanging around for the big paper pitch, where each panel has 60 seconds to pitch why you should come to their presentation.