Or how about 50x? Brent Oster's NVISION08 CUDA talk is now available at http://developer.nvidia.com/object/nvision08-advanced-cuda.html -- in it he describes the methods of using CUDA optimally, the new Tesla 10 architecture (hint: when viewing on the web, press the "fullscreen" button to see all the fine details in the diagrams), CUDA and OpenGL interoperability, optimal patterns of memory access, and introduces some staggering examples, such as a 1-million object interactive particle system interacting with 100 collision spheres in only 20ms....
Posted on 09/18/2008
