3 ms·

Yeah, a lot seems kind of weird. Many things get mentioned in odd places or credit seems somewhat misplaced. * It mentions James Kajiya, but David Immel usual

by a_e_k 4y ago

Yeah, a lot seems kind of weird. Many things get mentioned in odd places or credit seems somewhat misplaced. * It mentions James Kajiya, but David Immel usually also gets co-credited for the rendering equation. (He's credited in the glossary, but not in the main section on global illumination!) * Particle systems mentioned elsewhere, but not in the ILM (Lucasfilm) section. * "1968 also saw the arrival of Ray Tracing developed by Bell Labs and Cornell University." Here, it seems to be confusing Appel, who published on ray tracing as a visibility algorithm in 1968, with Whitted who published on recursive ray tracing as a shading algorithm in 1980 while at Bell Labs. One needs to be really careful about conflating ray tracing as a visibility algorithm vs as a shading algorithm. * In the timeline: 1987, "Willow (Lucasfilm) popularizes morphing" vs. 1988, "Willow (Lucasfilm) uses morphing in a feature film". * "One approach to taking more responsibility from the CPU was done by the GeForce 256 from NVIDIA (the first graphics processing unit, or GPU)." Next paragraph: "3dfx introduced their GPU, the Voodoo in 1996." Umm, the GeForce 256 came out in 1999; 3dfx was first here, though I believe that NVIDIA was first to coin the term GPU as a marketing term. And of course there were things like the SGI Onyx with its 3D accelerators before either (and SGI was discussed in the previous section). * Shading languages? BRDFs? Shadow mapping? Free form deformation, skinning, and non-rigid animation? Dithering and halftoning? Digital color? POV-Ray?

actionfromafar 4y ago
There should be a clean split between fixed function accelerations, like 3DFX did, and when you could compile simple code to be run on the GPU. (Shaders?) Also, at some point (NVIDIA?) you could start downloading graphics (or computing results) back to the CPU.
- a_e_k 4y ago
  Yes, definitely. IIRC, the fixed-function aspect is where NVIDIA's marketing made the distinction between their "GPU"s and the graphics accelerators that had gone before. That sort of thing would have tied in nicely with a discussion of the evolution of shading languages if this book had had such. And credit where it's due: I believe that NVIDIA also led the way on going from abusing the graphics pipeline to do GPGPU compute workloads to opening it up to much cleaner approaches like CUDA.

by a_e_k 4y ago

actionfromafar 4y ago

There should be a clean split between fixed function accelerations, like 3DFX did, and when you could compile simple code to be run on the GPU. (Shaders?) Also, at some point (NVIDIA?) you could start downloading graphics (or computing results) back to the CPU.

a_e_k 4y ago
Yes, definitely. IIRC, the fixed-function aspect is where NVIDIA's marketing made the distinction between their "GPU"s and the graphics accelerators that had gone before. That sort of thing would have tied in nicely with a discussion of the evolution of shading languages if this book had had such. And credit where it's due: I believe that NVIDIA also led the way on going from abusing the graphics pipeline to do GPGPU compute workloads to opening it up to much cleaner approaches like CUDA.