Important detail that I mentioned in class but neglected to include in the slides: Technically, perspective projection is a nonlinear operation so we can't shove it in a matrix alone. We need to do some tricks with the W coordinate from earlier. More info to come!
SlimShady
I guess one example of nonlinear projection matrices is the one we used in the very first lecture, because translation was involved.
bunbohue
How is GPU good at transformation matrices? Is it good at generating the matrix, or is it good at matrix arithmetic?
motoole2
The key advantage of the GPU is that it performs a lot of computations in parallel compared to a CPU. This is because the GPU can have thousands of cores, where as a CPU typically only has a handful of cores, as Adrian discussed here. Given a transform matrix and a set of vertices, a GPU parallelizes the matrix-vector computation required to transform all of these vertices. The generation of these transformation matrices is also easy, since it is small and only has to be done once for all of your vertices.
Important detail that I mentioned in class but neglected to include in the slides: Technically, perspective projection is a nonlinear operation so we can't shove it in a matrix alone. We need to do some tricks with the W coordinate from earlier. More info to come!
I guess one example of nonlinear projection matrices is the one we used in the very first lecture, because translation was involved.
How is GPU good at transformation matrices? Is it good at generating the matrix, or is it good at matrix arithmetic?
The key advantage of the GPU is that it performs a lot of computations in parallel compared to a CPU. This is because the GPU can have thousands of cores, where as a CPU typically only has a handful of cores, as Adrian discussed here. Given a transform matrix and a set of vertices, a GPU parallelizes the matrix-vector computation required to transform all of these vertices. The generation of these transformation matrices is also easy, since it is small and only has to be done once for all of your vertices.