Graphical processor units (GPUs), as found in nearly all computers nowadays, can be used, not only for pretty 3D graphics, but also for number crushing.
GPUs have a potential massive computational power, but programming them are harder and more error-prone than most "modern" programming techniques on the CPU.
The architecture of a GPU is radically different from a CPU - making it much more tricky to use, but under the right circumstances GPU can provide a phenomenal speedup due to the parallel hardware design.
This speech focus on the programming of the GPU, and gives a thorough walk-through on the various techniques used to make software run stable and fast on a GPU.
Carsten Eie Frigaard is a civil engineer and astrophysicists, working as a consultant and developer at Mergeit Aps. He primarily use the GPUs to run astronomical simulation, but otherwise he work with a broad variety of
open-source software tools - from databases to embedded computers.
| Attachment | Size |
|---|---|
| talk_OSD_2010_Programming_GPUs.pdf | 2.12 MB |