Abstract
Floating-point arithmetic performance of GPU (Graphics Processing Unit) keeps increasing and it has been well studied to use GPU for general purpose numerical calculation (GPGPU). GPGPU can be applied to an implementation of the SOM learning algorithm, and some researchers have achieved significant speedup on SOM learning process. Therefore GPU has an appropriate memory for example texture-cache, shared-memory and so on, further speedup is expected by using these memories. In this paper, we discuss how to implement SOM learning algorithm in GPU focused on the GPU’s appropriate memory systems and degree of speedup in execution of SOM learning process.
