H. Sato, S. Takase, A. Ozaki, and T. Wakayama (Japan)
Parallel Computing, Supercomputing, Synthetic ApertureRadar, Graphics Processing Unit.
We increased the speed of image formation processing for a synthetic aperture radar (SAR) by using a graphics processing unit (GPU), which is a high-performance processor whose speed can be increased at a low price. When a GPU is used for image formation processing in a SAR, data must be transferred between the CPU and the GPU every time the direction of processing changes. This leads to a data flow bottleneck. In order to solve this problem, we adopted a “sub-aperture” method that is an algorithm for image formation processing. This algorithm reduces the amount of memory and is suitable for parallel processing. We also developed a method by which GPU operation and the data transfer can be executed simultaneously by installing two data areas on the GPU. The results of our study show that image formation processing by the GPU is 13 times faster than that by one CPU processor and 2.8 times faster than that by eight CPU processors. A cluster server of general-purpose CPUs that is capable of image formation processing at the same speed as a GPU board would have to be 7 times larger, 3.8 times more expensive, and consume 5 times more electric power.
Important Links:
Go Back
