Fast generation of computer-generated hologram by graphics processing unit

A cylindrical hologram is well known to be viewable in 360 deg. This hologram depends high pixel resolution. Therefore, Computer-Generated Cylindrical Hologram (CGCH) requires huge calculation amount. In our previous research, we used look-up table method for fast calculation with Intel Pentium4 2.8 GHz. It took 480 hours to calculate high resolution CGCH (540,000 x 63,000 pixels and the average number of object points are 27,000). To improve quality of CGCH reconstructed image, fringe pattern requires higher spatial frequency and resolution. Therefore, to increase the calculation speed, we have to change the calculation method. In this paper, to reduce the calculation time of CGCH (912,000 x 108,000 pixels), we employ Graphics Processing Unit (GPU). It took 4,406 hours to calculate high resolution CGCH on Xeon 3.4 GHz. Since GPU has many streaming processors and a parallel processing structure, GPU works as the high performance parallel processor. In addition, GPU gives max performance to 2 dimensional data and streaming data. Recently, GPU can be utilized for the general purpose (GPGPU). For example, NVIDIA's GeForce7 series became a programmable processor with Cg programming language. Next GeForce8 series have CUDA as software development kit made by NVIDIA. Streaming processor of GPU has 1.35 GHz clocks, and GeForce8800GTX has the 128 processors. Theoretically, calculation ability of GPU is announced as 500 GFLOPS. From the experimental result, we have achieved that 47 times faster calculation compared with our previous work which used GPU. Therefore, CGCH can be generated in 95 ours. So, total time is 110 hours to calculate and print the CGCH.