Real-time electroholography using a single spatial light modulator and a cluster of graphics-processing units connected by a gigabit Ethernet network

Abstract

Systems containing multiple graphics-processing-unit (GPU) clusters are difficult to use for real-time electroholography when using only a single spatial light modulator because the transfer of the computer-generated hologram data between the GPUs is bottlenecked. To overcome this bottleneck, we propose a rapid GPU packing scheme that significantly reduces the volume of the required data transfer. The proposed method uses a multi-GPU cluster system connected with a cost-effective gigabit Ethernet network. In tests, we achieved real-time electroholography of a three-dimensional (3D) video presenting a point-cloud 3D object made up of approximately 200,000 points.

Fast and accurate phase processing in off-axis digital holography combining adaptive spatial filtering and an embedded GPU platform

“…To allow faster phase processing, graphic processing unit (GPU) acceleration technology has also been applied to achieve highly efficient data processing for offaxis DH [17][18][19][20][21][22][23]. Sannomiya et al [24] proposed a rapid GPU packing scheme, that significantly reduced the data transmission required for real-time DH of GPU clusters, further improving the reconstruction speed of holograms. Yin et al [25] proposed an optimized semi-global matching algorithm using GPU, which achieved efficient and accurate depth reconstruction dynamically.…”

“…Virtual memory and shared memory technologies are used in an Ubuntu system, and different computer unified device architecture (CUDA) libraries according to the characteristics of the key steps of the phase reconstruction are used to accelerate GPU parallel computing. Compared with traditional parallel-phase processing methods [24][25][26][27], an embedded GPU platform, such as Jetson Nano Developer Kit is chosen as the hardware platform. It is small, with low power consumption and low cost [28,29].…”