Performance analysis of the CG processor SIGHT
Tadashi Naruse, Masaharu Yoshida, Tokiichiro Takahashi · Systems and Computers in Japan · 1992
Abstract SIGHT is a high‐speed computer dedicated to ray tracing. Its architecture supports parallelism at two processing levels: pixel‐level and operation‐level. This paper discusses the performance analysis of the processing element (PE) of SIGHT. PE consists of four units: a TARAI unit for three‐dimensional (3‐D) vector operation, an MP unit for control functions, a DBM unit for data storage, and a DMA unit for data transfer. There are two major aspects of ray tracing analyzed here: 1) concurrency of three floating‐point units (FPU's) of TARAI, and 2) concurrency of functional units of PE. The results show that 1) 2.20 to 2.62 FPUs of TARAI are active during each execution step, and all three FPUs are active for 57.4 percent to 85.0 percent of the total processing time; and 2) the pipeline operation involving the operation of TARAI and the data transfer between DBM and TARAI are achieved at a very high efficiency. The PE architecture of SIGHT has been demonstrated to be suitable for concurrent ray tracing at the operation level.