GRAPE-MP: An SIMD Accelerator Board for Multi-precision Arithmetic

We describe the design and performance of the GRAPE-MP board, an SIMD accelerator board for quadrupleprecision arithmetic operations. A GRAPE-MP board houses one GRAPE-MP processor chip and an FPGA chip which handles the communication with the host computer. A GRAPE-MP chip has 6 processing elements (PE) and operates with 100MHz clock cycle. Each PE can perform one addition and one multiplication in every clock cycle. The architecture of the GRAPE-MP is similar to that of the GRAPE-DR. It is implemented using the structured ASIC chip from eASIC corp. A GRAPE-MP processor board has the theoretical peak quadruple-precision performance of 1.2 Gflops. As a preliminary result, we present the performance of the GRAPE-MP board for two target applications. The performance of the numerical integration of Feynman loop is 0.53 Gflops. The performance of a N-body simulation with the second order leapfrog schema is 0.505 Gflops for N = 1984, which is more than 10 times faster than the performance of the host computer