The NumberSmasher-RPU system is designed for multi-server co-processing development. It incorporates one dual-core AMD Opteron processor, DRC's RPU (Reprogrammable Processor Unit), Linux OS and PGI MPI and compiler software. DRC's RPU coprocessor module provides direct access to adjacent DDR memory and Opteron processors at HyperTransport™ speed and nanosecond latency. Tight coupling between CPU and memory virtually eliminates bandwidth and latency bottlenecks.

The system contains everything you need to modify your application subroutines to run on DRC hardware. Providing high value through a savings on space, heat generation and power consumption, this platform delivers a competitively priced, scalable solution for scientific applications.

The DRC RPU contains specialized cells that can be used to accelerate numeric tasks which work on arrays stored in the memory address space of an AMD Opteron. With 96 such cells on chip, it can perform both SIMD and pipelined parallel operations at very high speeds. These operations can be integrated with Opteron applications that communicate with the coprocessor using the message passing interface sent over the HTX bus that drives the FPGA using code generated by tools that start with kernels written in C.