Speaker: Seth Goldstein, Carnegie Mellon University
Date: September 23, 1999
Architectures and Compilers for Reconfigurable Computing
Future computing workloads will emphasize an architecture's ability to perform relatively simple calculations on massive quantities of mixed-width data. In this talk I will describe a novel reconfigurable fabric architecture being developed at CMU, PipeRench, optimized to accelerate these types of computations. PipeRench enables fast, robust compilers, supports forward compatibility, fault tolerance, and virtualizes configurations, thus removing the fixed size constraint present in other fabrics. In addition, I will discuss the compilation technology developed to target PipeRench. I will demonstrate extreme performance speedup on certain computing kernels, up to 190x versus a modern RISC processor and even, in one case, 25% faster than custom hardware.
Seth received his PhD from UC-Berkeley in 1997. Since arriving at CMU he has focused on reconfigurable computing. He is involved in the construction of the PipeRench chip, the development of tightly integrated general-purpose processors with reconfigurable fabrics, and the development of software tools for programming reconfigurable computing systems.