PDF, English Lizenz: Print on Demand Download (1148Kb)

Abstract

HyperTransport provides a flexible, low latency and high bandwidth interconnection between processors and also between processors and peripheral omponents. Therefore, the interconnection is no longer a performance bottleneck when integrating application specific accelerators in modern computing systems. Current FPGAs providing huge computational power and permit the acceleration of compute-intensive kernels. We therefore present a general purpose architecture based on HyperTransport and modern FPGAs to accelerate time-consuming computations. Further, we present a prototypical implementation of our architecture. Here we used an AMD Opteron-based system with the HTX Board [6] to demonstrate that common applications can benefit from available hardware accelerators. A cryptographic example showed that the encryption of files, larger then 50 kByte, can be successfully accelerated.