Accelerator technologies


In the multicore era, hybrid systems that combine general-purpose processor cores with application-specific accelerator cores receive considerable attention because they offer better scaling properties and power savings compared to monolithic systems. The current leader (May09) in the TOP500 list of supercomputers [1] is a hybrid system that combines an x86-based chip with the Cell broadband engine (Cell B.E) to surpass the petaflop barrier.

The main focus of the accelerator technologies group at the Zurich Research Lab is on hardware-based accelerators for data-processing applications, for example regular expression matching engines for intrusion-detection [2].

Whereas the majority of the high-performance computing (HPC) community focuses on the acceleration of scientific computation (e.g., matrix multiplication), our research focuses on applications that process large amounts of data for edge-of-the-network applications (e.g., intrusion detection and host-ethernet adapters), cryptographic cores and Web 2.0 applications. This research has led to application-specific cores that are now integrated into hybrid processors.

In addition to the work on data-accelerators, we also pursue research on compute-accelerators for parallel hybrid systems. This project uses the Cell B.E. to accelerate image-registration algorithms towards online registration of patient data for computer-assisted surgery.