Chapter 35. Active Pages: Memory-centric Computation
Diana FranklinDepartment of Computer Science California Polytechnic State University
Although field-programmable gate arrays (FPGAs) excel at tailoring the computation and interconnect to an application’s needs, we can go one step further. In many applications, regardless of the speed of the computation, memory performance always will be the limiting factor. This problem, referred to as the memory wall, is broken up into two parts—memory latency and bandwidth. For large-scale data-parallel applications, the computation can be moved to memory. This allows for both parallel computation and increased bandwidth. The replication of small computation units provides parallelism, and the sum of their ...
Get Reconfigurable Computing now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
