The Evolution from Chip Performance to System Performance
To meet the ever-increasing computing needs, especially driven by the emerging application of Machine Learning in several fields, the computing industry has entered an era of parallelization, in which tens of thousands of compute servers are connected in warehouse-scale data centers. As the merits of progressing Moore’s Law are slowing down, these parallel, general-purpose computing systems face serious challenges in terms of improving performance, energy and cost. Moreover, requirements of real time performance, privacy and communication bandwidth, driven by the emerging fields of Autonomous Cars and 5G wireless, create the need for more powerful and efficient computing capabilities at the edge of the network.
In this context domain-specific computing holds significant advantages. Domain-specific computing utilizes a customizable architecture and programming environment tailored to a particular application domain. Whereas faster compute hardware can be essential to improve the performance of a system, other aspects such as efficient memory access, interconnect bandwidth, deterministic behavior and low latency can be of equal importance.
Modern All Programmable FPGA platforms are well suited to assume a central role in the implementation of computing systems in the cloud and the edge infrastructure. They provide a parallel and heterogeneous programmable platform that can be customized to optimally meet the performance requirements of different application domains.