Authors
Niket K Choudhary, Salil V Wadhavkar, Tanmay A Shah, Sandeep S Navada, Hashem H Najaf-abadi, Eric Rotenberg
Description
A high-performance superscalar processor achieves good performance across a wide range of applications. For any one application, however, its generalized design is not the highestperforming one. Tailoring the design to an application will yield the highest-performing combination of clock frequency and instruction-level parallelism (ILP) for the application, and the lowest power consumption for this level of performance. Suboptimal performance on individual applications was overlooked in the past, when conventional microarchitecture and technology scaling yielded exponential growth in single-thread performance. Now, customizing superscalar processors to applications, classes of applications, or classes of application behavior, may be the chief source of single-thread performance scaling available to designers in the near future.
A single application-customized superscalar processor is not robust across …
Scholar articles