Files in this item



application/pdfExascale Research.pdf (198kB)
(no description provided)PDF


Title:Exascale Research: Preparing for the Post-Moore Era
Author(s):Snir, Marc; Gropp, William D.; Kogge, Peter
High-Performance Computing
Moore's Law
Abstract:(i) Achieving exascale performance at the end of this decade or the beginning of next decade is essential for progress in science – including progress on problems of major societal impact (such as weather or environmental impact); essential for the continued certification of the nuclear stockpile; and essential to our national security. (ii) The rate of advance in the performance of CMOS technology is slowing down and is likely to plateau mid next decade. No alternative technology is ready for deployment. (iii) Therefore, achieving exascale performance in 20 years may not be significantly cheaper than achieving it in 10 years – even if we could afford the wait. (iv) It is essential (for continued progress in solving major societal problems, nuclear stockpile, security) to have a sustained growth in supercomputer performance and sustained advantage over competitors and potential enemies. (v) To achieve this continued growth, we need research on (a) using CMOS more efficiently and (b) accelerating the development and deployment of a CMOS replacement. (vi) (a) is (or should be) the focus of exascale research: How to get significantly higher compute efficiencies from a fixed transistor or energy budget. (b) is essential to explore, even if not for exascale in 10 years, as it will be necessary to continue beyond exascale. 
Issue Date:2011-06-19
Genre:Working / Discussion Paper
Publication Status:unpublished
Peer Reviewed:not peer reviewed
Date Available in IDEALS:2011-06-20

This item appears in the following Collection(s)

Item Statistics