Power Point Slides aus Rutgers - und anderswo http://archive.dimacs.rutgers.edu/Workshops/Parallel/slides/… wie gehts weiter in der nach - Moore - Ära? einige Impulse, sehr subjektiv gesammelt Computational Electromagnetics in Post-Moore’s Law Era: Verification, Validation, and Benchmarking Adapting Mobile To A Post-Moore’s Law Era IC 'megatrends' point to post-Moore's Law era, warns TSMC's Chang predicted feature size in 2024 (7.5 nm) = ~32 silicon atoms (Si-Si lattice distance is 0.235 nm) Rock’s Law: Cost of semiconductor chip fabrication plant doubles every four years low entropy computations CS is lacking a good algorithmic theory of entropy Communication can be much cheaper if “known in advance” Self-Organizing Computations • Hardware continuously changes (failures, power management) • Algorithms have more dynamic behavior (multigrid, multiscale – adapt to evolution of simulated system) • Mapping of computation to HW needs to be continuously adjusted • Too hard to do in a centralized manner -> Need distributed, hill climbing algorithms Resiliency • HW for fault correction (and possibly fault detection) may be too expensive (consumes too much power) • and is source of jitter • Current global checkpoint/restart algorithms cannot cope with MTBF of few hours or less • Need SW (language, compiler, runtime) support for error compartmentalization • Catch error before it propagates • May need fault-tolerant algorithms • Need new complexity theory The Von Neumann Bottleneck Quantum Computing Neuromorphic Computing Optical Computing Bioinformatics is It Time To Rethink The Scientific Method? Bioinformatics Big Data machine learning algorithms the Democratization of Exploratory Research The End of Moore's Law: A New Beginning for Information Technology |
|
aus der Diskussion: | LPKF Laser & Elektronics AG: Die Analyse! |
Autor (Datum des Eintrages): | bezzz (08.04.19 14:55:24) |
Beitrag: | 87 von 377 (ID:60304525) |
Alle Angaben ohne Gewähr © wallstreetONLINE |