The Ultimate DataFlow for Ultimate SuperComputers-on-a-Chip, for\n Scientific Computing, Geo Physics, Complex Mathematics, and Information\n Processing

2020/09/20 by Veljko Milutinović, Milutinovic, Veljko, Erfan Sadeqi Azer +37
Computer Science · #Advanced Data Storage Technologies #Distributed #Distributed and Parallel Computing Systems #FOS: Computer and information sciences #Parallel #Parallel Computing and Optimization Techniques #and Cluster Computing (cs.DC)

paper · pdf · doi:10.48550/arxiv.2009.10593

openalex publication_date 2020/09/20 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28

Abstract

This article starts from the assumption that near future 100BTransistor\nSuperComputers-on-a-Chip will include N big multi-core processors, 1000N small\nmany-core processors, a TPU-like fixed-structure systolic array accelerator for\nthe most frequently used Machine Learning algorithms needed in bandwidth-bound\napplications and a flexible-structure reprogrammable accelerator for less\nfrequently used Machine Learning algorithms needed in latency-critical\napplications.\n

Related

• High-Performance Special-Purpose Computers in Science
• Challenges Towards Deploying Data Intensive Scientific Applications on Extreme Heterogeneity Supercomputers
• Solving the compute crisis with physics-based ASICs
• High Performance Optimization at the Door of the Exascale
• TPU v4: An Optically Reconfigurable Supercomputer for Machine Learning with Hardware Support for Embeddings