Multiscale Dataflow Computing - Competitive Advantage at the Exascale Frontier

Explore the cutting-edge world of Multiscale Dataflow Computing in this Stanford University seminar. Delve into the competitive advantages of this technology at the Exascale frontier, presented by Brian Boucher from Maxeler Technologies. Learn about the motivation behind dataflow computing, its role in overcoming frequency scaling limitations, and its push towards exascale machines. Discover the Maxeler dataflow ecosystem, including MaxJ code and DFE hardware, and see how dataflow principles are applied to the Quantum ESPRESSO HPC software package. Gain insights into energy-efficient high-performance computing and its applications in industries ranging from energy to finance and defense. Understand how dataflow computing maximizes performance density by building computers around specific problems, eliminating the need for elaborate caching and decoding machinery found in standard processors.

Intro
Moores Law
Control Flow Model
Dataflow Model
About MaxSeller
Success Stories
Building a Dataflow Computer
Dataflow Code
Complex Graphs
Building a Machine
Physical Representation
PCIe
Memory
Optical Interconnect
Network
Hardware
Example
Memory Hierarchy
Aerial View
Data Flow
Performance
Compute
Power Efficiency
Software Integration
App Gallery
Finite Difference
Machine Learning
QA
Graph