SC21 Invited Talk: Kunle Olukotun, Let the Data Flow!

As the benefits from Moore’s Law diminish, future computing performance improvements must rely on specialized accelerators for applications in high performance computing, artificial intelligence, and traditional data processing. These challenging applications are characterized by terabyte sized models, data sparsity and irregular control flow. In this talk, we explain how Reconfigurable Dataflow Accelerators (RDAs) can be used to accelerate a broad set of data-intensive applications with these characteristics. SambaNova Systems is using RDA technology contained in Reconfigurable Dataflow Units (RDUs) to achieve record-setting performance on challenging machine learning tasks. We will describe how RDAs can also be used to accelerate database and HPC applications with irregular control and data flow using a new execution model called Dataflow threads.

Presenter
Kunle Olukotun, Stanford University, SambaNova Systems Inc

—
SC21: The International Conference for High Performance Computing, Networking, Storage, and Analysis

November 14–19, 2021
America's Center, St. Louis, MO

https://sc21.supercomputing.org