FPGA Accelerated HPC and Data Analytics
Mike Strickland · 2018
Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. There are increasing opportunities in the data center for FPGA algorithm, networking, and data access acceleration. Microsoft has announced that they accelerate Bing search, machine learning, and networking with FPGAs - and they recently announced integration of FPGA based Project Brainwave with Azure Machine Learning for real-time AI. FPGAs also deliver some future-proofing for AI, with current support for reduced precision floating point and bfloat16 for better performance with minor loss of scoring accuracy. Intel and partners are also integrating FPGAs to data analytics frameworks and existing databases to enable enterprise customers to run unmodified applications without requiring any FPGA expertise for use with unstructured, NoSQL, and traditional relational databases. One such partner, Swarm64, is using a single FPGA for multiple acceleration roles to deliver 2X+ performance on the industry standard TPC-H benchmark.