Monica Chiosa gives a presentation at AMD Research Lab

Monica Chiosa gave a presentation at AMD Research Lab about her work on accelerating correlation computation using FPGAs.

Computation of data correlation using FPGAs

Abstract:

The characterization of data entering a system may play a secondary role in the architecture of the system, but it is of paramount importance when knowing the data can help make better decisions. In this talk I am going to introduce a stream analytics system that implements correlation computation on the FPGA. The deployment considers the FPGA as a co-processor and as a SmartNIC, and its design goes beyond the matrix multiplication paradigm and offers a customized solution for correlation computation. The system can easily sustain streams arriving at 100Gbps over an RDMA network, while requiring only tens of milliseconds to compute the correlation coefficients of streams with millions of items each, orders of magnitude better than the CPU design.