Altera and Baidu Collaborate on FPGA-Based Acceleration for Cloud Datacenters
Altera Corporation and Baidu are collaborating on using FPGAs and convolutional neural network (CNN) algorithms for deep learning applications set to play a critical role in the development of more accurate and faster online search.
The Altera-Baidu demonstration illustrates how much faster image classification can take place using FPGA-accelerated CNNs. In key search functions, such as image classification and recognition tasks, CNNs are considered to be the state-of-the-art and provide record-setting accuracy. Baidu is leveraging Altera Stratix V FPGAs and the Altera SDK for OpenCL, which achieved Khronos OpenCL conformance testing certification in May 2013, to dramatically simplify the implementation of parallel processing applications.
Baidu Research Distinguished Scientist Dr. Ren Wu said, “Baidu is a pioneer and leader in both deep learning and heterogeneous computing, and we believe FPGA acceleration has great potential. OpenCL support is a game changer and will help FPGAs penetrate the mainstream heterogeneous computing world. It opens doors for countless opportunities.”
Altera’s data center technology offerings are based on the company’s high performance Stratix V and Arria 10 FPGAs, and next-generation Stratix 10 FPGAs and SoCs, which are manufactured using the Intel 14 nm Tri-Gate process and feature Altera’s high-performance HyperFlex architecture. Altera’s FPGAs combine unprecedented reconfigurable logic with on-chip memory and DSP blocks, enabling the high performance and flexibility required by the demanding data center environment.
“Baidu and Altera are demonstrating a compelling heterogeneous computing approach to CNN algorithm acceleration,” said Altera Compute and Storage Business Unit Director Michael Strickland. “The programmability and features, such as hard IEEE 754 floating point multipliers and adders in Altera FPGAs, enable servers and data centers to keep up and evolve with complex requirements in search, big data and deep learning.”
