2–6 times faster than FPGA competitors, accelerating adoption of Xilinx FPGAs in hyperscale data centers

Xilinx announced new technology that enables the world's largest cloud service providers to rapidly develop and deploy acceleration platforms.

The Xilinx® Reconfigurable Acceleration Stack, powered by FPGAs designed for cloud-scale applications, includes libraries, framework integration, developer boards, and OpenStack support.

This enables over 40 times the computing efficiency of x86 server CPUs through Xilinx FPGAs and provides up to 6 times the computing efficiency of other FPGA competitors. Xilinx enables silicon optimization for performance-intensive workloads such as machine learning, data analysis, and video transcoding using dynamic reconfiguration. This workload optimization can be executed within a few milliseconds by swapping to the optimal design bitstream.

Today, Xilinx FPGAs enable hyperscale data centers that achieve 2 to 6 times computing efficiency in machine learning inference, powered by the structural advantages of DSPs for limited precision data types, abundant on-chip memory resources, and technology that is more than a year ahead of competitors.

Xilinx’s reconfigurable acceleration stack includes compute libraries designed for cloud computing workloads, application libraries with integrated major frameworks such as Caffe for machine learning, PCIe®-based development boards and reference designs for high-density servers, and OpenStack support packages, making it easy to supply and manage Xilinx FPGA-based accelerators.

Karl Freund, Senior Machine Learning Analyst at Moor Insights & Strategy, stated, “Flexibility and feature turnover—that is, the ability to quickly add new features through software—are critical for complex and constantly changing applications like hyperscale,” adding, “As Xilinx points out, the ability to reconfigure and optimize accelerators as algorithms evolve becomes an advantage in hyperscale environments.”

Nazeem Noordeen, Vice President of IP Solutions at Xilinx, said, “This new stack is leading to increased adoption of Xilinx FPGAs in hyperscale data centers. It is the fastest way to achieve computing efficiency 2 to 6 times better than other FPGA competitors.”