Advanced Computing in the Age of AI | Saturday, December 3, 2022

Hyper-scalers Speeding Apps Via FPGAs 


Nimbix is the latest HPC cloud vendor to collaborate with FPGA pioneer Xilinx Inc. to accelerate cloud datacenter workflows that increasingly include computing-intensive data analytics, machine learning and video processing workloads.

The partners announced the availability Monday (Nov. 14) of a software-defined Xilinx development platform for deploying workflows accelerated by FPGAs in the Nimbix cloud. The deal coincides with the release by Xilinx of its "reconfigurable acceleration stack" aimed at reducing programming hurdles linked to speeding up computing-intensive workloads running in hyper-scale datacenters.

The development environment combines a reconfigured compiler with libraries and other FPGA development tools. The compiler supports combinations of the OpenCL framework that supports FPGAs, CPUs, GPUs and other processing platforms along with C and C++ kernels. Meanwhile, the libraries cover deep neural networks, and SQL "data-mover" library along with an SQL-based computing kernel.

Xilinx (NASDAQ: XLNX), San Jose, Calif., has been promoting mainstream adoption of FPGAs in hyper-scale datacenters as a way of tackling the steep processing requirements of emerging data and video workloads. The partners said their collaboration would deliver "on-demand access" to accelerators and development tools based on reconfigurable FPGAs.

Xilinx notes that large cloud vendors such as Microsoft Azure (NASDAQ: MSFT) and China's Baidu have begun deploying its FPGA technology in their datacenters as a way to leverage "dynamic reconfiguration" to optimized hardware handling computing intensive workloads. Baidu said last month it has ganged FPGAs into "pools" to accelerate machine learning inference tools.

Along with Nimbix, Xilinx has also recently announced partnerships with IBM (NYSE: IBM) and Qualcomm (NASDAQ: QCOM) to accelerate datacenter operations. For example, IBM is using Xilinx FPGAs in an OpenPower storage and networking framework aimed at accelerating applications such as NoSQL databases.

Xilinx released its FPGA development platform in 2014, and unveiled details of its 16-nanometer Virtex FPGAs last week. These and other moves such as the formation earlier this year of an interconnect consortium underscore how FPGAs are moving into datacenters as traditional processing architectures fail to keep pace with big data and other workloads. Along with Xilinx, founding members of the industry consortium include Advanced Micro Devices (NASDAQ: AMD), ARM Ltd. (TYO: 9984), Chinese networking giant Huawei (SHE: 002502), IBM, Mellanox (NASDAQ: MLNX) and Qualcomm.

Add to that list HPC cloud provider such as Nimbix, Richardson, Texas, which noted in a statement announcing its collaboration with Xilinx that “enabling developers is key for large scale adoption of advanced technologies" in cloud datacenters.

The deal gives the cloud vendor access to Xilinx FPGA accelerator cards and tools that application developers can use for emerging datacenter workloads ranging from machine learning to video transcoding, the process of converting movie data files.

Nimbix also said it is ramping up FPGA-based data science workflows along with an application container-based continuous integration and deployment capability that allows developers to build, test and deploy FPGA-based applications on the company's cloud marketplace.

About the author: George Leopold

George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).

Add a Comment