Xilinx, a global leader in adaptive and intelligent computing, today announced the launch of the industry’s first “all-in-one SmartNIC Platform”—Alveo U25, which truly realizes the perfect integration of network, storage and computing acceleration functions on a single device. U25 is specifically aimed at those struggling with the ever-increasing networking needs and rising costs.cloud serviceProviders, telecommunications companies, and private cloud data center operators are designed to provide them with SmartNIC’s higher efficiency and lower total cost of ownership (TCO). U25 combines a highly optimized SmartNIC platform with a powerful and flexible FPGA engine to realize fully programmable and one-stop acceleration applications. U25 provides a powerful integrated SmartNIC platform to meet the industry’s most challenging requirements and workloads such as SDN, virtual switching, NFV, NVMe-oF, Electronic transactions, AI inference, video transcoding, and data analysis.
The industry’s first “all-in-one SmartNIC platform” — Alveo U25
In addition, Xilinx also released its first XtremeScale Ethernet adapter card, which conforms to the Open Computing Project (OCP) 3.0 size specification. At the same time, Xilinx today released the world’s first FPGA-based OCP accelerator module (OAM) proof-of-concept board.
The first OCP 3.0 new XtremeScale X2562 Ethernet adapter card
Baron Fung, research director of Dell’Oro Group, a market research organization, said: “It is estimated that by 2024, the SmartNIC market will exceed 600 million US dollars, accounting for 23% of the global Ethernet adapter market. With the vertical expansion of cloud service providers, they are The deployment of SmartNIC is constantly increasing to release valuable CPU cores for business applications and optimize server utilization. Telecom service providers are another market with strong growth potential. They are considering integrating SmartNIC from the core network to the edge network. , To provide services for applications such as NFV and AI inference. FPGA-based SmartNICs such as Alveo U25 cater to this growing market opportunity well.”
Integrated SmartNIC platform for cloud acceleration
As network port speeds continue to increase, Tier 2 and Tier 3 cloud service providers, telecom and private cloud data center operators are facing increasingly severe networking problems and networking cost challenges. At the same time, the large R&D investment required to develop and deploy SmartNIC has also become an obstacle to its widespread adoption. The true plug-and-play function provided by the Alveo U25 SmartNIC platform resolves the above-mentioned challenges and paves the way for the widespread deployment of SmartNIC.
Relying on Xilinx’s industry-leading FPGA technology, the Alveo U25 SmartNIC platform can provide higher throughput and a more powerful flexible adaptability engine than SoC-based NICs, allowing cloud architects to quickly create multiple types of functions and applications Speed up. The U25 SmartNIC platform supports “bump-in-the-wire” seamlessly embedded network, storage and computing offloading and acceleration functions, which can avoid unnecessary data transmission and CPU processing, thereby maximizing efficiency . This also significantly reduces the burden on the CPU and frees up more resources to run more applications. Embedded ARM processors provide unique and critical control layer processing functions that can support emerging bare metal server use cases. Basic NIC can provide ultra-high throughput, small packet performance and low latency. Standard full-featured NIC solutions and drivers use patented Onload application acceleration software, which reduces latency by up to 80%, and improves efficiency for server applications based on the Transmission Control Protocol (TCP) in cloud applications-the highest possible Up to 400%.
Donna Yasay, vice president of marketing for Xilinx’s Data Center Division, said: “The current cloud infrastructure is plagued by critical data bottlenecks caused by server I/O. Up to 30% of data center computing resources are allocated for networking. ／O processing, the overhead continues to grow with the increase in the number of CPU cores. The Smart NIC platform provided by Xilinx is easy to deploy, and has one-stop acceleration applications and out-of-the-box functions far beyond basic networking, which can help users easily deal with Challenges posed by the growing demand for networking.”
Speed up applications out of the box and speed up the product launch process
The Alveo U25 SmartNIC platform realizes a one-stop acceleration application, which can help non-level 1 cloud data center operators to deploy SmartNIC more conveniently and quickly gain results. At the same time, U25 SmartNIC also supports one-stop applications provided by Xilinx and independent software providers (ISV). Its programming model not only supports high-level network programming abstractions such as HLS and P4, but also supports computing acceleration frameworks such as the Vitis unified software platform to implement accelerated applications provided by Xilinx and third parties.
The first out-of-the-box acceleration application provided by U25 is support for Open vSwitch (OVS) offloading and acceleration. This plug-and-play solution will offload more than 90% of the OVS processing from the server, thereby increasing packet throughput by more than 5 times. In the future, Xilinx also plans to launch a one-stop solution for security features such as IPSec, SSL/TLS, AES-256/128, as well as distributed firewall and AI inference.
Currently, Alveo U25 SmartNIC platform is providing samples for early trial customers. It is expected to start mass supply in the third quarter of 2020.
The first OCP 3.0 Ethernet adapter and OCP accelerator module
Xilinx also launched a new XtremeScale X2562 10/25Gb Ethernet adapter card, which complies with OCP Spec 3.0 form factor specifications. X2562 is designed for high-performance electronic transaction environments and enterprise-level data centers. It can provide sub-microsecond latency, high throughput, and ultra-large-scale connectivity that can connect real-time data packets and information streams to thousands of virtual NICs . X2562 has already begun to provide samples and will be available in batches in the second quarter of 2020.
In addition, Xilinx also released the world’s first FPGA-based Open Computing Accelerator Module (OAM) proof-of-concept board. The mezzanine card is based on Xilinx UltraScale+VU37P FPGA and is equipped with 8GB HBM memory, conforms to the Open Accelerator Infrastructure (OAI) specification, can support seven 25Gbps x8 links, and provides a rich inter-module system topology for distributed acceleration.
Xilinx is committed to supporting the rapid innovation of many different technologies from the endpoint to the edge to the cloud by developing a highly flexible and adaptive processing platform. Xilinx is the inventor of FPGA, hardware programmable SoC and ACAP, aiming to provide the industry’s most dynamic processor technology to realize an adaptive, intelligent and interconnected future world.
Copyright 2020 Xilinx, Inc. All rights reserved. Xilinx, the Xilinx logo, Zynq and other designated brands mentioned in this article are trademarks of Xilinx in the United States and other countries. All other trademarks are the property of their respective owners.