Versal, providing heterogeneous acceleration for every application
Alveo, designed to enhance the performance of industry-standard servers


Xilinx is rapidly transforming into a platform company by announcing on the 3rd the ACAP 'Versal' that enables rapid innovation with software programmability and scalable AI inference, and launching the accelerator card 'Alveo' that provides 4x the performance of GPUs, 90x the performance of CPUs, and superior adaptability.


Versal, combining SW programmability and domain-specific HW acceleration
Xilinx officially announced Versal, an adaptive compute acceleration platform (ACAP), at the Xilinx Developer Forum (XDF) held in San Jose, California, USA.

Versal ACAP combines scalar processing engines, adaptive hardware engines, and intelligent engines with cutting-edge memory and interface technologies to deliver heterogeneous acceleration for any application. Versal ACAP hardware and software can be programmed and optimized by software developers, data scientists, and hardware developers. It is also enabled by a variety of tools, software, libraries, IPs, middleware and frameworks, enabling industry-standard designs.

Built on TSMC’s 7nm FinFET process technology, the Versal portfolio is the first platform to combine software programmability with domain-specific hardware acceleration. The portfolio includes six series—Versal Prime Series, Premium Series, HBM Series, AI Core Series, AI Edge Series, and AI RF Series—that deliver scalability and AI inference capabilities for a range of applications from cloud to networking, wireless, edge computing, and endpoints.

All six series feature high AI engine specifications. The AI Engine is a new hardware block designed to address the need for low-latency AI inference capabilities in a variety of applications. It supports enhanced DSP implementations for applications such as wireless and radar, and is closely linked to the Versal Adaptable Hardware Engines family to accelerate the entire application. This allows users to tune both hardware and software to maximize performance and efficiency.

“With the explosion of AI and big data and the decline of Moore’s Law, the industry has reached a critical inflection point—silicon design cycles can no longer keep up with the pace of innovation,” said Victor Peng, CEO, Xilinx. “Four years in development, Versal is the industry’s first ACAP that enables developers of all types to accelerate their entire applications with optimized hardware and software, designed to instantly adapt both to keep pace with rapidly evolving technology.”

Xilinx is currently engaging with a number of key customers through its Early Access Program. Versal is scheduled to be released in the second half of 2019.


Alveo Accelerates Data Centers and AI with Support for 14 Ecosystems
On the same day, Xilinx launched its accelerator card portfolio, 'Alveo'.

Alveo is designed to improve performance of industry-standard servers across cloud and on-premise data centers. When running critical data center applications such as image processing, genomics, data analytics, and real-time machine learning inference, Alveo can help you improve performance while reducing latency.

The Alveo U200 and U250 are based on Xilinx UltraScale Plus FPGAs, enabling customers to optimize for changing workloads, new standards, and updated algorithms without replacement costs by reconfiguring the hardware.

Alveo accelerator cards improve performance across a wide range of applications. For example, for machine learning, the Alveo U250 delivers 20x the real-time inference throughput of high-performance CPUs, and more than 4x the high-performance GPU for applications with sub-2ms latency, all at one-third the latency of GPUs. Some applications, such as database searches, can be dramatically accelerated by more than 90x over CPUs.

Alveo is supported by an ecosystem of partners and OEMs developing and certifying key applications in AI/machine learning, image processing, data analytics, financial risk modeling, security, and genomics.

The 14 ecosystem partners include Algo-Logic Systems Inc., Bigstream, BlackLynx Inc., CTAccel, Falcon Computing, Maxeler Technologies, Mipsology, NGCodec, Skreens, SumUp Analytics, Titan IC, Vitesse Data, VYUsync, and Xelera Technologies.

Top OEMs including Dell EMC, Fujitsu, and IBM are working with Xilinx to certify multiple server SKUs with Alveo accelerator cards, with more to come.

“With the launch of Alveo accelerator cards, Xilinx is rapidly transitioning to a platform company, enabling us to innovate faster than ever before with a growing ecosystem of application partners,” said Manish Muthal, vice president, data center, Xilinx. “We are seeing increasing customer interest in Alveo accelerators, and we are excited to collaborate with the application ecosystem to deliver Alveo-based solutions to our customers.”