Xilinx ships 7 nanometer ACAP, in response to artificial intelligence heterogeneous computing needs

Xilinx, a provider of self-adjusting and intelligent computing equipment, announced on the 19th that it will begin shipping its VersalAICore and VersalPrime components built on TSMC's 7nm process and provide it to a number of first-line customers involved in early-stage trials. Xilinx pointed out that Versal is the industry's first self-adaptive computing acceleration platform (ACAP), a new heterogeneous computing component that far exceeds traditional CPU, GPU and FPGA.

Xilinx further pointed out that ACAP is a highly integrated multi-core heterogeneous computing platform that can be modified at any time in hardware and software to respond to a wide range of applications in data centers, automotive, 5G wireless, cable and defense. Job load requirements. Moreover, when Xilinx built VersalACAP, its software programmability was taken into consideration, and its architecture consisted of a highly flexible network single-chip (NoC) with a transmission rate of terabits per second (terabit). Its NoC seamlessly integrates all engines and critical interfaces to enable the platform's resources to be used at startup, and is easy for both hardware and software developers and data scientists to program. In addition, through a variety of tools, software, libraries, IP, mediation software and framework resources, ACAP allows users to develop customized accelerated computing solutions at any time through the industry's standard design process.

Currently, Xilinx's VersalACAP series is built on TSMC's 7nm process technology and is the industry's first configurable hardware acceleration for software programmability, application-specific areas, and enabling companies to keep up with today's rapid innovations. The flexible strain characteristics of the footsteps form a platform. This series combines a new generation of scalar engines for embedded computing, a self-adaptive engine for FPGA chip programming, and a smart engine for AI inference and advanced signal processing, in terms of computing power and performance-to-power ratio. Both greatly exceed the CPU and GPU.

It is understood that the VersalAICore series can provide the highest computing power and lowest latency performance in this product line, and bring breakthrough AI inference transmission volume and performance through the AI ​​engine. In addition, it optimizes the cloud, networking and autonomous technologies, providing the industry's largest range of AI and workload acceleration capabilities. The VersalPrime range is designed for a wide range of applications and optimized for networking and online acceleration for a wide range of workloads.

In addition, Xilinx's VersalAICore and VersalPrime series include a variety of components, each of which includes a dual-core ArmCortex-A72 application processor, a dual-core ArmCortex-R5F real-time processor, and self-tuning hardware components with more than 2 million logic cells. And more than 3,000 DSP engines optimized for high precision floating point and low latency. The VersalAICore platform also provides up to 400 AI engines that have optimized AI inference and advanced signal processing workloads. The other four product lines in the Versal series, each with its own unique architecture, provide scalability for a wide range of applications from cloud, networking, wireless communications, edge computing to end-of-line devices. With the AI ​​inference function, related products are also expected to be fully available in the second half of 2019.