The Alveo SN1000 SmartNICs provide software-defined hardware acceleration for a wide range of networking, security, and storage offloads, such as Open vSwitch and virtualisation acceleration (Virtio.net). Security offloads include IPsec, kTLS and SSL/TLS and accelerated storage applications including Virtio.blk, NVMe over TCP, Ceph, and compression and crypto services.
The composable, programmable SmartNICs are believed to be the industry’s first composable SmartNIC. It is characterised by 100Gb/s performance and software-defined hardware acceleration for all function offloads. They offload CPU-intensive tasks and have an open architecture that can accelerate a broad range of network functions at line rate.
Software developers can use the company’s Vitis Networking platform, with programming languages such as P4, C, and C++, to create network functions, protocols, and applications that operate in hardware on the SmartNIC. The software can be used to compose new, and tweak existing, network functions to handle new protocols and applications without replacing hardware, explained Xilinx.
The SmartNICs are based on the Xilinx 16nm UltraScale+ architecture, and are powered by the company’s low latency XCU26 FPGA and a 16-core Arm processor. The SN1000 SmartNICs deliver dual-QSFP ports for 10/25/100Gb/s connectivity with a PCIe Gen 4 interconnect. The 75-W SmartNICs are available in a full height, half length (FHHL) form factor.
The company also launched the Smart World AI video analytics platform, with an ecosystem of partners, to accelerate complex and latency-sensitive AI video inferencing applications. Xilinx Smart World is powered by the Video Machine-learning Streaming Server (VMSS), and can support multiple neural networks on a single Alveo accelerator card at deterministic sub-100ms pipeline latency.
Ecosystem partner solutions that are available now include those from Aupera, for smart city and smart retail using video processing and Alveo accelerators. There is also a toolset from Mipsology to migrate existing AI applications from GPU-based architectures to the Alveo platform, and there is also a means to deliver AI training at the edge on Aleveo accelerators, from DeepAI. This is claimed to boost performance by a factor of 10, compared to GPU-based solutions.
A third announcement was the sub-microsecond latency in algorithmic trading delivered by the Vitis development platform. The addition of an accelerated algorithmic trading (AAT) reference design means that no custom hardware development is need for fast trading performance. The AAT is implemented on Alveo accelerator cards. The modular design includes all the necessary components for an end-to-end, low-latency trading solution. Each module can be customised using C and C++ in the Vitis platform. The AAT reference design is available now and at no cost for Alveo accelerator card customers.
Xilinx’s fpga App Store has ready-to-deploy accelerated applications from Smart World AI video analytics to anti-money laundering and live video transcoding. It has been developed by Xilinx ecosystem partners, and is described as a way to access containerised, pre-built applications that can be deployed in minutes.