Xilinx’s New Customizable FPGA for Data Centers is “Born to Run”

May 26, 2020 by Robin Mitchell

With an increasing burden on networking systems, Xilinx has unveiled a new FPGA with high data processing and flexibility to meet evolving connectivity standards.

Last week Xilinx announced their latest member of the UltraScale family, the Virtex UltraScale+ VU23P. What makes this new FPGA unique? And how do its features aim to tackle the growing issues with network storage and high-bandwidth demands?


Processor Speed vs. Bandwidth

The increasing burden on data centers, especially as more employees are working at home during COVID-19, has presented engineers with a new challenge: processor speed vs. bandwidth. Moore's law holds that the processing power of silicon devices doubles every two years. However, network port speeds have been exponentially growing due to the increasing demand from internet services, and it is feared that server processors and current silicon technology will quickly become inadequate.


Xilinx says port speeds are outstripping Moore's law

Xilinx says port speeds are outstripping Moore's law. Image used courtesy of Xilinx

This is also problematic for server design because the exponential increase in port speed will require servers to either be upgraded or replaced frequently in order for them to handle the increase. Imagine for a moment a server whose core hardware was configurable that could help to offload tasks from the main CPU while providing high-speed bandwidth capabilities. These dual capabilities, Xilinx claims, is what inspired its latest UltraScale+ device. 


Highlights of Xilinx Virtex UltraScale+ VU23P

To help meet the exponential demand in internet services, Xilinx has created an FPGA specifically designed for servers and data centers. The latest device, called the Virtex UltraScale+ VU23P, is an FPGA for adaptable networking and storage with the highest ration of LUTs and block RAM to DSP in the Virtex portfolio.

Xilinx says the FPGA allows maximizing packet processing throughput per port while also allowing bandwidth scaling within a fixed power envelope.



VU23P. Image used courtesy of Xilinx

The FPGA, whose purpose is to offload work from SDN and NFV as a Smart Network Interface Card (SmartNIC) can be used in applications requiring 200 Gbps speeds while also integrating 100G Ethernet MAC with KR4-FEC, 34 transceivers operating at 32.75 Gbps, and integrated blocks for PCIe Gen 4 for maximized bandwidth and low latency. 

The physical size of the Virtex UltraScale+ VU23P measures at 35 mm x 35 mm, making it a useful option for high-density server environments while also providing flexibility to support a range of standard server form factors. 


How Does the FPGA Appeal to Designers?

While these main features give an overview of the capabilities of the VU23P, the product brief details what engineers will be most interested in. The FPGA, based on 16nm FinFET+ technology integrates the same scalable technology as other 20nm UltraScale FPGA devices, making migration easy.

The VU23P allows for both 28G and 58G backplane support while also offering 32.75G and 58G chip-to-chip as well as chip-to-optics. The integrated 100G Ethernet saves typically between 60K-100K system logic cells per port, freeing up resources for other tasks while also providing up to 90% power saving when compared to software implementations.


Some of the columnar resources of the FPGA family.

Some of the columnar resources of the FPGA family. Image used courtesy of Xilinx

The PCIe Gen4 x8 and PCIe Gen3 x 16 allows for up to 16GB/s bandwidth while the use of expanded virtualizations can help with data center applications. On-chip deep buffering provides 79 Mb Block RAM for deed FIFO while also supporting DDR4 with speeds up to 2,666 Mbps.

Advanced clock routing techniques help to lower clock skew for use in high-performance networks and the efficient use of CLBs help to reduce interconnection delays.


The More Customizable, the More Applications

The VU23P is targeted at mostly server-related applications, including full OVS offloading, network security acceleration, SmartNIC, NVMe-oF acceleration, flash storage controller, front-haul gateway acceleration, and PON access.

The FPGA can be custom programmed to any application, something that current network solutions lack. A good example of where the VU23P could be potentially game-changing is the video streaming industry. As the resolution of displays increases with 4K becoming the norm, the demand for high-quality video increases.

Increasing the quality of a video increases its storage size. Larger videos require a higher bandwidth if a video is to be streamed without buffering. Plus, videos require multiple copies stored around the world to minimize latency and download time. Thus, a solution is needed that can access data on drives while minimizing the size of the file on disk simultaneously.

The VU23P in this situation would be helpful not only to accelerate data compression and decompression (to reduce the size on network storage) but to stream the content at high speed, freeing up server CPU resources. 


A Bright Horizon for Network Systems

The VU23P is an FPGA designed to be customizable and reprogrammable while offering hardware features aimed specifically at networking. Xilinx claims that this new device will get engineers to “run” to beat the competition. While the VU23P may not be groundbreaking in its own right, it may affect how network systems are designed.

This concept of replacing ridged silicon hardware and soft implementations with customizable systems may be the way forward since it can be adjusted, improved, and scaled over time.



If you work with networking systems, what is your greatest challenge with FPGAs? Share your experiences in the comments below.