Xilinx Introduces Kria SoM and Vision AI Devkit Based on Zynq Ultrascale+ FPGA MPSoC for AI Inferencing

Recently, our partner Xilinx declared its intentions to enter the expanding system-on-modules (SoM) market. Xilinx introduced the Kria portfolio of adaptive system-on-modules for accelerating innovation at the edge. The solution is a real challenger for the existing competitors, as it has the best price/functionality balance. Besides, it’s right on time: the SOM market is showing robust growth. Global Market Insights reports that the system on module market size will exceed USD 2.5 billion by 2026.

The Kria K26 SOM employs the Xilinx Zynq UltraScale+ MPSoC architecture. It has a small form factor which makes it an ideal fit for Smart Surveillance, Smart City, Computer Vision, Security Applications, and Retail Analytics. Kria SOM ships in both commercial (250$) and industrial (350$) grades.

Custom-built XCK26 SoC based on the Zynq® UltraScale+ Source: Xilinx

Setting aside the utmost attractive features of the Kria, Xilinx makes a real venture: the thing is that Nvidia, with its Jetson Nano family, currently reigns the edge inferencing market. Apparantly, Xilinx has the resources of a large company to compete. Recently, they published a paper with a comparison of the performance of K26 SOM against the Nvidia Jetson Nano and Jetson TX2 based on the most popular industry models, including ResNet50, SSD_Mobilenet_V1, Tiny Yolo V3, and others. 

The results show that the K26 SOM outperforms both Jetson Nano and Jetson TX2 devices. Furthermore, Kria achieved a massive throughput gain, which is greater than 4X over Jetson Nano. Besides, it also had a significant advantage in performance/watt – more than 2X over Jetson TX2.

The comparative analysis concerning the performance of a real vehicles licence plates detection & recognition application also showed a significant performance advantage of the K26 SOM. Furthermore, thanks to adaptable hardware in the K26 SOM, developers can reach lower latency and lower power consumption while accelerating complete applications. Assuming those advantages and lower total ownership cost, Xilinx has a market winning solution for Edge AI applications and devices.

Promwad, being a trusted member of the Xilinx Partner Ecosystem, is ready to provide our clients with hardware and embedded software design services on Kria for video processing, AI and Machine Learning, embedded computing and IIoT systems integration. If you plan projects on Xilinx, we can help you reduce the design cycle with a custom carrier board based on Kria. Don’t hesitate to contact us!