Showing results for 
Show  only  | Search instead for 
Did you mean: 

Xilinx AI in Action: Baidu Edgeboard and Snowlake at XDF China 2019

Xilinx Employee
Xilinx Employee
0 0 2,218

We briefly mentioned Baidu's Edgeboard last year. At XDF 2019 in Beijing Baidu were demonstrating Edgeboard. 

As a quick reminder, Edgeboard is Baidu’s FPGA-based embedded AI solution that uses a Model-Driven Architecture to make Software-Defined Hardware possible. Their software stack is shown below.



Their stack supports multiple forms of hardware including SoMs, cards, boxes and cameras. Users can choose from a variety of suitable hardware, depending on their end application.


This solution has already been used in many interesting applications to enable on-board AI inference, some of which are shown below. The advanced AI capability helps their end customers achieve better accuracy, lower cost and higher productivity.


Thanks to the great device scalability of the Xilinx ZU+ MPSoC family, three series of Baidu Edgeboard are available, each enabling a range of of end users.


The performance of some popular AI models on Edgeboard (ZU3 based) can be found in the following table: 





Snowlake was founded in 2017 and is headquartered in Shanghai China. It focuses on deep-learning computing development and offers FPGA-based solutions and services for applications in the data center, autonomous driving and others.  

At XDF 2019, Snowlake demonstrated a Wide & Deep Recommender System using an Alveo U200. They achieved lower latency and higher throughput compared with CPUs, enabling them to efficiently meet the rapidly increasing data requirements of recommender systems.

Snowlake used a distributed pipeline calculation based on the operators contained in their TensorFlow model. These optimizations enabled the inference performance to be dramatically improved.

They also developed a toolkit to enable model deployment. This allows models to be easily renewed and deployed into real systems. Many open source models are already supported by the Snowlake toolkit.



Compared with unaccelerated CPUs, Snowflake technology can achieve significantly higher throughput and lower latency, leading to a 5x reduction in TCO (total cost of ownership.



For more information, please refer to


Tags (3)