Artificial Intelligence

Machine Learning, Video Processing

Machine Learning Inferencing

Seamless Deployment, Broad Network Support, Power Efficient

Video Transcoding

Edge-based, Low Latency Video Transcoding Use Case

Custom Processing

FPGA Powered Custom Acceleration

Machine Learning (ML) is fast becoming part of our everyday life. Western Digital has engineering projects investigating their uses in our products and processes. Given that data is the key to the learning, accuracy and intelligence of algorithms/networks, ML is a growing focus for us. Western Digital’s U.2 FPGA based card enables edge processing applications like machine learning inferencing and video transcoding. The solutions contains a FPGA based hardware board combined with various firmware configurations. A wide array of processing solutions can be supported for power conscious edge applications.

Machine Learning at the Edge

Machine Learning Inferencing is implemented on Western Digital’s U.2 FPGA based card coupled with Mipsology’s Zebra IP. This solution provides compelling performance in a power conserving form factor. Numerous popular networks are supported to enable seamless deployment. No retraining or pruning of your algorithms are required. This FPGA based solution is ideal for industrial automation, medical, smart cities and other edge based ML applications.

Edge Based Video Transcoding

Western Digital’s U.2 FPGA based card can support a variety of processing accelerations at the edge. Xilinx’s application of the FPGA-based card can implement video transcoding. Encode and Decode of multiple H.264 or H.265 is supported. Resolution up to 4k@60fps are supported. This transcoding solution is available from Xilinx. The ordering part number from Xilinx is A-U2MA-P04G-PQG-021.

Seamless Deployment, Broad Network Support, Power Efficient

Often edge based applications are constrained both in terms of power and size. The Western Digital’s U.2 FPGA based card is compelling for a wide array of machine learning applications. Specially designed accelerators are becoming popular for their specific machine learning capabilities. Western Digital’s U.2 FPGA based card with Mipsology’s Zebra IP provides offload for traditional processors saving power and improving performance.

ZebraIP allows trained networks to run seamlessly for cloud trained implementations. The solution provides compelling performance/mWatt for edge-based machine learning inferencing applications.