Lanner Electronics Falcon H8 AI accelerator card powered by Hailo 8 AI processor...

Lanner Electronics Falcon H8 AI accelerator card powered by Hailo 8 AI processors

May 11, 2022 — by Giorgio Mendoza

130 views

Hailo and Lanner Electronics launched a cost efficient PCIe accelerator card called the Falcon H8. The device is designed to enable legacy devices such as NVRs, Edge AI boxes, Industrial PCs to efficiently run intensive workloads related to manufacturing, smart cities, retail, etc. 

The Falcon H8 is capable of accommodating 4, 5 and 6 Hailo-8 AI processors to customize performance and cost effectiveness. Each Hailo-8 edge AI processor is capable of running 26 tera-operations per second (TOPS). Additionally, the Falcon H8 accelerator cards are available in commercial grade and industrial grade variations (-40 – 85C). 

Falcon H8 side view (left) and top view (right)
(click images to enlarge) 

According to both companies, the Falcon H8 is capable of delivering machine learning inference processes of over 15,000 Frames Per Second (FPS) for MobilNet-v2 and 8,000 FPS for ResNet-50.

For software development, Hailo provides their own Hailo Dataflow Compiler API which supports well known ML frameworks such as TensorFlow, ONNX, Keras and Pytorch. According to Hailo’s documentation, a Deep Learning model is used as an input to the Hailo Dataflow Compiler. The resulting output is a binary file which gets loaded to the Hailo device. The library built is used by the runtime applications. 

High Dataflow Compiler by Hailo
(click images to enlarge) 

Lanner Electronics also released a short review for their Falcon H8 on their social media account. 

Falcon H8 Lanner Electronics Review 

Specifications listed for the Falcon H8 include:

• Processor System:
  • 4-6 Hailo-8 AI Processors w/ Hailo Patented Structure Dataflow Architecture
  • Up to 156 TOPs 8000 FPS of ResNet-50, 1300 FPS of YOLOv5m 
• PCI Express Interface:
  • PCI Express x16 Compliant with PCI Express specification v3.0
• AI Frameworks:
  • Hailo AI Dataflow Compiler W/ Profiler And Emulator supports TensorFlow, ONNX and PyTorch frameworks
• System Compatibility:
  • Intel x86 & ARM devices

• Power:

• Temperature:
  • 0 – 70ºC (commercial grade)
  • -40 – 85ºC (industrial grade)
• Dimensions:
  • 167.6 x 111.15 mm without bracket

 Further information

For information regarding price and availability refer to Lanner’s Electronics website.