Specifications
Accelerators
4, 5 or 6 x Hailo-8 AI processors
On-card Fabric
Integrated PCIe switch linking all Hailo-8 devices
Peak AI Throughput
Up to 156 TOPS (6 x 26 TOPS)
4-Channel SKUs
FALCON-H8C (commercial grade), FALCON-H8F (industrial grade)
5-Channel SKUs
FALCON-H8B (commercial grade), FALCON-H8E (industrial grade)
6-Channel SKUs
FALCON-H8A (commercial grade), FALCON-H8D (industrial grade)
Per-processor Power
Hailo-8 typical 2.5 W per processor
Form Factor
PCIe add-in card with heatsink
Host Interface
PCIe slot with onboard switch for device fan-out
Sister Product
Falcon Lite - 1, 2 or 4 x Hailo-8 in multiple configurations
Host Architecture
x86 or Arm hosts; demonstrated on Raspberry Pi 5 via PCIe HAT
Operating System
Linux
AI Frameworks
TensorFlow, TensorFlow Lite, Keras, PyTorch, ONNX
Grade Variants
Commercial-grade and industrial-grade card options
Positioning
Cost-efficient multi-stream deep learning inference offload for appliance CPUs
Overview
The Lanner Falcon-H8 is a PCIe AI acceleration card that packages four, five or six Hailo-8 AI processors behind an on-card PCIe switch. Each Hailo-8 delivers 26 TOPS with typically 2.5 W power draw, so a fully populated six-processor card reaches roughly 156 TOPS while remaining passive-cooled and slot-friendly.
Six ordering variants cover commercial and industrial grades: FALCON-H8A/H8B/H8C for four-to-six processors in commercial grade, and FALCON-H8D/H8E/H8F for the equivalent industrial-grade configurations. The smaller Falcon Lite card carries one, two or four Hailo-8 processors for lower-density deployments.
Because each Hailo-8 is an independent inference device, the card scales linearly across camera streams and model instances, making it a practical way to offload CPU load in edge video analytics appliances. The off-the-shelf PCIe form factor and standard Hailo software stack - TensorFlow, TFLite, Keras, PyTorch and ONNX runtimes on Linux - allow rapid integration into existing x86 or Arm edge systems.
Key Benefits
Up to 156 TOPS on a single PCIe card without consuming host CPU cycles. Onboard PCIe switch fans out to all Hailo-8 devices through one slot. Commercial and industrial grade SKUs let integrators match environmental requirements. 2.5 W per accelerator keeps total power and thermals low. Broad framework support including TensorFlow, PyTorch and ONNX on Linux reduces porting effort, and the Falcon Lite option scales the same architecture down to one to four processors.
Applications
Multi-camera video analytics appliances, industrial machine vision and inspection, smart city and traffic monitoring, retail analytics, access control and authentication, and CPU offload for edge AI servers and gateways.
Request a Quote — LANNER FALCON-H8 — MULTI-HAILO-8 PCIE AI ACCELERATION CARD
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →