Lacelli™ Edge Inferencing Server AI Acceleration Subsystem

Industry's Best Ratio of Performance-to-Power

Product overview

Integrates leading AI performance combined with low power use provided by Gyrfalcon’s Lightspeeur® 2803 AI Accelerator chips, together with Arm® core based multi-core processor ICs. The Gyrfalcon Accelerator chips are mounted on M.2 cards, and they can function independently or aggregated to work on large AI models. Each sub-system is connected to the server via a PCIe interface. Each AI Module delivers 67 TOPS, and 1,200 FPS, consumes only 5 watts of power, delivering 240 FPS/W.

High Performance

67 - 2,144 TOPS

Low Cost

Priced for scale

Best Peak Performance

240 FPS/Watt/ module

Low Power

5 Watts/module

Server - AI Accelerator Subsystem

– 32 x AI Module Slots and Ports
– High performance CPU
– Redundant power supplies
– Twin type fans

ai module

 – PCle Accelerator Card 
– Gen2 x16 as 4 x 4 bifurcated lanes 
– 4 x Lightspeeur® 2803S 


Flexible on-chip model processing capabilities allow the Lacelli™ Edge Inferencing Server AI Acceleration Subsystem to be used with different configurations. Batch parallel process images and video through single model-to-chip config or larger and more complex models such as ResNet-101 or ResNet-152 cascaded through multiple chips.

TARGETED inference

 Image Recognition · Object Detection & Tracking · Natural Language Processing · Natural Language Understanding · Business Intelligence · Visual Analysis

MODELS & performance

VGG – 2,914 FPS
ResNet – 20,480 FPS
MobileNet – 12,800 FPS

Suite of Development and Training Tools

GTI offers the tools to build and deploy Artificial Intelligence solutions on edge and cloud deployments.

Train your models and develop powerful AI applications:

► Software Development Kit
► Model Development Kits
► Hardware Integration Kit
► Application Suite