GTI Developer Portal

Access tools and resources to design, test and deploy high-performance AI models, optimized to run on GTI’s Artificial Intelligence accelerator solutions.

Gyrfalcon Technology’s (GTI) silicon chips are high performance and low power hardware accelerators used for convolutional neural network (CNN) solutions. They contain an optimized, expandable parallel CNN processing engine and efficient on-chip memory. The hardware accelerators can be integrated into embedded systems deployed at the edge or in servers in the data centers.

To get started developing CNN applications using GTI accelerators, we provide the following development toolsets:

  • Dev Hardware:  USB 3.0 sticks containing Lightspeeur® 2801 and 2803 accelerator chips. 
  • Model Development Kits (MDK): train and fine-tune CNN models optimized for GTI accelerators using Caffe, TensorFlow or PyTorch.
  • Software Development Kits (SDK): drivers, libraries, APIs and sample code to develop applications using GTI accelerators.

General Workflows

There are two typical general workflows:

1. Quick Evaluation


If you are just getting started and want to familiarize yourself with GTI’s accelerators, start by installing the SDK. Within the SDK there are a few out-of-the-box application demos with pre-trained models you can run to experience and familiarize yourself with our solutions.

Get Started with our SDK >

2. Typical Development Workflow


At a high level, a full development workflow utilizes the MDK to train and create models to work on GTI accelerators. Then with the SDK, a CNN application can be developed to use those trained models for inference on GTI accelerators.

Specifically, during your training phase, the MDK allows you to start training from an existing model or from scratch.  The MDK supports various popular AI frameworks to help you build machine learning models while sample pre-trained models are available to help shorten training time. 

When your network is ready for inferencing, you can deploy it onto a GTI accelerator. GTI also has different reference designs to help you prototype and build custom AI-based systems or integrate AI features into existing designs. 

GTI Plai™ Plug

The Plai™ Plugs along with the SDK, allows you to easily prototype and develop CNN solutions.

GTI now offers both 2801 and 2803 versions of the Plai™ Plug. Both contain the Lightspeeur accelerator along with various components managed by the control logic unit.

To use the Plai™ Plug, install the appropriate SDK and then connect the Plai™ Plug to your host machine via its USB port. The host machine contains the CPU that acts as the head unit to orchestrate the interactions with and computations of the Plai™ Plug accelerator.

Applications running on the host machine communicate with the accelerator through GTI APIs. When you initialize and open a device through the API, model files are transferred and loaded onto the Plai™ Plug. From there, the Plai™ Plug is ready to accept instructions to execute inferences.

GTI Architecture

Gyrfalcon accelerators support various interfaces including FTDI, eMMC and PCIe on different platforms to provide flexibility in your development.  The SDK provides the drivers needed to enable hardware integration to communicate with the host processor. It also has APIs to interface and communicate with the accelerator.