awesome-edge-ai 
A curated list of edge devices for AI applications.
Contents
Development Platforms
- Baidu EdgeBoard - Xilinx FPGA based edge computing board.
- BeagleBone AI - Linux Board powered by ARM CPU,TI C66x digital-signal-processor (DSP) cores and embedded-vision-engine (EVE). It has 1GB RAM.
- Google Coral Dev Board - It is single board solution which contains ARM CPU and Edge TPU as accelerator with 1GB RAM.
- Google Coral USB Accelerator - Edge TPU accelerator module connects with USB interface. It needs host device.
- Intel Neural Compute Stick 1 - First generation ASIC chip designed by Movidius that runs as module board via USB interface. It needs host device.
- Intel Neural Compute Stick 2 - Second generation module stick which is faster than previous one based TOPS measure. It needs host device.
- Nvidia Jetson AGX Xavier - Most power single board computer from NVIDIA designed for autonomous systems (cars etc.). Contains 512 Cuda core Volta GPU with Tensor Cores and 16 GB RAM.
- Nvidia Jetson Nano - Cost efficient alternative to NVIDIA boards. Performance is equivalent to the TX1. Contains ARM CPU and 128 Cuda core Maxwell GPU and has 4 GB RAM.
- Nvidia Jetson TX1 - Contains ARM CPU and 128 Cuda core Maxwell GPU.
It seems like this board is replaced by Nano. - Nvidia Jetson TX2 - Contains ARM CPU and 256 Cuda core Pascal GPU and has 8 GB/4 GB RAM.
- OrangePi AI Stick - ASIC based neural accelerator. It needs host device.
- Sipeed MAIX Go Suit - Single board computer based on RISC-V AI chip, KPU(Neural Network Processor) and APU(Audio Processor). It contains on-board DVP camera and LCD screen for visual output.
- SmartEdge Agile - Modular Edge Device with IOT capabilities.
- SparkFun Edge Development Board - Apollo3 Blue microcontroller based board that contains microphones and 3-axis accelerometer and OV7670 camera interface. Runs on single coin battery. Supports Tensorflow-lite.
- UP Squared AI Vision X - Single board computer contains Intel ATOM CPU and Movidius Myriad as neural accelerator. It has 4 GB RAM.
- Xnor.ai Solar Powered Module - Very low power unit which can be running with only using solar panels.
More technical specifications needed.
Custom Accelerators
- Google Edge TPU
- Intel Movidius Myriad VPU 2
- Intel Movidius Myriad X VPU
- Kendryte K210
- Lightspeeur 280-X Neural Accelerators
- Sipeed MAIX
- TI Vision AcclerationPac
Software Libraries
Contribute
Contributions welcome! Read the contribution guidelines first.
License
To the extent possible under law, Refik Can MALLI has waived all copyright and related or neighboring rights to this work. See LICENSE.