Features: Low consumption and stable performance.notes: 1. Compatibility: Can be used with H2, H3, H5, H6, A64 Orange Pi development boards.2. Provide model conversion tools to support caffe-based convolutional neural network (CNN) model decomposition.3. The training tool can directly apply the original , video, voice, natural language rapid training prototype to the neural network computing stick.Model: for Orange Pi AI Stick LiteAI Processor:Chip Series: LightspeeurChip model: SPR2801sEnergy efficiency: 9.3TOPs/wattUltra-low consumption: 2.8TOPs@300mW: 5.6TOPs@100MHzHardware interface: SDIO3.0/eMMC 4.5Chip package: BGA (7mm*7mm)Manufacturing process: 28nmInterface standard: USB2.0, USB3.0 Type-ATransmission bandwidth: read bandwidth=68.00MB/s, write bandwidth=84.69MB/sOpen source frameworks: Caffe, Tensorflow, PyTorch (Pytorch only supports classification tasks for the time being)Supported models: VGG, SSDOperating System: X86 Linux (Ubuntu 16.04) ARM Linux/ ARM (ARM v7, ARM v8)Working voltage: DC 5V 200mAOperating temperature: 0C to 40CStorage temperature: -20C to 80CDimensions: 66.5mm x 20.5mm x 10.8mmSoftware provided: (based on Pytorch's "full-stack" framework)Software provided: Open ARM, X86 SDK, built-in 1 to 2 model demos; model training tool Plait.Material: ABScolour: WhitePackage Contents:1 x for Orange Pi AI Stick LiteOnly the above package content, other products are not included.Note: Light and different displays may cause the color of the item in the picture a little different from the real thing. The measurement allowed error is +/- 1-3cm.