Intel introduced the Movidius Neural Computing Stick, the world’s first USB-based depth learning reasoning tool, and an independent artificial intelligence (AI) accelerator. It provides dedicated depth neural network processing for a wide range of edge host devices. The sleek Movidius Neural Computing Stick is designed for product developers, researchers and founders to provide dedicated high-performance deep neural network processing performance to reduce the time spent on development, tuning, and deployment of artificial intelligence applications.
Movidius Neural Compute Stick
Movidius Neural Compute Stick is the world’s first USB-based depth learning plug and play tool and an independent AI accelerator. As more developers start using advanced machine learning solutions to develop innovative applications and solutions, Intel is committed to providing the most comprehensive development tools and resources to ensure that developers using artificial intelligence get some assistance. Training of Artificial Intelligence Neural Networks on Intel® Nervana Clouds and the use of Intel® Xeon® Scalable processors for artificial intelligence, virtual reality (VR) and enhanced reality (AR) and unmanned (AD) Field, or through the Movidius Visual Processing Unit (VPU) technology will make artificial intelligence more accessible. Intel offers a full range of artificial intelligence tools, training and deployment options for next generation products and services that incorporate artificial intelligence.
The Myriad 2 VPU built into the Movidius TM neural stick provides a powerful and efficient performance – it can deliver more than 100 billion power per second, according to Remi El-Ouazzane, vice president and general manager of Movidius (Intel Corp.) The performance uses floating point operation, in order to run the real time deep neural network (DNN) directly on the device, which allows a variety of artificial intelligence applications to be deployed offline.
The development of machine intelligence mainly includes two stages: (1) training the algorithm with a large number of sampling data through modern machine learning technology; (2) running the algorithm in terminal applications that need to interpret real-world data. Among them, the second stage is also called “reasoning”. The reasoning is used in reducing the delay, power consumption and privacy which brings many benefits. Here are some components:
- Compiler: Automatically transforms a trained, Caffe-based convolution neural network (CNN) into an embedded neural network that is optimized to run on the Movidius Myriad 2 VPU.
- Tuning: Layer-by-layer performance metrics for industry standards and custom neural networks enable efficient tuning in low-power conditions for optimal performance in the real world. The validation script allows the developer to compare the accuracy of the optimized model with the original PC model on the device.
- Acceleration: As a unique feature of the Movidius TM neural stiick, the device can be used as a single neural network accelerator to improve performance and energy efficiency by adding dedicated depth learning reasoning capabilities to existing computing platforms.
The Movidius Neural Compute Stick is now available for sale at a retail price of $ 79 and is available for purchase at the Computer Vision and Pattern Recognition (CVPR) conference in Honolulu, Hawaii, from 22 to 25 July. Computer Vision and Pattern Recognition (CVPR) conference is an annually held global conference on computer vision for academics and industry researchers.