Intel Movidius Myriad-2 12 x 128-bit VLIW SHAVE Vector Processor Neural Compute Stick Development Board NCSM2450.DK1

Intel NCSM2450.DK1 Neural Compute Stick Development Board integrates the Movidius Myriad-2 Vision Processing Unit (VPU) operating at 600 MHz with twelve 128-bit VLIW SHAVE vector processors and 2MB of fast 400Gbps on-chip memory. Designed as a USB 3.0 stick, this development board enables developers to prototype and deploy deep neural networks efficiently without expensive supercomputer hardware and without cloud dependency.

The device supports floating point (FP16, FP32) and integer operations at multiple bit widths (8, 16, 32 bits), facilitating a broad range of machine learning models. The included SDK allows seamless conversion of trained Caffe-based and TensorFlow convolutional neural networks (CNNs) into embedded forms optimized for the on-board Myriad-2 VPU.

Developers can tune models using layer-wise performance metrics and validate performance against PC-based models to maximize efficiency and accuracy. The stick's compact dimensions of 72.5 x 27 x 14 mm provide portability for development and deployment.

Powered by a 600MHz Movidius Myriad-2 SoC with 12 vector processors. 2MB 400Gbps on-chip memory ensuring rapid data transfer. Supports FP16, FP32, and integer operations at multiple accuracies. Single USB 3.0 port delivers power and data connectivity.

Enables local, real-time deep learning inference without cloud. Compatible with Caffe and TensorFlow frameworks via SDK. Compact form factor for versatile development environments. Ideal for applications in smart home automation, robotics, surveillance, retail analytics, and healthcare diagnostics.

This Development Board is essential for industrial, laboratory, and commercial AI research and development, offering high computational performance in an energy-efficient, portable design. Explore more Intel development kits and AI acceleration tools at EnvMart.