Intel has officially launched its Neural Compute Stick 2, a low-cost plug-and-play USB-connected deep-learning accelerator based on the company’s Movidius Myriad X vision processing unit (VPU) and boasting an eightfold performance improvement over the previous generation.
Based on the same design as its predecessor, the Neural Compute Stick, Intel’s Neural Compute Stick 2 swaps out the Movidius Myriad vision processing unit (VPU) for its successor the Myriad X – a chip which is claimed to boost performance eightfold for deep neural network (DNN) processing. Internally, the device features 16 programmable streaming hybrid architecture vector engine (SHAVE) cores with a high-throughput memory fabric interconnect.
Intel is supporting the device, as with its previous releases, via a distribution of the OpenVINO toolkit which it claims speeds the development of computer vision applications through the inclusion of pre-trained models, optimised algorithms, and sample code. The company is also claiming that its application programming interface (API) offers “write once, deploy everywhere” development, supporting use on the Neural Compute Stick 2, its predecessor, and non-VPU hardware including CPUs, graphics processors, and field-programmable gate arrays (FPGAs).
More information on the Neural Compute Stick 2 can be found on the official website.
Gareth Halfacree is a technology journalist and technical author best known for his work on the Raspberry Pi User Guide and an upcoming book covering the BBC’s micro:bit project. A keen open source advocate and RISC-V fan, Gareth is responsible for curating our fortnightly community round-up posts.