Intel Neural Compute Stick was first introduced in early 2017 as a USB compute that allows AI inference at the edge with low power consumption. The stick is based on Myriad Movidius 2 VPU (Vision Processing Unit), and was found to significantly improve inference performance on Raspberry Pi 3 board over a proprietary GPU accelerated solution.
However, a little later last year, Intel also announced Movidius Myriad X VPU with claims of up to 10 times DNN performance over Myriad 2 VPU. But so far, we would only see solutions launched with the latter, and it looks like Intel is finally ready to bring Myriad X VPU to the market with the company announcing Intel Neural Compute Stick 2 at Intel’s artificial intelligence (AI) developer conference in Beijing taking place on November 14 and 15.
- Processor – Intel Movidius Myriad X Vision Processing Unit (VPU) with 16 SHAVE cores (128-bit VLIW Vector Processors)
- Host Interface – USB 3.0 Type-A port
- Dimensions – 72.5 mm x 27 mm x 14 mm
- Operating temperature – 0° C to 40° C
The Neural Compute Stick 2 works with Ubuntu 16.04.3 LTS (64 bit), CentOS 7.4 (64 bit), and Windows 10 (64 bit), supports TensorFlow and Caffe frameworks, and development can be done with Intel’s OpenVINO toolkit.
The company is not claiming 10 times the performance anymore, but NCS 2 has still be tested to outperform the first generation Intel Neural Compute Stick by about 8 times for image classification and object detection workloads. There’s no information about power consumption that I could find but it’s still low power with the company showing it used on drones. Just like with Movidius NCS, you can combine multiple NCS 2 sticks on a USB 3.0 hub for even better performance.