Intel announces AI-centric Movidius Myriad X VPU, promises more autonomy for embedded AI solutions
Intel has just released Movidius Myriad X vision processing unit (VPU). It’s an AI-centric, system-on-chip (SoC) featuring a dedicated Neural Compute Engine that is “designed to design to run deep neural networks at high speed and low power...." As deep learning platforms require inference processes that are fast and accurate, according to Intel, the Movidius Myriad X VPU is capable of delivering over 4 TOPS (trillion operations per second) of total performance.
Therefore, the new Intel VPU can be applied across a myriad of product categories that include drones, robotics, smart cameras and virtual reality. Such end devices can leverage on trained, deep-learning models to process new data and act upon it, with speed and accuracy. The Intel Movidius Myriad X VPU is a result of the chip maker’s purchase of computer vision startup Movidius in September 2016.
The system-on-chip has sixteeen, 128-bit VLIW programmable vector processors that are optimized for computer vision workloads. Up to 8 HD resolution sensors can be connected to the VPU, due to the presence of its 16 MIPI lanes. The VPU supports an instruction set, which is “tuned to computer vision and deep learning workloads.” The VPU has enhanced computer accelerators, and 2.5MB of homogenous on-chip memory.
The Intel Movidius Myriad X VPU is targeted at autonomous AI end devices, due to its compact form factor and powerful on-board deep neural network (DNN) inference processing capabilities.