Intel unveils Neural Compute Engine for AI applications

Posted Sep 11, 2017 by Tim Sandle
Imagine a tiny chip that can put the power of human vision into every device. The launch of the Neural Compute Engine in Intel’s new Movidius Myriad X VPU moves this closer to unleashing visual intelligence in drones and robots.
The Intel logo is displayed outside of the Intel headquarters in Santa Clara  California
The Intel logo is displayed outside of the Intel headquarters in Santa Clara, California
Justin Sullivan, Getty/AFP/File
With the unveiling of the Neural Compute Engine in Intel’s new Movidius Myriad X VPU, Intel has stated they are one step closer to bringing true visual intelligence in drones, robotics, cameras, wearables and smart home solutions to the fore. The dedicated Neural Compute Engine is designed to give machines the ability to see, understand and react to their environments in real time.
This concept of high-performance, low-power artificial intelligence is set to improve the future capabilities of devices. The video below explains more about the device capabilities:
Myriad X’s tiny form factor and on-board processing have been designed for autonomous devices. The main features include a proprietary Neural Compute Engine, which is a dedicated hardware block designed for accelerating neural network inferences with more than 1 TOPS of performance. TOPS refers to ‘trillion operations per second’, and this is a standard performance measure for the chip based on peak floating-point computational throughput of Neural Compute Engine. The chip is designed to require only minimum power. The TOPS concept is an extension of the standard floating point operations per second (FLOPS), the industry accepted measure of computer performance. This measure is useful in fields of scientific computations that require floating-point calculations.
Other features include the ability to run multiple simultaneous neural networks for enhanced understanding. This will enable devices to react more autonomously to their surroundings. Furthermore, there are sixteen vector processors optimized for computer vision workloads and with support capability for up to eight high definition camera inputs. This enables an equipped device to handle 700 million pixels per second of image signal processing throughput. There are also over twenty hardware accelerators which undertake tasks like optical flow and stereo depth. In all there is 2.5 MB of on-chip memory.
The chip represents an important offering for designers and developers of artificial technology systems. According to Remi El-Ouazzane, who is the vice president and general manager of Movidius, Intel New Technology Group: “We’re on the cusp of computer vision and deep learning becoming standard requirements for the billions of devices surrounding us every day. Enabling devices with humanlike visual intelligence represents the next leap forward in computing.”
He adds: “With Myriad X, we are redefining what a VPU means when it comes to delivering as much AI and vision compute power possible, all within the unique energy and thermal constraints of modern untethered devices.” This makes the chip particularly suited for sustained high-performance on deep learning and computer vision workloads.