Intel unveils AI accelerator capable of Trillion operation per second

Aug. 30, 2017, 11:38 a.m. By: Pranjal Kumar

Intel AI

Intel has released its new Movidious Myriad X Vision processing unit which will advance the Intel’s end to end portfolio of artificial Intelligence solution. This processing unit will deliver more autonomous capabilities across a wide range of product categories including Virtual Reality, drones and much more.

Intel’s new Myriad X Vision processing unit is world’s first system-on-chip with the dedicated Neural Compute Engine for accelerating deep learning inferences at the edges. The Neural compute engine is a hardware block which is specially designed to run deep neural networks at high but with low power consumption that too without compromising accuracy. It will enable the devices to understand and respond to their environments in real time. Another plus point of Myriad X is that it can perform 1 Trillion operations per second (on deep neural network inferences).

It merges three architectural elements to provide high performance.

  1. Array of programmable VLIW vector processors with an instruction set tuned to computer vision and deep learning workloads.

  2. Collection of hardware accelerators supporting image signal processing.

  3. Commonly accessible intelligent memory fabric that minimizes data movement on chip.

Remi El-Ouazzane, vice president of Movidious (Intel New Technology Group) said,” Computer Vision and deep learning will become a standard requirement for the billions of devices surrounding us. The next thing in the computing world is to enable the devices with human-like visual intelligence.

Myriad X will be redefining what a VPU means when it comes to delivering as much AI and vision compute power possible.”

Besides the dedicated Neural Compute Engine, Myriad X also uniquely combines deep learning inference in real time with many other like

1) Programmable 128-bit VLIW Vector Processors:

Users can run multiple imaging and vision application pipelines simultaneously (with the flexibility of 16 vector processors).

2) Vision Accelerators:

One can use more than 20 hardware accelerators to perform tasks like stereo depth.

3) 2.5 MB of Homogenous On-Chip Memory:

The 450 GB per second of internal bandwidth, minimizing latency and reducing power consumption by minimizing off-chip data transfer.

Myriad X is specially built for embedded visual intelligence and inference.