AI Optimized Hardware

• AI-optimized hardware is a technology that makes hardware much friendlier. These are new graphics and central processing units and processing devices that are specifically designed and structured to execute AI-oriented tasks.

• AI-optimized hardware is primarily used in making a difference in deep learning applications.

• Some of the companies that are offering AI-optimized hardware are Google, IBM, Intel, Nvidia, Alleviate, and Cray.

• An AI accelerator is a class of microprocessor or computer system designed to accelerate artificial neural networks, machine vision and other machine learning algorithms for robotics, internet of things and other data-intensive or sensor-driven tasks.They are often many core designs and generally focus on low-precision arithmetic, novel dataflow architectures or in-memory computing capability. A number of vendor-specific terms exist for devices in this space.

• Computer systems have frequently complemented the CPU with special purpose accelerators for specialized tasks, most notably video cards for graphics, but also sound cards for sound, etc. As Deep learning and AI workloads rose in prominence, specialized hardware units were developed or adapted from previous products to accelerate these tasks.

• Graphics processing units or GPUs are specialized hardware for the manipulation of images. As the mathematical basis of neural networks and image manipulation are similar, embarrassingly paralleltasks involving matrices, GPUs became increasingly used for machine learning tasks. As such, as of 2016 GPUs are popular for AI work, and they continue to evolve in a direction to facilitate deep learning, both for training and inference in devices such as self-driving cars.and gaining additional connective capability for the kind of dataflow workloads AI benefits from (e.g. Nvidia NVLink).As GPUs have been increasingly applied to AI acceleration, GPU manufacturers have incorporated neural network specific hardware to further accelerate these tasks.Tensor cores are intended to speed up the training of neural networks.

• Deep learning frameworks are still evolving, making it hard to design custom hardware. Reconfigurable devices like field-programmable gate arrays (FPGA) make it easier to evolve hardware, frameworks and software alongside each other.

      AI accelerating co-processors:

• The processor in Qualcomm's mobile platform Snapdragon 845 contains a Hexagon 685 DSP core for AI processing in camera, voice, XR and gaming applications

• PowerVR 2NX NNA (Neural Net Accelerator) is an IP core from Imagination Technologies licensed for integration into chips.

• Neural Engine is an AI accelerator core within the Apple A11 Bionic SoC

• Cadence Tensilica Vision C5 is a neural networks optimized DSP IP core

• The Neural Processing Unit is a neural network accelerator within the HiSilicon Kirin 970.

           If you enjoyed this blog post, share it with a friend!      

          Next week I will post more about AI...so stay tuned!