Survey of Machine Learning Accelerators (2020-09-01T00:00:00.000000Z)

TL;DR

This paper collects and summarizes the current accelerators that have been publicly announced with performance and power consumption numbers and highlights interesting trends regarding power consumption, numerical precision, and inference versus training.

Abstract

New machine learning accelerators are being announced and released each month for a variety of applications from speech recognition, video object detection, assisted driving, and many data center applications. This paper updates the survey of of AI accelerators and processors from last year's IEEE-HPEC paper. This paper collects and summarizes the current accelerators that have been publicly announced with performance and power consumption numbers. The performance and power values are plotted on a scatter graph and a number of dimensions and observations from the trends on this plot are discussed and analyzed. For instance, there are interesting trends in the plot regarding power consumption, numerical precision, and inference versus training. This year, there are many more announced accelerators that are implemented with many more architectures and technologies from vector engines, dataflow engines, neuromorphic designs, flash-based analog memory processing, and photonic-based processing.

Authors

A. Reuther

1 papers

P. Michaleas

1 papers

Michael Jones

1 papers

TL;DR

Abstract

Authors

References132 items

Domain-specific hardware accelerators

Efficient Processing of Deep Neural Networks

Think Fast: A Tensor Streaming Processor (TSP) for Accelerating Deep Learning Workloads

AI Chips: What They Are and Why They Matter

Compute Solution for Tesla's Full Self-Driving Computer

A Survey of Accelerator Architectures for Deep Neural Networks

Habana Labs Purpose-Built AI Inference and Training Processor Architectures: Scaling AI Training Systems Using Standard Ethernet With Gaudi Processor

Digital neuromorphic chips for deep learning inference: a comprehensive study

A Survey and Taxonomy of FPGA-based Deep Learning Accelerators

Survey and Benchmarking of Machine Learning Accelerators

Towards artificial general intelligence with hybrid Tianjic chip architecture

[DL] A Survey of FPGA-based Neural Network Inference Accelerators

A new golden age for computer architecture

Deep Neural Network Approximation for Custom Hardware

A 68 Parallel Row Access Neuromorphic Core with 22K Multi-Level Synapses Based on Logic-Compatible Embedded Flash Memory Technology

A survey of FPGA-based accelerators for convolutional neural networks

A domain-specific architecture for deep neural networks

DLA: Compiler and FPGA Overlay for Neural Network Inference Acceleration

Programming Spiking Neural Networks on Intel’s Loihi

Mixed Precision Training

The microsoft catapult project

A survey of neural network accelerators

A Survey of Neuromorphic Computing and Neural Networks in Hardware

A survey of neural network accelerators

TETRIS: Scalable and Efficient Neural Network Acceleration with 3D Memory

Efficient Processing of Deep Neural Networks: A Tutorial and Survey

The End of Moore's Law: A New Beginning for Information Technology

Deep learning with coherent nanophotonic circuits

Eyeriss: A Spatial Architecture for Energy-Efficient Dataflow for Convolutional Neural Networks

An Analysis of Deep Neural Network Models for Practical Applications

Convolutional networks for fast, energy-efficient neuromorphic computing

EIE: Efficient Inference Engine on Compressed Deep Neural Network

Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks

TrueNorth: Design and Tool Flow of a 65 mW 1 Million Neuron Programmable Neurosynaptic Chip

PuDianNao: A Polyvalent Machine Learning Accelerator

Deep Learning with Limited Numerical Precision

DaDianNao: A Machine-Learning Supercomputer

A million spiking-neuron integrated circuit with a scalable communication network and interface

1.1 Computing's energy problem (and what we can do about it)

AHaH Computing–From Metastable Switches to Attractors to Machine Learning

ImageNet classification with deep convolutional neural networks

NeuFlow: A runtime reconfigurable dataflow processor for vision

SpiNNaker: Mapping neural networks onto a massively-parallel chip multiprocessor

Survey of neural network hardware

G ROQ R OCKS N EURAL N ETWORKS Startup Creates New Architecture: One Core, 1,000 TOPS

Digital Neural Network Accelerators

GrAI Matter research gives rise to AI processor for the edge

Kneron’s Next-Gen Edge AI Chip Gets $40m Boost

“FSD Chip - Tesla,”

GAP application processors -GreenWaves Technologies

GAP9 for ML at the Edge EEJournal

Horizon Robotics Journey2 Automotive AI Processor Series

Intel Lays Out Strategy For AI: It’s Habana

Arm Ethos is for Ubiquitous AI At the Edge — WikiChip Fuse

XMOS adapts Xcore into AIoT crossover processor’ — EE Times

“Dell DSS8440 Graphcore IPU Server,”

Low-Power AI Startup Eta Compute Delivers First Commercial Chips - IEEE Spectrum

Performance Estimation and Benchmarks for Real-World Edge Inference Applications

Deep dive into Amazon Inferentia: A custom-built chip to enhance ML and AI

“Ascend 910 AI Processor,”

“Deﬁning AI Innovation with NVIDIA DGX A100,”

“Tenstorrent Scales AI Performance: Architecture Leads in Data-Center Power Efﬁciency,”

Perceive Exits Stealth With Super Efficient Machine Learning Chip For Smarter Devices

The Silicon Photonics Key to Building Better Neural Networks

Startup Accelerates AI at the Sensor

Quadric Edge Supercomputer

Habana Wins Cigar for AI Inference

Blaize Fires up GSP for AI Processing

Details of Hailo AI Edge Accelerator Emerge

Habana Offers Gaudi for AI Training

Startup reveals prototype optical AI processor, says report

NovuMind: An Early Entrant in AI Silicon

Bill Gates just backed a chip startup that uses light to turbocharge AI

Bitmain SoC Brings AI to the Edge

Samsung, Toshiba Detail AI Chips

Qualcomm Targets AI Inferencing in the Cloud — EE Times