AI Engine Architecture: Data Movement, Synchronization, Reconfiguration and Application Mapping
Offered By: Scalable Parallel Computing Lab, SPCL @ ETH Zurich via YouTube
Course Description
Overview
Explore the AI Engine (AIE) architecture developed by AMD/Xilinx in this 54-minute conference talk by David Clarke. Gain insights into the 2D array of vector processors that form part of the Xilinx Versal 7nm family and next-gen AMD APU devices. Delve into the intricacies of the architecture, including its VLIW SIMD vector processors, scratchpad memory, data movement engines, and streaming interconnect. Learn about target applications such as machine learning inference in datacentre, automotive, edge, and 5G acceleration. Discover key aspects of data movement, synchronization, reconfiguration, and application mapping onto hardware. Recorded at SPCL_Bcast on December 15, 2022, this talk offers a comprehensive overview of AIE architecture and its potential impact on high-performance computing.
Syllabus
Intro
Talk
Announcements
Taught by
Scalable Parallel Computing Lab, SPCL @ ETH Zurich
Related Courses
Stanford Seminar - MIPS Open, Wave ComputingStanford University via YouTube Loop Analysis and Vectorization in Julia - JuliaCon 2020
The Julia Programming Language via YouTube Intrinsic Functions and Vector Processing Extensions for SIMD Parallel Operations in C++
javidx9 via YouTube Intrinsics - Low-Level Engine Development with Burst - Unite Copenhagen
Unity via YouTube Aggregating Ticks to Manage Scale in Sea of Thieves - Unreal Fest Europe 2019 - Unreal Engine
Unreal Engine via YouTube