Heterogeneous Serverless Computing for High-Performance Computing and AI Workloads
Scalable Parallel Computing Lab, SPCL @ ETH Zurich via YouTube
Overview
Explore the future of high-performance computing in this 47-minute conference talk by Dejan Milojicic at SPCL_Bcast. Delve into the concept of Heterogeneous Serverless Computing (HSC) and its potential to revolutionize HPC, HPDA, and AI workloads. Learn about the convergence of traditional simulation with data analytics and AI/ML, the increasing efficiency of heterogeneous hardware, and the growing demand for flexible delivery models. Discover how HSC aims to combine the efficiency of heterogeneity, the productivity of serverless computing, and the granularity of Functions-as-a-Service to create a new architecture for modern computing challenges. Gain insights into innovative uses of accelerators, such as SmartNICs and Fabric Attached Memories, and their impact on application performance and hardware efficiency. Understand the implications of HSC for edge-to-exascale deployment and its potential to create a more maintainable, scalable, and reusable development model in the evolving landscape of high-performance computing.
Syllabus
Intro
Talk
Taught by
Scalable Parallel Computing Lab, SPCL @ ETH Zurich