Explore a technical talk from NVIDIA's Rob Davis that delves into the critical role of memory fabric technologies in modern data-centric and AI applications. Learn about three mature memory fabric solutions - Ethernet, InfiniBand, and NVLink - and understand their distinct performance characteristics and price points. Discover how Ethernet serves as a cost-effective solution with broad compatibility for general-purpose applications, while InfiniBand offers enhanced data throughput and reduced latency for mid-range performance needs in HPC and AI environments. Gain insights into NVLink's superior capabilities as the premium solution for demanding applications, particularly in advanced AI and machine learning workloads, delivering unmatched bandwidth and efficiency.
Memory Fabrics: Comparing Ethernet, InfiniBand, and NVLink Technologies
Open Compute Project via YouTube
Overview
Syllabus
Mature at Scale Memory Fabrics for all Performance and Price Points
Taught by
Open Compute Project
Related Courses
-
Exploring the Wilderness - Optimizing Ethernet Fabrics for AI Workloads
-
Ultra Ethernet Overview for AI and HPC Workloads
-
Networking for AI and HPC: Understanding Ultra Ethernet Technology
-
Source Routing for AI Fabrics - Optimizing Network Traffic in Multi-tenant AI Clusters
-
Fabric Attached Memory: Hardware and Software Architecture for HPC Systems
-
Fabric Attached Memory: Hardware and Software Architecture for HPC Systems
