Network Design for AI training
Learn how to design networks for AI training! We cover why packet loss kills AI jobs, Infiniband vs Ethernet, Ultra Ethernet Consortium, Arista Etherlink, and network designs from 10s to 100k+ XPUs.
Topics Covered in This Video
Subscribe to watch and navigate by chapter
0:00
Introduction
1:15
AI training jobs and why packet loss is bad.
2:49
Infiniband vs Ethernet design options.
5:30
Ultra Ethernet Consortium
6:36
Arista Etherlink
9:24
Small AI network design(10’s of XPUs)
10:05
Medium sized AI network design(100’s of XPUs)
11:20
Large Scale AI network design (up to 100k+ XPUs)