Can Ethernet meet all the interconnect challenges for AI and ML?
In this video, Moray McLaren, Principal Engineer from Google, discusses:
– Ethernet has displaced many application-specific networks, but its suitability for ML depends on node characteristics
– Google’s TPU systems use a proprietary network based on Ethernet, incorporating features similar to Ultra Ethernet
– While Ethernet is likely the future for scale-out and host networks, latency-sensitive architectures may require alternative solutions for scale-up networks
Want to be involved our video series? Contact [email protected]
Check out full showcase at: https://ngi.fyi/tef24yt to get executive perspectives from the show floor of the Ethernet Alliance’s Technology Exploration Forum 2024
