It’s been a busy conference season, with the AI Hardware and Edge AI Summit, Yotta 2024, and OCP’s Global Summit all taking place in the past month or so. The OCP Global Summit has become a personal favorite of mine; the diversity of presenters and industry verticals is unmatched, along with more focus on technical and engineering talks, rather than sales pitches.
As in previous years, I’ve been reflecting on the conference and poring over the 22 tracks and more than 430 presentations — thousands of hours of content!
Having spent the week prior at Yotta in Vegas, it was great to delve deeper into the engineering and technical depth in the workshops and expo hall, where OCP really shines.
Attendance this year jumped from 4,500 to 7,500, and booths in the expo went from 70 to 100. It was BIG, and while there’s plenty that I find interesting, I'm merely scratching the surface on topics and depth.
The themes, like many conferences in the past two years, were AI with generous servings of power, heat, and cooling/sustainability.
Given the wealth of content, I’ve decided to share my reflections across two articles.
In this first part, I’ll share my initial impressions of the keynotes, and in the next, I’ll dive deeper across power challenges, cooling updates, and (for me) highly anticipated interconnectivity news, specifically UALink and UEC (Ultra Ethernet Consortium).
Keynotes Highlights
Meta’s Advancements in AI Infrastructure
Omar Baldonado from Meta kicked things off sharing their latest AI infrastructure initiatives. Meta is harnessing AMD’s MI300X inference systems, built on their Grand Teton platform. They’re also introducing a new NIC ASIC in collaboration with Marvell, aiming to boost network performance and efficiency. On the networking front, they’re utilizing 51.2 Tbps switches powered by Broadcom’s Tomahawk and Cisco technologies, coupled with lossless fabric solutions from Arista Networks and Broadcom’s Ramon and Jericho chips (respectively). It’s clear that Meta is pushing the envelope to support the massive computational demands of AI workloads, and awesome they continue to share their experiences with the wider OCP community.