An in-person
afternoon event in
Boston

There seems to be an awful lot of new technology lately, but is it really new or just the next logical step in the evolution and intertwining of compute and fabric? This talk sets the stage for the rest of the afternoon. We look at how hardware has been evolving to help us see where it goes next.

This session includes a number of short talks that address the challenges of selection, provision, sharing, and scheduling, GPU resources. The gist is that traditional HPC batch job scheduling is inefficient for AI/ML. We need new methods for sharing hardware and scheduling work.

AI/ML is the hottest field of research computing, both literally and figuratively. On the literal front, we are at the point where traditional air cooling will no longer cut it. This session frames the problems and looks at some new, affordable, and practical cooling solutions.

The last thing you want is to keep those expensive GPUs of yours waiting for storage to deliver. Fortunately, the same kinds of technologies that enable the modern GPU and HPC data center also take storage to the next level. In this session we will hear perspectives from NVIDIA, Weka, and Vast - three of the industry leaders in storage for AI/ML.

Do we really need another processing unit, or are DPUs at worst just decorations on the NIC card, or at best glorified offload engines? We will try to get close to this by asking some of the pioneers in this space. We are lucky to have Fungible, AMD, and Nvidia on hand to share their respective product visions.

Do you ever dream about conjuring up bespoke physical servers from components that are sprinkled across your data fabric? No? Well, it’s time to start because composable infrastructure is an actual thing now and there are practical applications. Our panelists come from three leaders in the field: Liquid, Fungible, and GigaIO.