ARISE

Abstract

Why this tutorial matters

The rapid growth of foundation models and large-scale AI systems has transformed modern computing infrastructure. As model sizes and cluster scales continue to grow, system performance is increasingly constrained not by compute alone, but by communication, memory capacity, memory bandwidth, collective operations, and interconnect fabrics. Emerging technologies such as Ultra Ethernet, InfiniBand, NVLink, UALink, CXL-based memory expansion, and memory-semantic fabrics are redefining the architecture of AI clusters and introducing new research challenges spanning systems, networking, and computer architecture.

This tutorial introduces simulation-driven methodologies for studying and designing next-generation AI infrastructure. Participants will learn the fundamentals of distributed AI training and inference, understand the role of communication and memory bottlenecks, and gain hands-on experience using Astra-Sim and related simulation frameworks. The tutorial further covers emerging research directions in AI fabrics, memory disaggregation, large-scale model serving, and memory-semantic networking.

The tutorial combines conceptual foundations, practical demonstrations, and guided hands-on exercises, enabling researchers and practitioners to evaluate AI infrastructure designs without requiring access to expensive production-scale clusters.

Tutorial Schedule

Agenda

Exact session timings are TBD and subject to final confirmation. Additional presenters may be included depending on the final tutorial team composition.

Tutorial Website

Materials & Resources

Upon acceptance, this site will host the full set of tutorial materials below. All materials will remain publicly available after the conference to maximize community impact.