Speaker: Jérémie Leguay (Nokia Bell Labs)
Abstract
Network acceleration is critical for modern AI workloads, particularly in optimizing communication and data transfer among distributed GPUs. The network has emerged as a significant bottleneck for Large Language Model (LLM) training and execution, driving rapid advancements in GPU network technologies. This evolution, however, introduces complex research challenges spanning load balancing, congestion control, collective communications, and novel network architectures. This presentation will first establish the fundamental distinctions between AI data centers and conventional cloud computing environments, followed by an overview of ongoing research topics in this rapidly evolving field.
