R-CCS Cafe 第224回 併設 特別セミナー

AI/ML workloads (specially deep learning) are outpacing Moore's law, in both computational requirements and the number of users. This rapid development, increased scale, and widening adoption in the field of AI necessitate continuous foundational contributions from the HPC community to accommodate AI workloads. In this talk we give an overview of recent and on-going efforts dedicated to optimize and scale AI workloads on HPC systems, across the entire stack, from hardware co-design to applications. More specifically, we highlight cases of how "HPC in AI" help in understanding and alleviating the bottlenecks of AI workloads running on HPC systems.