Seminario: “Enabling Portable Collective Communication on Heterogeneous GPU Systems” di Salvatore Sirica e “Vettorizzazione Esplicita del Prodotto Matriciale su RISC-V con RVV” di Sergio Guastaferro
Speaker 1: Salvatore SiricaTitolo: "Enabling Portable Collective Communication on Heterogeneous GPU Systems" di Salvatore Sirica Abstract:Many distributed HPC applications and data-parallel AI training pipelines depend on MPI-style communication patterns. As systems scale to many GPUs and nodes, collective operations like Allreduce become a primary performance bottleneck, directly impacting scalability. In practice, the highest-performing GPU collectives […]
