The Future of HPC Systems: Towards Large-Scale Time-Slice Execution and Adaptive Accelerator Allocation

This talk will discuss challenges in High-Performance Computing (HPC) systems, such as long waiting times and low resource utilization due to batch queue systems, and underutilization of GPUs due to fixed allocation. The speaker will introduce two solutions: "Interactive HPC" and "AI Computing Broker". Interactive HPC, used in large-scale clusters, eliminates unnecessary waiting times and idle nodes, thereby improving system utilization and user experience. AI Computing Broker enhances GPU utilization by dynamically allocating GPUs only when necessary, thus improving efficiency.