Empowering the Research and Development of Taiwan through HPC and AI

In this regional update, Wei-Cheng Huang of Taiwan's National Center for High-performance Computing (NCHC) reviews how the center's mission and infrastructure have evolved under rapid AI growth. NCHC operates both GPU systems, including Nano5 and the newer four-nanometer-based Nano4 entering production, and CPU machines that academia continues to demand for traditional numerical simulation, producing an alternating pattern of GPU and CPU procurements. CPU workloads are dominated by environment and disaster mitigation, particularly climate, weather, and typhoon ensemble simulation, and by chemistry, while over sixty percent of GPU usage is AI-related. Huang describes a sharp budget increase against nearly flat staffing, new data center construction including a facility with a PUE of 1.3 and a future machine room designed for 90 megawatts, and funding through the Taiwan Chiplet-based Industry Innovation Program and the Southern Taiwan Silicon Valley Project targeting 480 petaflops by the end of 2029. Additional initiatives include an alliance with private-sector compute providers for overflow capacity, machine-learning-based prediction of system load with particle swarm optimization of queue configurations yielding around 21 percent performance improvement, backbone network upgrades toward 1.6 terabits, and interest in quantum communication. On the AI side, the center offers the Taiwan AI RAP platform with training, API, and inference tiers, a locally built OmniGPT service, earlier federated learning across four Taipei medical centers, and an experimental agentic workflow that automatically assembled and ran a 2D CFD airfoil simulation.