跳过导航至主内容
Zhenyu He · Jobs Stroustrup
English 中文
Skip to article

研究笔记 · Research Note

高性能计算 HPC

Zhenyu He · Jobs Stroustrup 阅读约 1 分钟

高性能计算 HPC (High Performance Computing)

熟练度

熟练

描述

  • 大涡模拟 (LES) 运行与调试（DAM 模型），单次运行约 46k core-hours
  • 代码扩展：添加诊断输出变量、新的 forcing 逻辑、I/O 节奏优化
  • 计算资源规划：在算力预算内设计可执行的敏感性实验方案
  • 理想化水球模拟 (aquaplanet simulations)

🌟 研究级 signature capabilities

PKU 本科 HPC infra mastery (2020-2021)

  • Tianhe-II NSCC-GZ: 3 accounts ((undergraduate Tianhe-II account) primary + 2 更多 Walker variants) yhrun -p work -n 48 48-PE SLURM (~本科生少见)
  • CAM3 aqua-planet 30-yr spin-up runs × 10 cases (D60-180 + Wonly + Conly + Aqua, 每 case 30 year integration)
  • 27,882 MATLAB+NCL analysis lines + 5-stage HPC pipeline architecture (CAM3 aqua-planet → Tianhe-II runs → 25 cal_X MATLAB diagnostic → NCL stream function + ω analysis → Python figure polish)

UC Berkeley PhD HPC mastery (2022-2025)

  • Lawrencium LBNL cluster: 8 production runs × ~46k core-hours each ≈ 44 days HPC wall time 工程规划
  • DAM quasi-2D arrays Lx 20km-50000km scan (Phase 4.5 killer figure convergence)
  • 115,363 Jupyter notebook lines / 62 ipynb / 3,793 code cells aggregate (PhD analysis code)

Cross-language HPC tooling

  • Fortran 77/90: NCAR CAM3 v3.1 source tree (615 .F90 files) + DAM LES source (40+ modules) 级 source code 理解
  • Python/MATLAB/R/NCL/IDL/Mathematica: 跨语言 scientific computing (Caltech DSCOVR 5-language fluency)
  • remote workstation CUDA bootstrap (Caltech, Windows CUDA incompatibility → manual CUDA driver + Anaconda bootstrap)

方法论 concept 页

在哪些经历中用到