*本課程僅開放給有意願成為學生叢集競賽隊員或教練的學生。
<br>
<br>
<br>教學方式:
<br>  討論、講習、實驗操作、口頭報告
<br>
<br>教學進度:
<br>  MPI 進階知識: MCA,arch,UCX,RDMA
<br>  Cluster 系統調控: Prometheus,Grafana,ELK
<br>  Application Bnechmark: IO500,MLPerf,osu_micro_benchmark
<br>  Application Profile方法學: Amdahl's Law,Roofline model,TMA
<br>  Application Profile 實作: CPU Perf,Vtune,Branch prediction,Vector instruction,etc
<br>  Profiler: GPU Nsigh System, Ucx_info, Iostat, Rdma ib command
<br>  Case Study: AI (GPU) 計算的介紹,平行化與效能優化,Deepspeed,tensorRT,Quantization
<br>  Power management: BIOS,IPMI,fan control
<br>
<br>成績考核:
<br>  每周進度報告: 40%
<br>  實驗操作: 20%
<br>  期末計畫: 40%