---
tags: Cluster Contest
---

# SC23 寒假作業

> [mail=[e-mail]]
> [name=翁君牧]

[toc]

# Compile WRF
## module load GCC ==(5%)==

- Check `module list`
- 

## Install HDF5 + NetCDF ==(15%)==

- Check `which ncdump`
- 
- Report (Why use this flag?)
- 編譯 NetCDF Fortran 時需要加上 `--disable-shared`, 因為 NetCDF C 也是 static 的，否則 HDF5 的參數也要修改

## Get WRF.exe ==(20%)==

- Check `tail -15 build_wrf.log`
- 

---

# Run WRF:
## Just run on headnode ==(5%)==

- Check Screenshot
- 

## Single Node with Slurm ==(15%)==

- Check `./validate.csh ./VALIDATION ./VALIDATE`
- 

## Optimization(Multiple Node, AVX512...) ==(20% ⬆️)==

- Check `./timing.csh TIMING/rsl.out.0000`
- 
- Report (What do you do?)
- Compile with `-march=cascadelake -mtune=native` to enable _AVX512_ and other optimizations & run on 4 nodes with 224 tasks in partition ct224

---

# Others
## Profile WRF ==(15%)==

- Check Vtune Screenshot
- 
- 
- Report (Based on the results, What are your assumptions?)
- 上圖是單一節點中 56 cores 的長條圖，下圖為所有節點共 224 cores 的長條圖。可以發現大部分的時候每個節點依然只有一個核心被利用到。

## Display output ==(5%)==

- Check Screenshot
- 
- 我是使用在 local 以 `brew` 安裝的 `ncview` 查看 `scp` 下來的結果

## 其他你想告訴我們的

- 繳交前：
- [x] 填寫進度追蹤表
- [x] 刪除沒完成的分數