机器学习环境搭建-硬件优化

使用 CPU GPU 加速等紧密贴合硬件的库时，务必查看官方文档的硬件要求

在测试或搭建硬件优化库时，务必备份原环境，以及及时的备份数据

1. Intel CPU + Intel GPU

• Installing Intel® Distribution for Python* and Intel® Performance…: https://www.intel.com/content/www/us/en/developer/articles/technical/using-intel-distribution-for-python-with-anaconda.html

# 添加 intel channel
conda config --add channels intel
# 核心发行版
conda create -n idp intelpython3_core python=3.x -c intel
# 完整发行版
conda create -n idp intelpython3_full python=3.x -c intel

2. Nvidia GPU

# 尽量在最初同时安装 cudnn cudatoolkit
# 后期再安装 cudnn 会比较麻烦
# 只安装 cudatoolkit 后期再添加 cudnn 会存在版本匹配等问题
mamba create -n lab-01 python=3.8 cudnn cudatoolkit

3. Intel CPU

# 一般各大框架都会默认使用 mkl（所以 intel cpu 一般不需要手动调整）
mamba create -n lab-01 python=3.8 mkl

4. AMD CPU

BLAS、OpenBLAS、ATLAS、MKL_openblas mkl - CSDN博客: https://blog.csdn.net/fuhanghang/article/details/122885265

暂时没有找到针对 AMD CPU 优化的库，ACML 也停止更新了，如果有关于 AMD CPU 的优化需求可以参考下面两篇文章，删除 MKL 限制 或 使用 OpenBLAS

• Daniël de Kok: Intel MKL on AMD Zen: https://danieldk.eu/Posts/2020-08-31-MKL-Zen.html
• AMD CPU 做科学计算还香吗？（2022 年 MKL 与 OpenBLAS 的 NumPy 简单测试）: https://xcel.me/amd-cpu-numpy-mkl-vs-openblas-2022/

• Let’s Build Everything - AMD GPUOpen: https://gpuopen.com/
• ACML Downloads & Resources - AMD: https://developer.amd.com/acml-downloads-resources/
• Best Processors for Machine Learning https://www.sabrepc.com/blog/Deep-Learning-and-AI/best-processors-for-machine-learning

5. 参考

• Anaconda | Open Sourcing Anaconda Accelerate: https://www.anaconda.com/blog/open-sourcing-anaconda-accelerate
• Anaconda | Getting Started with GPU Computing in Anaconda: https://www.anaconda.com/blog/getting-started-with-gpu-computing-in-anaconda
• 关于 scikit-learn-intelex 提速: Anaconda | Scikit-learn Speed-up with Intel and Anaconda: https://www.anaconda.com/blog/scikit-learn-speed-up-with-intel-and-anaconda
• Intel® Extension for Scikit-learn— Intel(R) Extension for Scikit-learn 2021.5 documentation: https://intel.github.io/scikit-learn-intelex/
• 关于 Modin 提速 pandas: Anaconda | Scale your pandas workflow with Modin – no rewrite required: https://www.anaconda.com/blog/scale-your-pandas-workflow-with-modin
• Installation — Modin documentation: https://modin.readthedocs.io/en/latest/getting_started/installation.html
• Intel® Distribution of Modin: https://www.intel.com/content/www/us/en/developer/tools/oneapi/distribution-of-modin.html
• Anaconda | Getting Started with GPU Computing in Anaconda: https://www.anaconda.com/blog/getting-started-with-gpu-computing-in-anaconda