1 服务器环境

功能	参数/工具名称	版本号
操作系统	CentOS 7.7	内核linux mstation 3.10.0-1062.18.1.el7.x86 64
CPU	12vCPU	-
内存	128GB	-
GPU 2卡	NVIDIA GP100GL Quadro	-

• 查看 硬盘信息确定安装路径

df -h
fdisk -l

• 后台运行生成文件 python x.py &>> xxx.out 参考https://askubuntu.com/questions/420981/how-do-i-save-terminal-output-to-a-file

访问方法

• SSH： putty
• SFTP：fileZilla

2 安装基础应用

2.1 运行环境安装参考

Python 3.8

下载链接：https://www.python.org/downloads/release/python-3811/ 安装方法：https://packaging.python.org/guides/installing-using-linux-tools/

Anaconda

https://docs.anaconda.com/anaconda/install/linux/ 可在conda上建立多个环境，以便用于不同的模型训练

NVIDIA CUDA-enabled

CUDA Installation Guide for Linux 参考 https://docs.nvidia.com/cuda/archive/10.2/cuda-installation-guide-linux/index.html

cuDNN

用于深层神经网络的gpu加速库 下载： https://developer.nvidia.com/rdp/cudnn-download 安装：https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html

• https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#installlinux Note: the tar file installation applies to all Linux platforms.
• get following installation from https://developer.nvidia.com/rdp/cudnn-download https://developer.nvidia.com/compute/machine-learning/cudnn/secure/8.2.4/10.2_20210831/cudnn-10.2-linux-x64-v8.2.4.15.tgz

wget https://developer.nvidia.com/compute/machine-learning/cudnn/secure/8.2.4/10.2_20210831/cudnn-10.2-linux-x64-v8.2.4.15.tgz
tar -xzvf cudnn-10.2-linux-x64-v8.2.4.15.tgz

2.2 安装过程命令

• Conda 2021.05
• Python 3.8.11
• CUDA 10.2

• KB:custom install path only for shell file:

sudo sh cuda_10.2.89_440.33.01_linux.run  --silent \
                --toolkit --toolkitpath=/home/hpc/cuda/toolkit \
                --samples --samplespath=/home/hpc/cuda/samples \ 
                --defaultroot=/home/hpc/cuda --tmpdir=/home/hpc/cuda/tmp --installpath=/home/hpc/install  --extract=/home/hpc/extract   --no-man-page   --kernel-source-path=/home/hpc/kernel --librarypath=/home/hpc/lib 

refer to:

1. How do I install the Toolkit in a different location? https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#faq1
2. how can I install CUDA on linux on a custom path (not in /usr/local) https://superuser.com/questions/1131813/how-can-i-install-cuda-on-linux-on-a-custom-path-not-in-usr-local

• download specified version, search in https://developer.nvidia.com/ e.g. https://developer.nvidia.com/cuda-10.2-download-archive?target_os=Linux&target_arch=x86_64&target_distro=CentOS&target_version=7&target_type=runfilelocal

• cuDNN 8.1.1.33_CUDA11.2

• 验证cuDNN正确安装和运行 https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#verify

3 路径设置

3.1 备用文件

• /home/hpc/files

3.2 Conda 环境

• tsai
• cnnlstmdense

3.3 代码

• /home/hpc/cnnlstmdense

• data
• models
• outputs

• /home/hpc/tsai

• data
• models
• output

4 运行报告

编号	运行时长（分钟或者小时)	epochs	batch_size	模型和运行参数	评价指标：正确率%等	GPU	cuDNN