CUDA安装

ybinu

已于 2024-06-18 21:12:39 修改

阅读量441

点赞数 1

文章标签：深度学习人工智能

于 2023-07-20 22:50:19 首次发布

本文链接：https://blog.csdn.net/ybinu/article/details/131841667

版权

LINUX

111111111111111111111111111111111111111111111111驱动
https://www.nvidia.com/download/index.aspx?lang=en-us 纯驱动
https://developer.nvidia.com/cuda-toolkit-archive
https://developer.nvidia.com/cuda-10.2-download-archive
https://developer.nvidia.com/cuda-11-8-0-download-archive
https://developer.nvidia.com/cuda-downloads?target_os=Windows

带cudakit要查询需要注册

https://blog.csdn.net/weixin_58283091/article/details/133349022对应关系
https://developer.nvidia.com/rdp/cudnn-archive

/usr/local/cuda/

wget https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux.run

https://www.nvidia.cn/download/driverResults.aspx/182489/cn/
apt list --installed
https://blog.csdn.net/Lucifer_zzq/article/details/88997018

https://developer.nvidia.com/cuda-11.3.0-download-archive?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=deb_local

wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run

=====

安装驱动相关
https://www.linuxidc.com/Linux/2017-12/149577.htm
安装 nvidia-driver 驱动，可以顺带安装 nvidia-dkms ，方便后续如果需要升降级内核的时候，减少不必要的麻烦：
https://zhuanlan.zhihu.com/p/616208372
nvidia-smi

2222222222222222222222222222222222222222222222222conda 导入
参考这里：https://editor.csdn.net/md/?articleId=129908570

conda env create -f freeze.yml
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch

333333333333333333333333333333333333333333333333cudatoolkit

sudo apt-get install nvidia-driver-515

测试

nvcc -V
cd /usr/local/cuda-11.4/samples/4_Finance/BlackScholes
make BlackScholes
./BlackScholes

pip install numba==0.56.4 numpy==1.23.4

pip install --upgrade setuptools
安装微软的VS Install，把你的操作系统包和改装的库装上：
pip install nes-py --no-cache-dir
pip install --upgrade pip setuptools wheel

444444444444444444444444444444444444444444444444444cub
并设置好CUB_HOME。

curl -LO https://github.com/NVIDIA/cub/archive/1.10.0.tar.gz
tar xzf 1.10.0.tar.gz
vim ~/.bashrc
export CUB_HOME=/mnt/cub-1.10.0
source ~/.bashrc

python3 -c 'import torch ;print(torch.cuda.is_available() );print (torch.cuda.get_device_name(0))'
python3 -c 'import torch ;print(torch.cuda.is_available())'

import torch
torch.cuda.is_available()       # cuda是否可用
torch.cuda.current_device()     # 返回当前设备索引
torch.cuda.device_count()       # 返回GPU的数量
torch.cuda.get_device_name(0)   # 返回gpu名字，设备索
import torchvision

55555555555555555555555555555555555555555555555555555pip
protobuf3.20.0
transformers4.27.1
pip install basicsr --use-pep517 --verbose
pip install torch1.6.0+cu101 torchvision0.7.0+cu101 torchaudio0.6.0 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch1.10.1+cu102 torchvision0.11.1+cu102 torchaudio0.10.1 -f https://download.pytorch.org/whl/cu102/torch_stable.html

39
https://download.pytorch.org/whl/torch_stable.html
pip install https://download.pytorch.org/whl/cu118/torch-2.0.1%2Bcu118-cp39-cp39-win_amd64.whl
pip install https://download.pytorch.org/whl/cu118/torchvision-0.15.2%2Bcu118-cp39-cp39-win_amd64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple some-package
pip install https://download.pytorch.org/whl/cu118/torchaudio-2.0.2%2Bcu118-cp39-cp39-win_amd64.whl

https://pytorch.org/get-started/previous-versions/

66666666666666666666666666666666666666666666666环境

vim ~/.bashrc
export CUB_HOME=/home/yb/cub-1.8.0

export TORCH_CUDA_ARCH_LIST="6.0+PTX" #设置计算能力
alias condax='conda activate xuniren'
alias condapy39='conda activate py39'

export PATH=usr/local/cuda-10.2/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64:$LD_LIBRARY_PATH
export PATH=$PATH:/opt/miniconda3/bin/

export PATH=/usr/local/cuda-11.3/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.3/lib64:$LD_LIBRARY_PATH
source ~/.bashrc

7777777777777777777777777777777777777777777777777777问题

The detected CUDA version (10.1) mismatches the version that was used to compile
PyTorch (11.3). Please make sure to use the same CUDA versions.

cuda11.8
namespace “thrust::cub” has no member “DebugSyncStream”

gcc 版本
https://huaweicloud.csdn.net/638071e7dacf622b8df884df.html
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-7 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 50

sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-7 100
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-9 50

sudo update-alternatives --config gcc

如何检查版本匹配：

check A

nvcc -V

…
Cuda compilation tools, release 10.1, V10.1.243

check B

nvidia-smi

…
CUDA Version: 12.1

check C

python3 -c 'import torch ;print(torch.cuda.is_available() );print (torch.cuda.get_device_name(0))'
python3 -c 'import torch ;print(torch.cuda.is_available())'

check D

$ conda list | sed -n “3p” && conda list | grep -i -E “cuda|torch”

Name Version Build Channel

cudatoolkit 10.2.89 hfd86e86_1 defaults
ffmpeg 4.3 hf484d3e_0 pytorch
pytorch 1.9.1 py3.8_cuda10.2_cudnn7.6.5_0 pytorch
torchaudio 0.9.1 py38 pytorch
torchvision 0.10.1 py38_cu102 pytorch

#check E
$ pip list | grep -i -E “cuda|torch”
nvidia-cuda-cupti-cu11 11.7.101
nvidia-cuda-nvrtc-cu11 11.7.99
nvidia-cuda-runtime-cu11 11.7.99
torch 2.0.1
torchvision 0.15.2

export PYTHONNOUSERSITE=True之后，看起来问题解决了。
参考：https://github.com/conda/conda/issues/7707

关于deb包的安装和删除
http://www.360doc.com/content/21/0301/09/7467661_964529101.shtml

sudo chmod a+x cuda_10.0.130_410.48_linux.run
运行cuda文件：
sudo ./cuda_10.0.130_410.48_linux.run

WINDOWS

sudo apt-get update
sudo apt install gcc-9 g+±9
ls /usr/bin/gcc*
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-1 200
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 150
sudo update-alternatives --config gcc

apt list --installed

sudo cp -P cudnn-linux-x86_64-8.5.0.96_cuda11-archive/lib/libcudnn* /usr/local/cuda-11.8/lib64/
sudo cp cudnn-linux-x86_64-8.5.0.96_cuda11-archive/include/cudnn.h /usr/local/cuda-11.8/include/
sudo chmod a+r /usr/local/cuda-11.8/include/cudnn.h
sudo chmod a+r /usr/local/cuda-11.8/lib64/libcudnn*

ins-quupycr0hk.qq.ybinu.cn
https://download.pytorch.org/whl/torch-2.0.1-cp310-cp310-manylinux2014_aarch64.whl

https://download.pytorch.org/whl/cu118/torch-2.0.1%2Bcu118-cp310-cp310-linux_x86_64.whl
https://download.pytorch.org/whl/cu118/torch-2.0.0%2Bcu118-cp310-cp310-win_amd64.whl

python
import torch
print(torch.cuda.is_available())

cuda主要有以下参数：
–no-opengl-libs：表示只安装驱动文件，不安装OpenGL文件。必需参数，原因同上。注意：不是-no-opengl-files。
–uninstall (deprecated)：用于卸载CUDA Driver（已废弃）。
–toolkit：表示只安装CUDA Toolkit，不安装Driver和Samples
–help：查看更多高级选项。
nvidia-smi

http://www.dtmao.cc/java/83525.html

select vdisk file=“C:\Users\Administrator\AppData\Local\Packages\CanonicalGroupLimited.Ubuntu22.04LTS_79rhkp1fndgsc\LocalState\ext4.vhdx”
compact vdisk
detach vdisk

sudo docker pull nvcr.io/nvidia/l4t-pytorch:r35.2.1-pth2.0-py3
https://blog.csdn.net/hymnal/article/details/126960005
docker run --gpus all -it --rm nvcr.io/nvidia/pytorch:22.08-py3