paddlepaddle-gpu安装

置顶 DYF-AI

已于 2024-11-18 12:00:52 修改

阅读量4.7k

点赞数 12

文章标签： paddlepaddle 人工智能

于 2024-04-20 10:01:44 首次发布

本文链接：https://blog.csdn.net/weixin_40437821/article/details/137989723

版权

本文详细描述了解决在Windows和Linux环境下安装PaddlePaddleGPU时遇到的问题，包括推荐使用CUDA11.8，提供详细的安装步骤，并针对Linux中出现的Segmentationfault错误给出了修复方法。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

背景

之前安装paddlepaddle-gpu遇到各种问题，安装不成功，之前使用了wsl+docker的方式，可查看我之前博客：记录paddlepaddle-gpu安装，使用paddle-docker会导致我整个开发流程比较割裂，因此还是需要wsl中安装paddlepaddle-gpu，

cuda版本

强烈推荐cuda11.8，paddlepaddle对cuda版本要求及其严格，如12.1和12.0的差异都不能正常安装成功，如果需要同时使用pytorch和paddle，目前为止，只能安装cuda11.8：

安装

1.下载cuda11.8：
https://developer.nvidia.com/cuda-11-8-0-download-archive?target_os=Windows&target_arch=x86_64&target_version=11&target_type=exe_local

2.然后直接安装，安装教程就不赘述，一路下一步即可

3.下载对应版本的cudnn
在这里插入图片描述
https://developer.nvidia.com/rdp/cudnn-archive

4.拷贝

下面的路径自己修改即可，仅供演示
复制前要确保拷问文件夹权限（不然就算拷贝也会报错）：
在这里插入图片描述

（1）M:\software\cuda\cudnn-windows-x86_64-8.9.7.29_cuda11-archive\bin 下的文件全部拷贝到 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin
（2）M:\software\cuda\cudnn-windows-x86_64-8.9.7.29_cuda11-archive\include 下的文件全部拷贝到 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\include
（3）M:\software\cuda\cudnn-windows-x86_64-8.9.7.29_cuda11-archive\lib\x64 下的文件全部拷贝到 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\lib\x64

ubuntu:https://blog.csdn.net/enjoyyl/article/details/134893210

sudo chmod 777 -R include/
sudo chmod 777 -R lib64/

cp -r /mnt/n/software/cuda/cudnn-linux-x86_64-8.9.7.29_cuda11-archive/include/cudnn*.h /usr/local/cuda-11.8/include/

cp -r /mnt/n/software/cuda/cudnn-linux-x86_64-8.9.7.29_cuda11-archive/lib/libcudnn* /usr/local/cuda-11.8/lib64/
sudo chmod a+r /usr/local/cuda-11.8/lib64/libcudnn*
sudo chmod a+r /usr/local/cuda-11.8/include/cudnn*.h

5.安装paddlepaddle-gpu

https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/windows-pip.html

pip install paddlepaddle-gpu==2.6.1 -i https://pypi.tuna.tsinghua.edu.cn/simple

6.验证

在这里插入图片描述

备注：博客仅作为我个人文档记录，如觉得写的不好，还请谅解。

linux报错：

>>> import paddle
>>> paddle.utils.run_check()
Running verify PaddlePaddle program ...
I0528 18:27:59.434978  1489 program_interpreter.cc:212] New Executor is Running.
W0528 18:27:59.435155  1489 gpu_resources.cc:119] Please NOTE: device: 0, GPU Compute Capability: 8.9, Driver API Version: 12.3, Runtime API Version: 11.8
W0528 18:27:59.435673  1489 gpu_resources.cc:164] device: 0, cuDNN Version: 8.9.
W0528 18:27:59.669302  1489 dynamic_loader.cc:314] The third-party dynamic library (libcuda.so) that Paddle depends on is not configured correctly. (error code is libcuda.so: cannot open shared object file: No such file or directory)
  Suggestions:
  1. Check if the third-party dynamic library (e.g. CUDA, CUDNN) is installed correctly and its version is matched with paddlepaddle you installed.
  2. Configure third-party dynamic library environment variables as follows:
  - Linux: set LD_LIBRARY_PATH by `export LD_LIBRARY_PATH=...`
  - Windows: set PATH by `set PATH=XXX;


--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
No stack trace in paddle, may be caused by external reasons.

----------------------
Error Message Summary:
----------------------
FatalError: `Segmentation fault` is detected by the operating system.
  [TimeInfo: *** Aborted at 1716892079 (unix time) try "date -d @1716892079" if you are using GNU date ***]
  [SignalInfo: *** SIGSEGV (@0x0) received by PID 1489 (TID 0x7f3411621740) from PID 0 ***]

解决：https://blog.csdn.net/weixin_45921929/article/details/128853445

(torch) dongyongfei786@DESKTOP-3QNKDLL:~$ sudo find /usr/ -name libcuda.so*
/usr/lib/wsl/drivers/nv_dispig.inf_amd64_8f4dab92e290c42d/libcuda.so.1.1
/usr/lib/wsl/lib/libcuda.so
/usr/lib/wsl/lib/libcuda.so.1
/usr/lib/wsl/lib/libcuda.so.1.1
/usr/local/cuda-11.6/targets/x86_64-linux/lib/stubs/libcuda.so
/usr/local/cuda-11.8/targets/x86_64-linux/lib/stubs/libcuda.so
/usr/local/cuda-11.3/targets/x86_64-linux/lib/stubs/libcuda.so
(torch) dongyongfei786@DESKTOP-3QNKDLL:~$ cd /usr/l
lib/     libexec/ local/
(torch) dongyongfei786@DESKTOP-3QNKDLL:~$ sudo ln -d /usr/local/cuda-11.8/targets/x86_64-linux/lib/stubs/libcuda.so /usr/lib/x86_64-linux-gnu/libcuda.so
(torch) dongyongfei786@DESKTOP-3QNKDLL:~$ python
Python 3.8.13 | packaged by conda-forge | (default, Mar 25 2022, 06:04:18)
[GCC 10.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import paddle
>>> paddle.utils.run_check()
Running verify PaddlePaddle program ...
I0528 19:49:38.749671  1566 program_interpreter.cc:212] New Executor is Running.
W0528 19:49:38.749820  1566 gpu_resources.cc:119] Please NOTE: device: 0, GPU Compute Capability: 8.9, Driver API Version: 12.3, Runtime API Version: 11.8
W0528 19:49:38.750424  1566 gpu_resources.cc:164] device: 0, cuDNN Version: 8.9.
I0528 19:49:41.227173  1566 interpreter_util.cc:624] Standalone Executor is Used.
PaddlePaddle works well on 1 GPU.
PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.
>>>