Appendix

Nvidia GPU architecture and installation mode supported by PaddlePaddle

Compile Dependency Table

Compile Option Table

BLAS

PaddlePaddle supports two BLAS libraries, MKL and OpenBlAS. MKL is used by default. If you use MKL and the machine contains the AVX2 instruction set, you will also download the MKL-DNN math library, for details please refer to here.

If you close MKL, OpenBLAS will be used as the BLAS library.

CUDA/cuDNN

PaddlePaddle automatically finds the CUDA and cuDNN libraries installed in the system for compilation and execution at compile time/runtime. Use the parameter -DCUDA_ARCH_NAME=Auto to specify to enable automatic detection of the SM architecture and speed up compilation.

PaddlePaddle can be compiled and run using any version after cuDNN v5.1, but try to keep the same version of cuDNN in the compiling and running processes. We recommend using the latest version of cuDNN.

Configure Compile Options

PaddePaddle implements references to various BLAS/CUDA/cuDNN libraries by specifying paths at compile time. When cmake compiles, it first searches the system paths ( /usr/liby and /usr/local/lib ) for these libraries, and also reads the relevant path variables for searching. Can be set by using the -D command, for example:

Cmake .. -DWITH_GPU=ON -DWITH_TESTING=OFF -DCUDNN_ROOT=/opt/cudnnv5

Note: The settings introduced here for these compilation options are only valid for the first cmake. If you want to reset it later, it is recommended to clean up the entire build directory ( rm -rf ) and then specify it.

You can find and download the corresponding PaddlePaddle-gpu release for your CUDA environment at the following official PaddlePaddle path:

• CUDA 11.8 Installation Package

• CUDA 12.6 Installation Package

• CUDA 12.9 Installation Package

• CUDA 13.0 Installation Package

Multi-version whl package list - Release

Execute the PaddlePaddle training program in Docker

Suppose you have written a PaddlePaddle program in the current directory (such as /home/work): train.py ( refer to PaddlePaddleBook to write), you can start the training with the following command:

cd /home/work

docker run -it -v $PWD:/work ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.3.0-gpu-cuda12.9-cudnn9.9 /work/train.py

In the above commands, the -it parameter indicates that the container has been run interactively; -v $PWD:/work specifies that the current path (the absolute path where the PWD variable in Linux will expand to the current path) is mounted to the :/work directory inside the container: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle specifies the container to be used; finally /work/train.py is the command executed inside the container, ie. the training program.

Of course, you can also enter into the Docker container and execute or debug your code interactively:

docker run -it -v $PWD:/work ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.3.0-gpu-cuda12.9-cudnn9.9 /bin/bash

cd /work

python train.py

**Note: In order to reduce the size, vim is not installed in PaddlePaddle Docker image by default. You can edit the code in the container after executing ** apt-get install -y vim (which installs vim for you) in the container.

Perform GPU training using Docker

In order to ensure that the GPU driver works properly in the image, we recommend using nvidia-docker to run the image. Don't forget to install the latest GPU drivers on your physical machine in advance.For specific driver version requirements, please refer to here

Nvidia-docker run -it -v $PWD:/work ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.3.0-gpu-cuda12.9-cudnn9.9 /bin/bash

Note: If you don't have nvidia-docker installed, you can try the following to mount the CUDA library and Linux devices into the Docker container:

export CUDA_SO="$(\ls /usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') \
$(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')"
export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
docker run ${CUDA_SO} \
${DEVICES} -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-gpu