Skip to content

Docker Installation (SM75 GPU)

Nvidia-RTX20-Series-SM75

Suitable for users with a GPU who want to leverage GPU acceleration. sm-75 refers to the hardware platform architecture number, applicable to Nvidia RTX 20 series GPUs. To run the program correctly, the host machine must have the NVIDIA Driver installed, with a version that supports CUDA 12.4. Run the following command in the terminal to check if the GPU driver is ready:

nvidia-smi
  • step 0 Pull the Docker base image

For users in China, it is recommended to pull the base image from a domestic mirror first: 

sudo docker pull docker.m.daocloud.io/nvidia/cuda:12.4.1-devel-ubuntu22.04
sudo docker tag docker.m.daocloud.io/nvidia/cuda:12.4.1-devel-ubuntu22.04 nvidia/cuda:12.4.1-devel-ubuntu22.04

  • step 1 Enable CUDA and configure SM

Create a config.mk.local file in the project root directory (not inside the docker_install folder). CUDA is enabled by default, so you only need to configure SM: 

rm -f config.mk.local
echo "CUDA_ARCH := sm_75" >> config.mk.local

  • step 2 Build the Docker image
# a. Using Tsinghua mirror (recommended for users in China):
docker build -t psum:nvidia-sm75 \
  --build-arg MIRROR_URL=//mirrors.tuna.tsinghua.edu.cn \
  -f docker_install/dockerfile_cuda_12_4 .

# b. Using the official source:
docker build -t psum:nvidia-sm75 -f docker_install/dockerfile_cuda_12_4 .
  • step 3 Start the container

Must add the --gpus all flag to mount the physical GPU: 

docker run -it --gpus all --name psum_gpu_dev -v $(pwd):/workspace psum:nvidia-sm75 bash

  • step 4 Compile inside the container

After entering the container terminal, execute the following standard procedure: 

bash env_scan.sh
source env_load.sh
bash build.sh

Build results:

  • Field solver: CPU ✅ | Multigrid ✅ | Direct ✅ | GPU ✅
  • Pypsum serialization: ✅ Build succeeded | ✅ Python interaction test passed

Since -v $(pwd):/workspace is used, code modifications made via VS Code on the host machine will be synchronized inside the container, and make or test executions will take effect accordingly.

After configuration, you should be able to correctly compile and run PSuM code inside the container.

Finally, add the contents of env_load.sh from the project folder to the ~/.bashrc file inside the container, so that environment variables are automatically loaded when entering the container.

Later, to re-enter the container: 

docker start -i psum_gpu_dev

# Load environment variables inside the container
source env_load.sh