GPGPU 2020 (FING)

Tareas del curso de Computación de Propósito General en Unidades de Procesamiento Gráfico 2020, Facultad de Ingeniería (UdelaR)

• Práctico 1 - Impacto del acceso a los datos en CPU
• Práctico 2 - Primeros pasos con CUDA
• Práctico 3 - Blur y ajustar brillo en imágenes

Setup Environment

Prerequisites

• Recommended Drivers of your NVIDIA video card: sudo ubuntu-drivers autoinstall
• Nvidia Container Runtime:
  • Add the repository following the instructions
  • After doing sudo apt-get update, install the Container Runtime: sudo apt-get install nvidia-container-runtime

Build the image and name it gpgpu

docker image build -t gpgpu .

Run the container

docker run -it --gpus all --network host -v $(pwd):/workspace/src gpgpu

Remove dangling images

Useful command to free up space docker rmi -f $(docker images -f "dangling=true" -q)

Cluster

Connect

ssh gpgpu_6@login.cluster.uy

Add volume (Linux)

• Open Files
• Click on + Other Locations
• Connect to Server inputing the address: ssh://gpgpu_6@login.cluster.uy/clusteruy/home/gpgpu_6

You may relocate proyect files to this new volume in order to access them on the cluster

Run job

• Alter launch_single.sh to your liking
• sbatch launch_single.sh
• You can also swap the last line for $1 $2 $3 and run sbatch launch_single.sh ./program arg1 arg2 (in case enqueuing with multiple scripts/args is desired)