cuda docker containers

# gpu driver
nvidia-smi
# docker 19.03+
docker --version

# add nvidia apt key
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
  sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg

# add repository
curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
  sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
  sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

# install toolkit
sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit

# configure docker
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

# configure for user
nvidia-ctk runtime configure --runtime=docker --config=$HOME/.config/docker/daemon.json

# critical: disable cgroups
sudo nvidia-ctk config --set nvidia-container-cli.no-cgroups --in-place

# restart user docker
systemctl --user restart docker

docker run --rm --gpus all nvidia/cuda:12.8.0-base-ubuntu22.04 nvidia-smi

+-------------------------------------------------------------------------+
| NVIDIA-SMI 575.57.08    Driver Version: 575.57.08    CUDA Version: 12.9|
+-------------------------------------------------------------------------+

# all gpus
docker run --gpus all myimage

# specific gpu
docker run --gpus '"device=0"' myimage

# multiple gpus
docker run --gpus '"device=0,1"' myimage

version: '3.8'
services:
  ml-app:
    image: ml:latest
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

# limit visible gpus
docker run -e NVIDIA_VISIBLE_DEVICES=0 myimage

# capabilities
docker run -e NVIDIA_DRIVER_CAPABILITIES=compute,utility myimage

# pytorch
docker run --gpus all pytorch/pytorch:2.7.1-cuda12.8-cudnn8-runtime

# tensorflow
docker run --gpus all tensorflow/tensorflow:latest-gpu

# jax
docker run --gpus all ghcr.io/nvidia/jax:base

FROM nvidia/cuda:12.8.0-runtime-ubuntu22.04

WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt

COPY . .
CMD ["python", "app.py"]

FROM ghcr.io/astral-sh/uv:latest as uv
FROM nvidia/cuda:12.8.0-runtime-ubuntu22.04

# install minimal deps
RUN apt-get update && apt-get install -y \
    curl \
    ca-certificates \
    && apt-get clean \
    && rm -rf /var/lib/apt/lists/*

# copy uv
COPY --from=uv /uv /uvx /bin/

# install python
RUN uv python install 3.13

WORKDIR /app
COPY pyproject.toml uv.lock ./

# install deps with cache
RUN --mount=type=cache,target=/root/.cache/uv \
    UV_LINK_MODE=copy \
    uv sync --frozen --no-install-project

COPY . .
RUN --mount=type=cache,target=/root/.cache/uv \
    uv sync --frozen

CMD ["uv", "run", "python", "main.py"]

# build stage
FROM nvidia/cuda:12.8.0-devel-ubuntu22.04 as builder

WORKDIR /build
COPY . .
RUN make

# runtime stage
FROM nvidia/cuda:12.8.0-runtime-ubuntu22.04

COPY --from=builder /build/app /usr/local/bin/
CMD ["app"]

FROM ghcr.io/astral-sh/uv:latest as uv
FROM nvidia/cuda:12.8.0-runtime-ubuntu22.04

COPY --from=uv /uv /uvx /bin/
RUN apt-get update && apt-get install -y curl ca-certificates \
    && apt-get clean && rm -rf /var/lib/apt/lists/*

RUN uv python install 3.13

# pytorch-specific index
ENV UV_INDEX_URL=https://download.pytorch.org/whl/cu128

WORKDIR /app
CMD ["uv", "run", "--with", "torch", "python", "-c", \
     "import torch; print(f'cuda available: {torch.cuda.is_available()}')"]

FROM tensorflow/tensorflow:latest-gpu

# already includes cuda/cudnn
WORKDIR /app
COPY . .
CMD ["python", "train.py"]

FROM nvidia/cuda:12.8.0-cudnn9-runtime-ubuntu22.04

RUN pip install --upgrade "jax[cuda12]"

# cache python packages
RUN --mount=type=cache,target=/root/.cache/pip \
    pip install -r requirements.txt

# cache uv packages
RUN --mount=type=cache,target=/root/.cache/uv \
    uv sync --frozen

# bad: changes invalidate cache
COPY . .
RUN pip install -r requirements.txt

# good: deps cached separately
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .

# limit gpu memory growth
docker run -e TF_FORCE_GPU_ALLOW_GROWTH=true tensorflow/tensorflow:latest-gpu

# pytorch memory fraction
docker run -e PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512 pytorch

# real-time monitoring
docker exec -it container_id nvidia-smi

# continuous monitoring
docker exec -it container_id watch -n 1 nvidia-smi

services:
  dcgm-exporter:
    image: nvidia/dcgm-exporter:latest
    environment:
      - DCGM_EXPORTER_NO_HOSTNAME=1
    devices:
      - /dev/nvidiactl
      - /dev/nvidia0
    volumes:
      - /run/nvidia:/run/nvidia:ro

# check runtime config
docker info | grep nvidia

# inspect container
docker inspect container_id | grep -i gpu

# test minimal cuda
docker run --rm --gpus all nvidia/cuda:12.8.0-base-ubuntu22.04 \
  bash -c "echo 'int main() { return 0; }' > test.cu && nvcc test.cu && ./a.out"

# old
nvidia-docker run image

# new
docker run --gpus all image