Alibaba Cloud Linux with pre-installed NVIDIA GPU drivers

Supported instance families

• gn7e, gn7s, gn7i, gn6v, gn6i, gn6e, gn5, and gn5i

• ebmgn7e, ebmgn7i, ebmgn6v, ebmgn6i, and ebmgn6e

• ebmgn7ix, ebmgn7ex

• gn8is, ebmgn8is, gn8v, and ebmgn8v

Environment variables

• /etc/profile.d/nccl.sh

  export NCCL_HOME=/usr/local/nccl
  export LD_LIBRARY_PATH=${NCCL_HOME}/lib:$LD_LIBRARY_PATH

• /etc/profile.d/openmpi.sh

  export MPI_HOME=/usr/local/openmpi
  export LD_LIBRARY_PATH=${MPI_HOME}/lib:$LD_LIBRARY_PATH
  export PATH=${MPI_HOME}/bin:$PATH

• /etc/profile.d/cuda.sh

  export PATH=/usr/local/cuda/bin:$PATH
  export CUDA_HOME=/usr/local/cuda
  export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

Supported instance families

• ebmgn9g, ebmgn9gc, ebmgn9ge, ebmgn9t, gn9g, gn9t, and gn9ge

• gn8t, gn8te, ebmgn8t, ebmgn8te, ebmgn8ts, and gn8ep

Enable the keentune tool

Enable keentune with the following commands. Restart the operating system for changes to take effect.

systemctl stop tuned
systemctl disable tuned
systemctl start keentune-target
systemctl enable keentune-target
systemctl enable keentuned
systemctl start keentuned
keentune profile set ai_common.profile

To disable keentune, run keentune profile rollback and restart the operating system.

What should I note when using an Alibaba Cloud Linux image with pre-installed NVIDIA GPU drivers in an ACK cluster?

See How to create a custom image based on an existing ECS instance and use the image to create nodes and Usage notes and instructions for risky operations in the Container Service for Kubernetes documentation.

Switch CUDA versions

Run the nvcc command to check the current CUDA version.

nvcc --version

The image uses update-alternatives to manage CUDA versions. To switch from CUDA 13.0 to 12.8, use either of the following methods:

• Run update-alternatives --config cuda to switch interactively.

  update-alternatives --config cuda
  There are 2 choices for the alternative cuda (providing /usr/local/cuda).
  
    Selection    Path                  Priority   Status
  ------------------------------------------------------------
    0            /usr/local/cuda-13.0   20        auto mode
    1            /usr/local/cuda-12.8   10        manual mode
  * 2            /usr/local/cuda-13.0   20        manual mode
  
  Press <enter> to keep the current choice[*], or type selection number: 1  --> Enter the number for the version you want to use.
  update-alternatives: using /usr/local/cuda-12.8 to provide /usr/local/cuda (cuda) in manual mode

• Use the --set option to set the CUDA version directly.

update-alternatives --set cuda /usr/local/cuda-12.8