Overview of Advanced Compute Images

Artificial intelligence (AI), machine learning (ML), and high performance computing (HPC) workloads often require you to create a custom image that includes additional software packages and is tuned for optimal performance. You can now use Advanced Compute Images to set up an optimized environment for your AI/ML or HPC workloads.

Advanced Compute Images provide a standardized image stack for AI/ML and HPC infrastructure. These images can eliminate the need to manually build your own custom images for AI/ML and HPC workloads.

Benefits

Advanced Compute Images provide the following benefits:

  • Compute instances ready for AI/ML or HPC workloads out-of-the-box. There is no need to manually tune performance, manage instance reboots, install Slurm agents, or stay up to date with the latest Cloud de Confiance updates for tightly coupled AI/ML or HPC workloads.
  • Consistent, reproducible performance. Image standardization gives you consistent, reproducible application-level performance.

Types of Advanced Compute Images

Advanced Compute Images contain multiple layers of configurations to support running AI/ML and HPC workloads.

  • Operating system configurations: Includes disabling automatic updates, setting HPC-optimized ulimits, tuning kernel sysctl parameters, and configuring security settings like SELinux.
  • Network tuning: Contains the necessary scripts for network tuning, such as enabling multi-queue service.
  • Cloud de Confiance by S3NS integration: Installs and configures Google Cloud CLI and the Ops Agent.
  • Container tools: Installs and configures all container-related software, including:
  • NVIDIA: Installs the following tools:
  • MPI: Installs the Message Passing Interface (MPI) libraries. Installs Open MPI on Ubuntu-based images.
  • Slurm: Installs the core components required to create a Slurm cluster node, including:
  • Filesystem clients: Installs client-side tools for accessing various filesystems, such as Cloud Storage FUSE, NFS utilities (nfs-utils), and the Lustre client.
  • Developer tools: Installs common development and debugging tool chains and utilities, such as:
  • Environment management: Installs and configures environment management tools, primarily Lmod, and sets up the necessary profile scripts to make the module command available to users.

Supported hardware and image families

Advanced Compute Images support GPU workloads and have the following characteristics:

  • Are optimized for NVIDIA accelerators
  • Support Ubuntu LTS 22.04 and Ubuntu LTS 24.04
  • Come with pre-integrated CUDA and NVIDIA drivers
  • Are suitable for AI, ML, or HPC workloads
Hardware OS Orchestrator Features Architecture Supported machine series Image family
GPU Ubuntu LTS 22.04 Slurm 25.11
  • CUDA 13.0
  • NVIDIA 580
 AMD x86_64 A3 Ultra, A4 aci-gpu-u2204-slurm-2511-cuda-130-nvidia-580-amd64
GPU Ubuntu LTS 24.04 Slurm 25.11
  • CUDA 13.0
  • NVIDIA 580
 AMD x86_64 A4 aci-gpu-u2404-slurm-2511-cuda-130-nvidia-580-amd64
GPU Ubuntu LTS 22.04 none
  • CUDA 13.0
  • NVIDIA 580
 AMD x86_64 A3 Ultra, A4 aci-gpu-u2204-cuda-130-nvidia-580-amd64
GPU Ubuntu LTS 24.04 none
  • CUDA 13.0
  • NVIDIA 580
 AMD x86_64 A3 Ultra, A4 aci-gpu-u2404-cuda-130-nvidia-580-amd64

Pre-installed packages (or RPMs)

Advanced Compute Images come with a variety of tools, libraries, driver, and RPMs pre-installed. To see a list of what is installed for any image, refer to the manifest file for that image.

You can find the manifest file at /usr/share/google/manifest.txt.

Pricing

Advanced Compute Images are available at no additional cost. Because Advanced Compute Images run on Compute Engine, you might incur charges for Compute Engine resources such as vCPUs, GPUs, TPUs, disks, and memory. To learn more, see Compute Engine pricing.

What's next