Deep Learning VM Images are preconfigured virtual machine images for data science and machine learning. They ship with the most important frameworks and tools already installed, so an instance can be used for development and training right away - including on machines with GPUs.
What are Deep Learning VM Images?
According to the documentation, the images cover many combinations of framework and processor. Variants are available for TensorFlow Enterprise, TensorFlow, PyTorch, and generic high-performance computing, each in versions for CPU-only and GPU-enabled workflows.
The images are based on Debian 11 and Ubuntu 22.04. They can be configured to include a specific framework and its supporting packages, plus Python 3.10 with common libraries such as numpy, scipy, matplotlib, pandas, nltk, pillow, scikit-image, opencv-python, and scikit-learn. JupyterLab notebook environments are included for quick prototyping.
For GPU-enabled instances, the images include the Nvidia packages with the latest driver plus CUDA in versions 11.x and 12.x and CuDNN 7.x or NCCL 2.x; the specific versions depend on the chosen framework and the CUDA version.
Google updates the images regularly with bug fixes and package updates; the individual changes are documented in the release notes. According to the pricing page, Deep Learning VM Images instances are free to use; you pay for the Compute Engine resources consumed.
Core Features
- Preinstalled frameworks: Images for TensorFlow Enterprise, TensorFlow, PyTorch, and generic high-performance computing.
- CPU and GPU variants: Matching images for both operating modes.
- Complete GPU stack: Nvidia driver, CUDA, and CuDNN or NCCL are included.
- Python environment: Python 3.10 with the common data science libraries.
- JupyterLab: Notebook environments for quick prototyping.
- Regular maintenance: Ongoing updates with bug fixes and package updates.
Typical Use Cases
Fast project start
A team starts a new ML project and uses a preconfigured image instead of installing drivers, CUDA, and frameworks itself.
GPU training without setup effort
A GPU instance with a matching image is created for a training experiment, with the full Nvidia stack already in place.
Reproducible training environments
Several team members work with the same image and therefore identical framework and library versions.
High-performance computing tasks
For compute-intensive work without a specific ML framework, the generic HPC variant is used.
Prototyping in the notebook
The included JupyterLab environment is used to test ideas directly on the instance before moving them into pipelines.
Benefits
- No setup effort: Frameworks, drivers, and libraries are already included.
- Matching variants: Every common framework has its own image.
- No surcharge for the image: According to the pricing page, Deep Learning VM Images instances are free to use.
- Current GPU stack: Nvidia driver, CUDA, and CuDNN are shipped with the image.
- Maintained base: Regular updates with bug fixes and package updates.
Integration with innFactory
As a certified Google Cloud Partner, innFactory supports you in using Deep Learning VM Images: selecting suitable images and machine types, sizing GPUs, building reproducible training environments, and connecting them to your existing ML workflows.
Contact us for a consultation on Deep Learning VM Images and Google Cloud.
Typical Use Cases
Technical Specifications
Frequently Asked Questions
What are Deep Learning VM Images?
According to the documentation, Deep Learning VM Images are a set of virtual machine images optimized for data science and machine learning tasks. All images come with key ML frameworks and tools pre-installed and can be used out of the box on instances with GPUs.
Which frameworks are available?
The documentation lists images supporting TensorFlow Enterprise, TensorFlow, PyTorch, and generic high-performance computing, each in versions for CPU-only and GPU-enabled workflows. The complete list is in the image selection section of the official documentation.
What software is included?
The images are based on Debian 11 and Ubuntu 22.04. They include Python 3.10 with packages such as numpy, scipy, matplotlib, pandas, nltk, pillow, scikit-image, opencv-python, and scikit-learn, plus JupyterLab notebook environments. GPU-enabled instances additionally include the latest Nvidia driver, CUDA 11.x and 12.x, and CuDNN 7.x or NCCL 2.x, with versions depending on the framework and CUDA version.
What do Deep Learning VM Images cost?
According to the pricing page, Deep Learning VM Images instances are free to use. Because they run on Compute Engine, you pay for the Compute Engine resources consumed.
How often are the images updated?
The documentation states that Deep Learning VM images are updated regularly with bug fixes and package updates. Details on individual updates are in the release notes.
Note: All product information on this page has been compiled with care, but is provided without guarantee and may be outdated or incomplete. Cloud services evolve rapidly — features, pricing, SLAs, and availability change frequently. Authoritative and up-to-date information can only be found on the official product page of Google Cloud (official documentation). This page does not represent an offer by Google Cloud.
