Overview

On-Demand Cloud (ODC) provides on-demand access to Linux-based, GPU-backed virtual machine instances.

Instance types

ODC offers a variety of predefined instance types to support different workload requirements. Available GPUs include the state-of-the-art NVIDIA HGX B200 GPU, NVIDIA GH200 Grace Hopper Superchip, and NVIDIA H100 Tensor Core GPU, as well as several earlier models. Each instance you create is tied to a specific geographical region. For a list of available regions, see the section below.

Select instance types are backed by GPUs that feature NVIDIA SXM. SXM offers improved bandwidth between the NVIDIA GPUs in a single physical server.

Warning
Lambda prohibits cryptocurrency mining on ODC instances.

As of December 2025, ODC offers the following instance types:

Base images

Default image

By default, each ODC instance runs Ubuntu 22.04 LTS. Lambda also preinstalls Lambda Stack, a standard set of AI/ML-related drivers, tools, and frameworks, on the instance:

• NVIDIA tools, libraries, and drivers: CUDA, cuDNN, NCCL, NVIDIA container toolkit, NVIDIA driver
• Deep learning frameworks and libraries: TensorFlow, torchvision, Keras, PyTorch®, JAX, Triton
• Dev tools: Git, Vim, Emacs, Valgrind, tmux, screen, htop, build-essential

For more information on Lambda Stack, see Lambda Stack.

In addition, each ODC instance provides a JupyterLab installation for creating and managing Jupyter notebooks. You can access your instance's JupyterLab by visiting the Instances page in the Lambda Cloud console and clicking Launch in your instance's row.

Note
JupyterLab can take a few minutes after an instance launches to become accessible.

Alternative images

If you launch your instance using the Lambda Cloud API, you can choose to use an alternative base image instead. Available image types include:

• Lambda Stack: The default image used on most ODC instances.
  • Includes: NVIDIA driver, CUDA toolkit, OFED, cuDNN, NCCL, Lambda Stack Python packages (PyTorch, TensorFlow, JAX), Docker, JupyterLab
  • Use if: You want something that "just works" and don't need deep configurability.
• Lambda Stack Legacy: The default image on V100-backed ODC instances.
  • Includes: NVIDIA driver (closed-source), CUDA toolkit, OFED, cuDNN, NCCL, Lambda Stack Python packages (PyTorch, TensorFlow, JAX), Docker, JupyterLab
  • Use if: You're running an instance that requires NVIDIA's legacy closed-source driver stack, such as a V100 instance.
• GPU Base: An image based on Ubuntu Server that includes a minimal set of key AI/ML tools and drivers.
  • Includes: NVIDIA driver, CUDA toolkit, OFED, cuDNN, NCCL, Docker
  • Use if: You want a minimal working setup with a high degree of control over your Python environment.
• Ubuntu Server: A standard, unmodified Ubuntu Server image.
  • Includes: No additional packages.
  • Use if: You're an expert user who prefers maximum customizability, or you have unique tool or driver requirements.

The table below lists the available versions for each of the above image types, the architectures each version supports, and the default Python version each includes.

Note
As of December 2025, running sudo apt full-upgrade or sudo apt dist-upgrade on the Lambda Stack 24.04 or GPU Base 24.04 base images produces an error that prevents the upgrade from completing successfully. For instructions on how to avoid the error, see Troubleshooting > apt full-upgrade fails on Lambda Stack 24.04 and GPU Base 24.04 images.

Storage

Root volume

Each new instance comes with a root volume of a predefined size. The specific volume size depends on which instance type you choose. To see which volume sizes are associated with each instance type, check out the instance type table above.

Filesystems

When you create a new instance, you're prompted to attach a filesystem. A filesystem is a unit of networked persistent storage you can connect to your instance. Filesystems are typically several orders of magnitude larger than your root volume, and are an ideal place to store both your instance state and your large datasets.

To use a filesystem with your instance, you must attach it during the instance creation process. The filesystem must also reside in the same region and workspace as your instance.

For more information about filesystems, see Filesystems.

Mount point

When you mount a filesystem to your instance, the filesystem is available at:

/lambda/nfs/<FILESYSTEM_NAME>

By default, the filesystem is symlinked in the ubuntu user's home directory, using the filesystem's name. You can safely remove the symlink if it's not needed.

Billing

Filesystems are billed per GB used per month, in increments of one hour. If you delete an instance but not its associated filesystem, you'll continue to be billed for the filesystem. For more details on filesystem pricing, see Filesystems.

Size limits

ODC filesystem size limits vary by region. In general, filesystems have a capacity of 8 EB (8,000,000 TB), and you can have a total of 24 filesystems. In the Texas, USA (us-south-1) region, filesystem size is limited to 10 TB.

Network

Connection options

You connect to your instance through a standard SSH connection. For information on creating and managing SSH keys, see Lambda Cloud console > Add, generate, and delete SSH keys or the SSH keys section of the Lambda Cloud API browser.

Firewall defaults

You can define inbound TCP and UDP firewall rules on the Firewall page in the Lambda Cloud console. By default, only port 22 (SSH) is open.

ODC allows ICMP traffic by default, as many network diagnostic tools rely on ICMP to determine where connectivity issues are occurring. If you'd prefer to restrict ICMP traffic, you can do so on the Firewall page.

Regions

Lambda resources are hosted in multiple locations worldwide. Not every instance type will be available in every region.

Instance management

Lambda Cloud console

In addition to the firewall settings mentioned earlier, you can manage your instances, filesystems, SSH keys, API keys, and more through the Lambda Cloud console. For more information, see Lambda Cloud console.

Cloud API

You can perform administrative tasks such as creating, restarting, listing, and terminating your instances through the Lambda Cloud API. For more details, visit the Lambda Cloud API browser.

Next steps

• Visit the Lambda Cloud portal
• Explore the Lambda Cloud API

1. The NVIDIA GH200 Grace Hopper Superchip features an onboard ARM CPU. Verify that your intended workloads run on ARM before choosing this instance type. ↩