Lambda GPU Cloud for Deep Learning

Train the most demanding AI, ML, and Deep Learning models. Scale from a single machine to an entire fleet of VMs with a few clicks.

Training with Lambda

Carnegie Mellon
Los Alamos National Lab
Laptop with deep learning notebook connecting to a cloud GPU server.

Supporting research at every stage

Start or scale up your Deep Learning project with Lambda Cloud. Get started quickly, save on compute costs, and easily scale to hundreds of GPUs.

Discover more

Start training models right away

  • Major frameworks preinstalled

    Every VM comes preinstalled with the latest version of Lambda Stack — which includes major deep learning frameworks and CUDA® drivers.

  • Jupyter notebooks

    In seconds, access a dedicated Jupyter Notebook development environment for each machine directly from the cloud dashboard.

  • Direct SSH access

    For direct access, connect via the Web Terminal in the dashboard or use SSH directly with one of your provided SSH keys.


Spend less

Lower compute costs, without commitments

  • Save up to 50% on compute costs

    By building compute infrastructure at scale for the unique requirements of deep learning researchers, Lambda can pass on significant savings.

  • Reduce cloud TCO

    Benefit from the flexibility of using cloud computing without paying a fortune in on-demand pricing when workloads rapidly increase.

  • No multi-year commitments

    We provide on-demand pricing at or below typical one year commitment pricing — you benefit from lower costs without lock-in to a specific instance type.

Column chart showing Lambda GPU instance at 1.50 an hour vs the competitors at 1.95 and 3.06 an hour.

Scale up

Seamlessly scale GPU compute infrastructure

  • Multi-node distributed training

    Instances support up to 10 Gbps of inter-node bandwidth to enable distributed training with Horovod or another framework.

  • Hyper-parameter optimization

    Reduce model optimization times by scaling across large numbers of GPUs on one or more instances.

  • Growing research teams

    Supplement oversubscribed or underpowered compute resources with dedicated instances for new team members.

Stylized UI showing the ability to spin up many GPU instances.

Launch an instance.
Accelerate your model training today.


Deep learning doesn't require deep pockets

2x more GPU compute per dollar than other cloud providers

4-GPU Instance

For researchers, students, and hobbyists looking for extra power at low cost.

Instance Specifications

  • GPU Details

    4x NVIDIA® Pascal Based GPUs (11 GB)




    32 GB


    1.4 TB SATA SSD


    1 Gbps (peak)

Dedicated 4-GPU instance pricing

$1.50 / hr
8-GPU Instance NEW

For ML engineers and researchers looking for maximum training speed.

Instance Specifications

  • GPU Details

    8x NVIDIA® V100 Tensor Core (16 GB) + NVLink™




    448 GB


    6 TB NVMe


    10 Gbps (peak)

Dedicated 8-GPU instance pricing

$12.00 / hr
NVLink™ GPU-to-GPU Interfaces

NVLink offers up to 66% more bandwidth per connection for GPU-to-GPU communication than PCI Gen 3.0 x16 with some GPUs, including the NVIDIA® V100, having 6 NVLink connections per card.

Blazing Fast Local NVMe

Experience up to 6x faster performance than SATA3 SSDs and 14x faster performance than general purpose elastic cloud storage when loading training data from directly attached NVMe drives.

NVIDIA® V100 Tensor Core GPUs

The NVIDIA® V100 GPUs in the 8-GPU system offer 16 GB of HBM2 memory (faster than GDDR6) per card with each card also having 5120 CUDA® Cores and 640 Tensor Cores to accelerate training.

Frequently Asked Questions (FAQ)

Getting Started

Which operating systems are available on Lambda Cloud?

Lambda Cloud V100 & RTX 6000 instances run Ubuntu Server 20.04 LTS while 4-GPU Pascal based instances run 16.04 LTS. Instances are only accessible via SSH or the included Jupyter Notebook. We do not currently support any other operating systems or access via a typical desktop graphical user interface (GUI).

I'm new to SSH, how do I use the .pem private key?

You can follow the getting started guide available on our blog or our getting started video tutorial.

How do I upload and download models & data to my instances from my local machine?

Check out our how to guide on transferring data to and from your instance. Please note: If an instance is terminated, all data will be deleted, so it’s important to backup results to either your local machine or external cloud storage.

How do I shutdown / terminate the instance?

It is recommended to terminate instances using the Lambda Cloud dashboard and will be billed until terminated in the dashboard. Please note: Terminating an instance permanently deletes all data on the instance.

How do I upload and download models & data to my instances from S3 or Google Cloud Storage?

Check out our how to guide on transferring data to and from your instance. Please note: If an instance is terminated, all data will be deleted, so it’s important to backup results to either your local machine or cloud storage.

What libraries come preinstalled on Lambda Cloud instances?

Instances come preinstalled with CUDA, Python 3, Julia, Tensorflow, CuArrays, Pytorch, Plots, Flux, and Zygote.

Can I install my own DL/ML libraries and dependencies?

You can install any library that supports Ubuntu 18.04 or 16.04. Dependencies can be installed using the apt-get command for the Advanced Package Tool (APT) available in Ubuntu.


Can I use Blender or other rendering programs on Lambda Cloud?

Lambda Cloud instances do not support graphical user interfaces (GUI) and are optimized for deep learning. Using a cloud instance as a remote render farm instance is possible, but it is not a supported use case.

Can I mine Bitcoin or other crypto currency on Lambda Cloud?

Lambda Cloud is intended for deep learning use cases. Mining Bitcoin or other cryptocurrency mining is not allowed.

Do I lose my data when I terminate my instance?

If an instance is terminated, all data will be deleted, so it’s important to backup results to either your local machine or cloud storage.


When will I be billed for the instance time that I've used?

We currently bill at the end of the month for all usage from the previous month. You can change the credit card on file anytime under the billing section of the cloud dashboard.

What countries does Lambda Cloud support?

Lambda Cloud is currently only officially supported in the United States and Canada. We are currently unable to provide dedicated support for users outside of these regions.

Does Lambda offer credits for students or researchers?

We do not have a formal cloud credit program in place, but we do occasionally offer cloud credits for specific events or projects. Please email us at if you are interested in credits for an academic project.

How do I delete my account?

To delete your account please email us at from the email associated with your account stating that you would like for it to be deleted. Account deletions are permanent.


Does Lambda Cloud support multi-user access?

We do not currently support multi-user access to the dashboard. If you would like to set up different instances for different users, you can attach a unique private key to each new instance. If you would like to have multiple users share a single instance, you can use Ubuntu’s built in multi-user access access controls to give each user their own account.

Instance Information

Where are the Lambda Cloud data centers located?

We have data centers located in San Francisco, CA and Allen, TX.

What is the configuration of the instances available on Lambda Cloud?

8-GPU instances come with Ubuntu 18.04 LTS (4-GPU instances come with Ubuntu 16.04 LTS) and have a number of libraries pre-installed including: CUDA, Python 3, Julia, Tensorflow, CuArrays, Pytorch, Plots, Flux, and Zygote. For different hardware configurations, please refer to the pricing section.

Can I spin up multiple instances on Lambda Cloud?

Yes, you can spin up multiple instances in the dashboard. However, we do not currently have API support for spinning up instances.

What are the network speeds available on Lambda Cloud instances?

Depending on the instance type network burst speeds can range from 1 Gbps to 10 Gbps.

Can I only use a subset of GPUs available on each Lambda instance and pay only for the number of GPUs that I am using?

We currently have instance types with 4 or 8 GPUs. It is not possible to only pay for a partial number of GPUs in an instance.


How can I perform distributed training on multiple instances on Lambda Cloud?

Distributed training can be performed across multiple instances using a framework like Horovod. We recommend only using the same instance type for each node when clustering instances. The maximum inter-node connectivity is 10 Gbps.