Getting started with Foundry
Welcome!
Foundry Cloud Platform is a public cloud purpose-built for the unique requirements of AI/ML workloads. It offers ML engineers and researchers self-serve access to NVIDIA GPU compute for training, fine-tuning, and inference without long-term contracts or opaque procurement processes.
Using Foundry Cloud Platform, ML teams can:
Reserve compute for as little as 3 hours for critical tasks like training, verification, and debugging.
Burst preemptible Spot compute for interrupt tolerant workloads like inference, hyperparameter training, and fine-tuning.
Resell reserved compute when not in use (coming soon)
Today Foundry Cloud Platform offers access to NVIDIA H100 80GB SXM GPUs with Infiniband, A100 80GB SXM GPUs with Infiniband, and inference focused chips (e.g., A40, A5000). All of our servers are located in Tier 3 and 4 datacenters and follow stringent security protocols to ensure your data and workloads are secure.
Quickstart
Foundry Cloud Platform is currently in a limited-access rollout. To request access, join the waitlist.
To start using Foundry Cloud Platform:
Create a Foundry account. If your organization has not been added to the private beta, your account will remain in the waitlist. Please ensure that you have submitted our waitlist form with the same email as your account. You'll receive an email when you've been granted preview access.
Create an instance. Once your organization has been granted preview access, you can start to create compute instances. Read Getting started with compute for a step-by-step guide.
To discuss bespoke needs like long-term (2+ month) clusters or becoming a compute supplier for Foundry, contact the Foundry solutions team (solutions@mlfoundry.com)
Last updated