Unlock Rapid AI Development: Rent Nvidia GPUs with New AWS Service

Increasingly, companies are leveraging large language models that require GPU access, with Nvidia GPUs being the most sought after. However, these resources come at a high cost and are often limited in availability. For businesses needing GPUs for a one-off task, committing to a long-term cloud rental doesn't make financial sense.

To address this challenge, Amazon Web Services (AWS) has unveiled Amazon Elastic Compute Cloud (EC2) Capacity Blocks for Machine Learning. This new offering allows customers to secure access to GPUs for a specified duration, typically to execute AI-related tasks such as training a machine learning model or conducting experiments.

Channy Yun highlighted this innovative solution in a recent blog post, stating, “This is an innovative new way to schedule GPU instances, where you can reserve the number of instances you need for a future date for just the amount of time you require.”

With this service, customers can access Nvidia H100 Tensor Core GPU instances in clusters ranging from one to 64 instances, each comprising eight GPUs. Reservations can be made for up to 14 days at a time, and users can plan up to eight weeks in advance. Once the reserved period ends, the instances will automatically shut down.

This flexible product allows customers to reserve the necessary number of GPU instances for a specified time, much like booking a hotel room. This gives users clarity on the duration of their tasks, the GPU count required, and the total cost upfront, ensuring budgeting certainty.

From Amazon's perspective, this model optimizes resource allocation in a market-like environment, ensuring consistent revenue streams, provided customer demand aligns. Prices for GPU access will fluctuate based on market dynamics.

When users enroll in the service, they can see the total cost for their selected timeframe and resources. They have the flexibility to adjust their choices based on their resource needs and budget, fostering a more tailored purchasing experience.

This new feature is now generally available in the AWS US East (Ohio) region.