Amazon Web Services announced on November 18 the AWS Lambda GPU feature, which lets Event Functions run on GPU instances in Lambda Managed Instances. In practice, developers can run AI inference, media processing, 3D rendering and scientific computing on GPUs without provisioning servers or managing drivers.
How it works
With AWS Lambda GPU, GPU-accelerated functions are deployed as container (OCI) images and run on EC2 G-series instances managed by Lambda: the service handles provisioning, automatic scaling, drivers and event integration. There are no GPU drivers, CUDA versions or AMIs to reconcile, and no clusters or device plugins to operate, according to AWS.
AWS supplies ready-made base images with PyTorch, vLLM and SGLang, and each function can use up to 32 GB of NVMe ephemeral storage, lazy-load large models from S3 and work with 3 to 48 GB of accelerator memory. Functions inherit the extended Lambda Managed Instances timeout of up to 90 minutes for asynchronous invocations, and integrate with Lambda's 200+ event sources.
Instances, availability and pricing
At launch, the feature covers the G4dn (NVIDIA T4), G5 (NVIDIA A10G), G6, G6f, Gr6, Gr6f (NVIDIA L4) and G6e (NVIDIA L40S) families, with more families promised. It is available in all AWS commercial regions where Lambda Managed Instances and G-series instances exist.
Pricing follows the On-Demand price of the underlying EC2 instance, with Savings Plans and Reserved Instances applying, plus a 9.75% management fee for GPU instances and the standard $0.20 per million requests. AWS positions the feature for inference and small-to-medium custom models — not for training, and not for the large foundation models already available on Amazon Bedrock. New accounts may start with a zero quota for on-demand G instances, so a quota increase must be requested before using the feature.
Image credit: Amazon Web Services logo, via Wikimedia Commons.
Sources: AWS News Blog (anúncio oficial); iMasters; Oossa.
Transparency: this article was produced with the help of artificial intelligence, based on public posts and sources on the internet, and checked against the original sources cited. We recommend consulting the sources to confirm the details.
By Muse