AWS integrates NVIDIA T4 GPUs into Outposts to support on-premises AI inference and rendering workloads.

Amazon Web Services Inc. today enhanced its AWS Outposts platform with the addition of G4 instances powered by Nvidia Corp.'s T4 graphics processing unit. Outposts enables organizations to deploy AWS infrastructure in their own data centers and have the Amazon.com Inc. subsidiary manage the hardware on their behalf. This offering targets companies that wish to take advantage of AWS's technological capabilities but cannot deploy their workloads on its public cloud for regulatory, compliance, or security reasons. For example, healthcare companies may prefer to keep patient records and proprietary research files within their own controlled data centers rather than moving them to external facilities.

Justin Boitano, the senior director of product for Nvidia's enterprise and edge computing group, emphasized the business case for this approach: "Most companies still keep their data inside their own walls because they see it as their core intellectual property. But for deep learning to transition from research into production, enterprises need the flexibility and ease of development the cloud offers — right beside their data. That's a big part of what AWS Outposts with T4 GPUs now enables."

The T4-powered G4 instances are designed to address two primary use cases: inference, which involves running trained AI models on live data, and graphics rendering. G4 instances come in seven sizes ranging from four to 96 virtual central processing units. The five lowest-end configurations come with a single T4 GPU, while the top-end models feature four and eight GPUs respectively, plus a maximum of 1.8 terabytes of local NVMe flash storage.

Nvidia's T4 chip occupies a middle position in the company's data center product hierarchy, sitting below its premium Volta V100 card. Unlike the V100, which is oriented strictly toward machine learning workloads, the T4 features a more general-purpose architecture and includes 40 specialized circuits called RT Cores designed to accelerate graphics processing. These RT Cores speed up the generation of light and shadow effects in media files, including computer-generated imagery for film production.