- .

(Seoul=NSP NEWS) = Samsung SDS has set out to expand its enterprise AI inference infrastructure by introducing GPUaaS based on the NVIDIA B300 GPU.

This service is characterized by its design that aligns with the flow of enterprise AI utilization shifting from a model training-centric approach to actual service operations and inference phases.

The B300 GPU utilizes 12-layer HBM3E to provide 288GB of memory per GPU and a bandwidth of 8TB per second. Compared to the existing H100, the memory capacity is 3.6 times and the bandwidth is 2.4 times expanded.

As a result, an environment has been established that reduces memory bottlenecks occurring when running large-scale language models and enhances the processing efficiency of high-performance AI services, such as AI agents and image, video, and code generation.

The service operates on a subscription basis, reducing the burden of initial infrastructure investment and applying a usage-based cost structure. Additionally, it is designed to support enterprise data processing environments by integrating security features.

Samsung SDS plans to add serverless-based inference services and automated distributed learning-based AI learning services in the third quarter of this year.

By Soon-ki Lee(s8789@nspna.com) and Bok-hyun Lee(bhlee2016@nspna.com)

ⓒNSP News Agency·NSP TV. All rights reserved.