The release of a technology that allows numerous GPU resources to be used as if they were a single computer raises expectations that large-scale GPU cluster resources will be utilized more efficiently.

SK Telecom (CEO Yoo Young-sang) announced on the 8th that it has launched 'SKT Enterprise AI Cloud Manager', an AI-based B2B solution that can optimize AI infrastructure resources and integrate management of AI development environments.

'AI Cloud Manager' is a product that commercializes the management and operation know-how of SKT's large-scale GPU resources, and is a GPU cluster-specific solution based on an AI job scheduler that can efficiently utilize GPU resources. It can manage numerous GPU resources owned by corporate customers as if they were a single computer, maximizing GPU performance and shortening the learning time required for AI development.

Since large-scale data learning consumes a lot of GPU resources, such AI resource optimization is an essential function for AI cloud solutions. For example, real-time monitoring of GPU usage and performance for each project, detection of unused GPUs, and redistribution to other projects. It also improves resource efficiency by first allocating cloud resources to high-priority projects.

In addition, by utilizing 'AI Cloud Manager', you can systematically manage the entire process of AI development. AI service development goes through many stages such as △data preprocessing and storage/management △model development/learning △model distribution △model inference. 'AI Cloud Manager' provides functions that are easy for customers to utilize in each development process and provides an 'MLOps (machine learning operations)' environment that allows all processes to be organically linked.

In the AI model learning process, a large-scale parallel learning environment is provided to distribute and process large amounts of data across multiple computers, thereby increasing the learning speed. In addition, the 'AI Cloud Manager' is equipped with various convenient functions that allow developers to easily develop AI services. It can be used based on a 'web browser (Web UI)' without having to install specific software, and a collaborative environment has also been created where multiple developers can access the same AI development project at the same time.

In the model inference process, customers can choose the type of GPU and inference engine combination that is optimized for the project or cost. Various combinations are supported, such as the highest performance, lowest cost, and minimum GPU usage. In addition, RAG (Retrieval Augmented Generation) technology is introduced to minimize the 'hallucination phenomenon' that recognizes and provides past data as the latest information.

Last September, SKT collaborated with Hana Financial TI to build a large-scale GPU cluster. In this project, Hana Financial TI utilized 'AI Cloud Manager' to build an environment where Hana Financial Group can respond to AI technology demands in a stable and flexible manner. This enabled efficient GPU resource operation, learning, and shortened work times.

Kim Myeong-guk, head of SKT Cloud CO, said, “With the launch of ‘AI Cloud Manager,’ companies preparing to introduce AI services can now maximize the performance of GPU resources and develop AI services effectively.” He added, “In the future, we will lead the AI data center solution business sector by stably supplying AI data centers and GPU servers, as well as providing essential solutions to manage them.”