KT Cloud Applies H200 to 'AI Train' in Q3

Expansion of Subscription-Based GPU Service Lineup
Twice the Inference Speed Compared to H100
Strengthening AI-Based Enterprise Solutions

NVIDIA's latest GPU H200.

KT Cloud announced that it will introduce NVIDIA's latest graphics processing unit (GPU) H200 to its AI training infrastructure service 'AI Train' in the third quarter of this year.

At the 'KT Cloud AI Business and GPUaaS' online briefing held on the 20th, Jeong Mijin, head of KT Cloud AI Business Team, said about the recently secured NVIDIA H200 subscription-based GPU service (GPUaaS) introduction plan, "We plan to provide enhanced infrastructure during the third quarter by applying the H200 GPU to AI Train." The H200 is NVIDIA's latest GPU product released last year, focusing on accelerating AI and high-performance computing (HPC) workloads. Based on Meta's open-source LLaMA 2, the H200 boasts approximately twice the inference speed compared to its predecessor, the H100.

AI Train is a service specialized for the training process that requires intensive use of high-performance GPUs in a short period when building large-scale AI systems. By utilizing container technology, multiple GPUs can be combined and used as one, making it efficient for large-scale AI development by enterprises. Currently, A100 and H100 GPUs are provided, and the more powerful H200 will be added this time.

KT Cloud did not disclose the exact number of H200 units secured. A KT Cloud official stated, "It is difficult to disclose the specific quantity of introduction," but added, "We plan to continue providing high-performance infrastructure services through large-scale investments in the future."

Additionally, KT Cloud has established a security system that meets the 'ISMS-P' certification standards, an information protection and personal information protection management system certification jointly hosted by the Ministry of Science and ICT and the Personal Information Protection Commission, ensuring data security even in multi-user environments. They also plan to pursue 'CSAP' security certification for public institutions.

KT Cloud is also providing a cloud service based on NPU (Neural Processing Unit) developed by the domestic company 'Rebellion.' The NPU has advantages such as higher power efficiency and cost reduction compared to GPUs. While GPU resources are basically operated in a shared manner among multiple users, dedicated usage services are also under consideration.

KT Cloud plans to launch an advanced AIOPS (AI-based IT Operations) solution in the first quarter and subsequently introduce RAG (Retrieval-Augmented Generation) services tailored to enterprise environments. Retrieval-Augmented Generation is a technology that combines search and generation in machine learning models. The RAG service is targeted for release in the third quarter, with necessary functions such as document processing, information retrieval, and AI language model integration being prepared.

This content was produced with the assistance of AI translation services.