(Seoul=News1) Reporter Yang Sae-rom = KT Cloud announced on the 24th that it has applied the NVIDIA H200 to its Graphics Processing Unit as a Service (GPUaaS).
KT Cloud provides GPUaaS, which allows for flexible GPU resource usage based on consumption, targeting domestic public institutions and AI startups that require AI infrastructure.
The services are specialized in areas requiring the supply of large-capacity, high-spec GPUs for a short period, known as ‘AI Train’, and in areas that need the continuous, seamless supply of small amounts of GPUs for inference, known as ‘AI SERV’.
KT Cloud plans to enhance the performance of its AI Train service, which was centered on the existing H100, by applying the H200.
The NVIDIA H200 is the latest GPU product based on the upgraded Hopper architecture, focused on accelerating AI and high-performance computing (HPC) workloads.
Compared to its predecessor, the H100, the H200 has improved computational processing power and energy efficiency, with a wider memory bandwidth that is expected to provide an optimized high-performance learning and inference infrastructure environment.
Meanwhile, KT Cloud is collaborating with Rebellions to provide ‘AI SERV NPU’ services dedicated to inference infrastructure in response to the neural processing unit (NPU) domain.
They plan to verify the NPU performance required by small language models (sLM) by applying Rebellions’ ‘Atom Max’, set to launch in May this year, to the AI SERV NPU.
Additionally, they are considering collaboration with the American semiconductor design company Ampere for AI semiconductor business cooperation.
Furthermore, KT Cloud plans to actively respond to public sector AI demand by conducting the Cloud Security Assurance Program (CSAP) certification process for inference-dedicated products targeting public institutions within the second half of this year.