NHN Cloud (hereafter NHN Cloud) said on the 19th that it was selected as the operator of Krafton's ultra-large GPU (graphics processing unit) cluster.
NHN Cloud was selected as the final operator for Krafton's "GPU cluster project," signed a contract, and will provide infrastructure based on GPU as a service (hereafter GPUaaS).
The project was pursued to meet rising AI computing demand following Krafton's declaration of "AI first" as a core management strategy in Oct. last year and the broader adoption of artificial intelligence (AI). The GPU cluster is core infrastructure to reliably implement Krafton's mid- to long-term AI strategy, aiming for flexible scaling and efficient operations by using NHN Cloud's GPUaaS.
NHN Cloud will build a GPU farm composed of about 1,000 of the latest Blackwell Ultra GPUs in a multi-cluster architecture and create a large-scale AI computing environment by applying an XDR-800G-class ultra-high-speed InfiniBand network. This will minimize data transfer latency between GPUs and stably support a range of tasks including AI model training and inference.
It also applies a dynamic management architecture so multiple tasks can share GPU resources, minimizing idle capacity and enabling efficient GPU use tailored to task size and characteristics, from small-scale AI development to large-scale large language model (LLM) training. Through a customized GPUaaS that applies a Slurm-based resource management solution suited for Kubernetes and high-performance computing (HPC) environments, it will also strengthen efficiency in AI development and operations.
The Blackwell Ultra GPU infrastructure will be built at the NHN Cloud Pangyo NCC (NHN Cloud Center), with construction to be completed in July and full operations to begin thereafter.
A Krafton official said, "In driving the AI first strategy, the GPU cluster is the core foundation for companywide AI operations," and added, "Through GPUaaS, we expect to strengthen scalability and efficiency across AI research and services."
An NHN Cloud official said, "This project is a case that proves our technology and operational capability to build and operate large-scale GPU clusters in a GPUaaS model," and noted, "We will continue to support Krafton's AI first strategy by providing stable clusters and operational support."