Education & Research

[GIST] Building an Integrated Platform for Large-scale GPU Infrastructure Operations and AI Validation

To boost AI R&D and validation across industry, academia, and research institutes on its large-scale shared HPC-AI infrastructure, the Gwangju Institute of Science and Technology (GIST) adopted an integrated AI development platform that supports infrastructure operations, user management, AI asset sharing, and model deployment.

CheetahMLOps

Efficient Operations for Large-scale Shared HPC-AI Infrastructure

GIST aimed to create an AI R&D and validation environment that local universities and companies could use together, based on Korea's largest shared HPC-AI infrastructure. This required an integrated operating framework covering large-scale GPU operations, management of hundreds of users, AI asset sharing, and validation services.

Securing Large-scale Infrastructure Operations and AI Collaboration and Validation

The key challenge was to reliably operate more than 160 A100 GPUs while hundreds of users conduct R&D simultaneously in independent environments, and to connect their results to validation services.

  • HPC infrastructure operations: A framework to efficiently operate, allocate, and reclaim more than 160 A100 GPUs

  • Support for many users: An environment where an average of 200 users can access the platform at once and conduct AI R&D independently

  • AI asset sharing and validation: A foundation to share AI assets such as data, code, and work outputs and connect them to validation services

Building an Integrated Platform for HPC-AI Operations and AI Service Validation

With Cheetah, we made HPC-AI infrastructure easy to operate even for non-specialists, and provided independent development environments for each user, AI asset sharing, and a validation framework based on one-click deployment.

  • Intuitive infrastructure operations: An easy UI for managing HPC-AI resources and allocating and reclaiming resources per user

  • Independent environments and collaboration: Per-user development environments and AI asset sharing support R&D and collaboration

  • Validation through one-click deployment: AI models can be deployed instantly on the platform and connected to validation services

A Shared Infrastructure Foundation for AI R&D and Validation by Industry, Academia, and Research

GIST now operates its large-scale shared HPC-AI infrastructure more efficiently and can manage resources and users with in-house staff alone. University and corporate users can conduct AI R&D simultaneously in independent environments and deploy their models immediately for validation, giving them a professional foundation for using AI.

  • More efficient infrastructure operations: In-house staff can easily manage large-scale GPU resources and user environments

  • Concurrent R&D environments: Up to 200 users can access the platform at once and work in independent development environments

  • Faster path to validation: Models deploy in one click, quickly connecting to AI service validation

  • Are you curious about an enterprise AI integration platform? → Cheetah

Have more questions?

Contact us anytime to learn more about AIFRICA solutions.

Contact Us