From Application Insight to Optimized HPC and AI Clusters: System architecture and Tuning in Practice

This presentation introduces our company’s cluster solutions for HPC and AI workloads, with a focus on how application understanding, system architecture, and liquid cooling technologies come together to deliver scalable and efficient computing platforms. Rather than treating infrastructure as a collection of hardware components, we advocate an application driven design approach that aligns system choices with real workload behavior.

Drawing on extensive experience in HPC environments such as CAE and research computing, the talk highlights how deep insight into application characteristics—compute intensity, memory access patterns, communication behavior, and I/O demands—forms the foundation of effective cluster design. Based on this understanding, we present our mature system tuning and optimization methodology, which combines workload analysis, performance modeling, and iterative validation to help customers select the most appropriate hardware platforms, including CPU, GPU, and heterogeneous architectures.

The presentation will also showcase a series of practical tuning and performance characterization case studies, illustrating how architectural decisions, InfiniBand fabric design, and cooling strategies directly influence application scalability, efficiency, and system stability. Particular attention will be given to the role of liquid cooling in enabling high density systems while maintaining energy efficiency and operational reliability.

The goal of this session is to demonstrate our solution level capabilities—from application analysis and architecture design to system tuning and delivery—and to share practical insights that help customers and partners extract real value beyond theoretical peak performance. By presenting proven methodologies and real world examples, we aim to foster deeper collaboration across the HPC ecosystem and support sustainable commercial and technical growth."