GIGABYTE Demonstrates AI TOP ATOM Cluster for Scientific Computing
GIGABYTE has demonstrated its AI TOP ATOM solution, enabling four-node clustering to scale local AI compute and memory for complex scientific simulations, including IC packaging technology development.
TAIPEI, July 9, 2026 – GIGABYTE showcased a solution that scales local AI compute power through a four-node AI TOP ATOM cluster. This approach aims to address the increasing memory and computational demands of more complex scientific simulations and AI workloads.
The core of the solution involves clustering AI TOP ATOM nodes, each offering up to 1 PetaFLOP of FP4 AI performance and 128GB of unified memory. Four interconnected nodes can, via a RoCE-enabled 200GbE switch, scale memory-intensive workloads beyond the limitations of a single system. The modular architecture allows for scaling from one to four nodes while maintaining local deployment and data control.
GIGABYTE demonstrated the solution's capabilities with an AI-powered workflow for scientific simulations, developed in collaboration with NVIDIA. This workflow utilizes NVIDIA NemoClaw blueprints and the Nemotron-3-Nano-30B model, combining AI-generated hypotheses with GROMACS simulations across the entire cluster. This integration shows how AI can drive research within cluster computing environments.
As an application example, the solution is applied to the development of thermal interface materials (TIM) for advanced semiconductor packaging. While standalone systems are typically limited to simulating around 10 million atoms before hitting memory limits, a four-node AI TOP ATOM cluster expands simulation capacity to over 30 million atoms for IC packaging technology research. This extends the application of AI TOP ATOM from AI development to scientific computing and simulation.