GIGABYTE Unveils Four-Node AI TOP ATOM Cluster for Scientific Computing

GIGABYTE has unveiled its four-node AI TOP ATOM cluster, engineered to scale local AI computing for increasingly complex scientific and business workloads. The cluster addresses limitations in standalone systems for memory and computational power, enabling organizations to run demanding applications locally without compromising data security.

Each AI TOP ATOM node provides 1 PFLOPS FP4 AI compute power and 128GB of unified memory. By interconnecting four nodes with a 200 GbE RoCE-compatible switch, memory-intensive workloads can extend beyond single-system limits. The modular architecture supports scaling from one to four nodes as workload demands grow, while maintaining local deployment and data sovereignty.

In collaboration with NVIDIA, GIGABYTE demonstrated an AI-driven scientific computing workflow on the AI TOP ATOM cluster. This workflow utilizes NVIDIA NemoClaw blueprints and NVIDIA Nemotron-3 Nano 30B NVFP4 models for research hypothesis generation and runs GROMACS for simulations across the cluster. This integration combines AI reasoning with scientific simulation.

The demonstration focused on developing thermal interface materials (TIM) for advanced semiconductor encapsulation. While standalone systems are typically limited to simulating around ten million atoms, a four-node AI TOP ATOM cluster expands this capability to over 30 million atoms for next-generation integrated circuit packaging research. This advancement extends GIGABYTE's utility from AI development into scientific computing applications.