Each node in the cluster delivers 1 PFLOPS of FP4 AI performance alongside 128 GB of unified memory. By utilizing a RoCE-capable 200GbE switch to interconnect four units, the system creates a unified architecture that scales computing resources while keeping data strictly on-premises. This modular approach allows research teams to expand their infrastructure as project demands grow, maintaining full sovereignty over sensitive datasets.

In a collaborative demonstration with NVIDIA, GIGABYTE showcased the platform’s utility by running an AI-driven scientific workflow. Using NVIDIA NemoClaw blueprints and Nemotron-3-Nano-30B-NVFP4 models, the cluster orchestrated research hypothesis generation alongside GROMACS-based simulations. When applied to thermal interface material development, the four-node setup successfully simulated systems exceeding 30 million atoms. This represents a significant leap over the 10-million-atom limit typically faced by independent systems, positioning the AI TOP ATOM cluster as a viable tool for complex, large-scale scientific discovery.