New AI hardware combines both CPU and GPU into one chip, which delivers supercomputer-level performance in a compact form factor.
GIGABYTE is introducing its new personal AI supercomputer called AI TOP ATOM. The system is built for developers and enterprises looking to run advanced AI workloads directly on-premises instead of relying on cloud infrastructure.
The new AI supercomputer is powered by NVIDIA’s Grace Blackwell GB10 chip, combining CPU and GPU architecture in a single platform. This design allows the system to deliver up to one petaFLOP of 4-bit floating point, allowing more parallel AI operations using less energy and energy.
The performance level enables users to train and deploy AI models with up to 200 billion parameters locally. For larger workloads, two units can be connected using the built-in NVIDIA ConnectX-7 network interface card, supporting models up to 405 billion parameters.
The device comes with 128GB of unified system memory and up to 4TB of solid-state storage. The compact system allows research labs, enterprises, and institutions to run AI tasks such as training, fine-tuning, and inference within limited physical space. The design aims to bring supercomputing capability to desktop-level environments where quick experimentation and data privacy are required.
The system includes the NVIDIA AI software stack, which provides frameworks, libraries, and tools for model training and inference. This stack allows developers to manage and optimise AI workloads efficiently across different domains, including language, multimodal, and machine learning applications.
The device also integrates GIGABYTE’s AI TOP Utility, a software interface that supports local model fine-tuning, inference, and deployment. The combined setup enables faster prototyping and testing cycles, reducing dependence on remote compute resources.
