Gigabyte AI TOP ATOM 4TB Gen 4 NVIDIA GB10 DGX-Based Personal AI Supercomputer

Gigabyte AI TOP ATOM

The Gigabyte AI TOP ATOM is a compact desktop AI supercomputer built on NVIDIA’s DGX Spark architecture, engineered to bring enterprise-grade AI compute to desktops. Delivering up to 1 petaflop of AI performance, it features 128 GB of coherent unified memory, a tightly integrated NVIDIA® GB10 Grace Blackwell Superchip, and 5th-generation Tensor Cores. With built-in ConnectX®-7 SmartNIC for high-bandwidth interconnect and support for linked scaling, the AI TOP ATOM empowers developers, researchers, and creators to prototype, train, fine-tune, and infer large models locally - no cloud dependency required.

1 Petaflop AI Performance

With support for FP4 precision and optimised sparsity, the AI TOP ATOM provides 1 petaflop (1 quadrillion operations per second) of compute throughput. This level of performance enables efficient handling of large-scale AI workloads, from model fine-tuning to inference, in a desktop form factor.

Coherent Unified Memory + Ultra-Fast Storage

The AI TOP ATOM features 128 GB of coherent unified LPDDR5X memory, shared seamlessly between CPU and GPU through a unified memory architecture. This eliminates the need for data transfer between separate memory pools, reducing latency and improving performance in complex AI tasks. Complementing this unified memory is 4TB of ultra-fast NVMe Gen 4 storage, providing ample space and rapid access for datasets, model checkpoints, and project files - ensuring efficient data handling during intensive AI workloads.

NVIDIA® GB10 Grace Blackwell Superchip

At its core lies the GB10 Grace Blackwell Superchip, combining a 20-core ARM-based Grace CPU with a Blackwell GPU connected via NVLink-C2C. This integrated design ensures low-latency coordination between compute units and efficient data sharing.

Fifth-Generation Tensor Cores

The Blackwell GPU in the GB10 chip incorporates 5th-generation Tensor Cores, with native support for FP4 precision and sparsity optimisations. These cores are designed to boost throughput for AI workloads while maintaining efficiency.

ConnectX®-7 SmartNIC & Scalability

A built-in ConnectX®-7 SmartNIC enables ultra-high-bandwidth networking and inter-unit connectivity. When two AI TOP ATOM units are linked, they can collectively support AI models of up to 405 billion parameters, enabling scalable AI workloads across a compact cluster.

Model Capacity & Use Case Support

Out of the box, the AI TOP ATOM supports AI models up to 200 billion parameters in a single unit configuration. With multi-unit scaling via ConnectX-7, it can stretch to models of 405 billion parameters - offering flexibility for advanced research, inference, and prototyping.

Compact, Efficient Design

Despite its high-end capabilities, the AI TOP ATOM maintains a compact form factor and operates with moderate power (around 170 W), running a customised Linux-based “DGX OS.” This makes it feasible for desktop deployment without excessive power or thermal demands.

I/O & Connectivity

The AI TOP ATOM provides comprehensive I/O designed for high-speed data access and flexible device integration. It includes a 10 GbE RJ-45 Ethernet port for stable wired networking, alongside Wi-Fi 7 and Bluetooth 5.3 for advanced wireless connectivity. Three USB Type-C ports deliver fast data transfer and peripheral expansion, with another for power; while a dedicated HDMI 2.1a output supports high-resolution external displays. Together, these connections ensure seamless integration with modern AI workflows, development tools, and data environments.

• 1 Petaflop AI compute.
• 128GB unified memory.
• 4TB NVMe Gen4 Storage.
• NVIDIA GB10 Grace Blackwell chip.
• 5th-gen Tensor Cores.
• ConnectX-7 scalable interconnect.
• 3 x USB-C, USB-C Power, HDMI 2.1a, 10G RJ45, NVIDIA ConnectX-7.