GIGABYTE AI TOP ATOM Gets 64GB Version for Local AI Development

GIGABYTE is expanding its compact AI workstation lineup with a new 64GB unified-memory version of the AI TOP ATOM, giving developers another option for running artificial intelligence workloads locally.

The new configuration is based on NVIDIA's DGX Spark platform and is aimed at developers, researchers, businesses and AI enthusiasts who want to experiment with models without depending entirely on cloud computing. GIGABYTE says the 64GB model will be available starting October 23, 2026.

A Smaller Memory Option for Local AI

GIGABYTE already offers the AI TOP ATOM with 128GB of unified memory. The new 64GB configuration gives users a lower-capacity alternative for workloads that do not require the full memory capacity of the flagship configuration.

The system continues to use NVIDIA's GB10 Grace Blackwell Superchip, combining CPU and GPU resources with unified memory in a compact desktop design. NVIDIA says the 64GB DGX Spark configuration is intended to support local AI models of up to around 100 billion parameters, depending on the workload and configuration.

NVIDIA GB10 Powers the System

At the heart of the AI TOP ATOM is NVIDIA's GB10 Grace Blackwell Superchip.

The platform combines a 20-core Arm CPU with Blackwell GPU architecture and NVIDIA's AI software ecosystem. GIGABYTE's existing AI TOP ATOM specification lists a 20-core Arm processor, up to 1 petaflop of FP4 AI performance, 273GB/s memory bandwidth and NVIDIA ConnectX-7 networking.

The combination is designed for AI inference, development, data science and other compute-intensive workloads rather than traditional desktop tasks.

64GB of Unified Memory

Unified memory is particularly important for local AI because CPU and GPU workloads can access the same memory pool.

The original AI TOP ATOM configuration offers 128GB of LPDDR5X unified memory. The new 64GB version provides a more accessible configuration for developers whose models and workloads fit within a smaller memory footprint.

For users working with increasingly efficient AI models, 64GB can still provide substantial capacity for local inference and development.

Compact Desktop AI Supercomputer

Despite its performance focus, the AI TOP ATOM is remarkably small.

GIGABYTE's existing specifications list a chassis volume of approximately 1 liter, with dimensions of about 150 × 150 × 50.5mm and a weight of around 1.2kg.

That makes it considerably easier to deploy than conventional AI servers or workstation systems.

The compact form factor is particularly useful for developers who want dedicated AI hardware on a desk without installing a large server or workstation.

Designed for Local AI

The AI TOP ATOM is built around the idea of running AI workloads directly on local hardware.

This can be useful for companies working with sensitive information because data can remain within their own infrastructure rather than being sent to an external cloud service.

Potential applications include:

• AI model inference
• Model prototyping
• Fine-tuning
• Data science
• Retrieval-augmented generation
• AI application development
• Edge AI
• Robotics and computer vision

GIGABYTE specifically positions the platform for developers, researchers, students, AI enthusiasts and data scientists.

AI TOP Utility Simplifies Local AI

Hardware is only part of the equation when running AI locally. GIGABYTE also provides its AI TOP Utility, which is designed to make local AI development easier.

The utility supports features including model downloads, inference, retrieval-augmented generation (RAG) and machine-learning workflows.

This can reduce some of the configuration work normally required when setting up a local AI environment.

ConnectX-7 Enables Scaling

Another major feature is NVIDIA's ConnectX-7 networking.

The AI TOP ATOM includes high-speed networking designed to allow multiple systems to communicate and work together. GIGABYTE says two AI TOP ATOM systems can be connected to expand the available resources and handle larger AI workloads.

NVIDIA is also introducing its Sync Cluster Assistant to simplify connecting multiple 64GB DGX Spark systems. Two 64GB units can pool their memory to provide a 128GB environment, while NVIDIA says certain workloads can achieve up to 1.7× the performance of a single system.

This gives users a possible upgrade path without immediately purchasing the highest-capacity configuration.

Built for AI Agents and RAG

The system is also aimed at newer AI workloads such as agentic applications.

AI agents can perform multiple connected tasks, including analyzing documents, writing or reviewing code and carrying out research workflows. Running these workloads locally can give developers more control over data and infrastructure.

RAG is another important use case. It allows AI applications to retrieve information from external documents or private datasets before generating responses.

For businesses, this can be useful for building internal AI assistants around company information while keeping that information on-premises.

64GB vs. 128GB

The choice between the two configurations will largely depend on the size and complexity of the AI workloads being run.

The 64GB version is therefore not necessarily a replacement for the 128GB model. Instead, it gives developers another entry point into the platform.

Availability

GIGABYTE says the new 64GB AI TOP ATOM will be available from October 23, 2026. It joins the existing 128GB model in the company's AI TOP lineup.

NVIDIA has also confirmed that 64GB DGX Spark systems will be offered through manufacturer partners including Acer, ASUS, Dell, GIGABYTE, HP and MSI.

The Bigger Push Toward Local AI

The arrival of a 64GB configuration reflects a broader shift toward running increasingly capable AI models locally.

Cloud platforms remain important for large-scale training and high-demand workloads, but local AI hardware offers advantages in privacy, latency and control. Developers can experiment with models directly on their own machines instead of renting cloud computing resources for every test.

NVIDIA's DGX Spark platform is designed specifically around this trend, combining Grace Blackwell hardware, unified memory, high-speed networking and its AI software ecosystem in a compact system.

Final Thoughts

GIGABYTE's new AI TOP ATOM 64GB gives local AI developers another way to enter the company's compact AI workstation ecosystem.

It retains the core NVIDIA DGX Spark platform while reducing unified memory from 128GB to 64GB. For developers working with models that fit comfortably within that capacity, the new version could offer a more practical alternative to buying a higher-memory configuration.

With support for local inference, RAG, AI development and multi-system scaling, the AI TOP ATOM is positioned as a compact machine for the growing local-AI market.