The Pulse
GIGABYTE Adds a 64GB AI TOP ATOM Desktop for Local AI
GIGABYTE says its AI TOP ATOM desktop will be available in a 64GB unified-memory configuration from October 23, alongside the existing 128GB version. The company also describes local AI software features and networking for clustering up to
AI.info Team · Written with AI ·
A second memory configuration arrives October 23
GIGABYTE announced on October 2 that its AI TOP ATOM desktop will gain a 64GB unified-memory configuration, adding an option alongside the existing 128GB model. The new version is scheduled to become available on October 23, 2026. The company says the system keeps the existing hardware design and is based on NVIDIA’s DGX Spark platform.
The announcement presents the new configuration as a choice for developers, researchers and enterprise teams whose AI workloads call for different memory capacities. GIGABYTE does not describe the 64GB system as a replacement for the 128GB model, or provide a performance comparison between the two. It frames the change as an expanded product lineup for local AI development, with users able to select a capacity based on model sizes, workflows and multitasking requirements.
Local inference, prototyping and document search
AI TOP ATOM is built for work such as model inference, prototype development and data analysis on a compact desktop system. GIGABYTE says running models and processing data locally can give users more control over development resources and project data while reducing reliance on cloud computing. The release positions the machine for offices, laboratories and educational settings.
The software package combines NVIDIA’s CUDA-accelerated AI software ecosystem with GIGABYTE AI TOP Utility. The company lists model downloading, inference and retrieval-augmented generation among the utility’s capabilities. Users can apply RAG to their own documents to build knowledge-based question-and-answer applications, according to the announcement. The release also describes exploring open models and testing AI assistants, with developers able to refine and validate solutions against project requirements. It does not provide test results or quantify how much data those workflows can handle.
ConnectX-7 links systems for larger workloads
GIGABYTE says the desktop’s built-in ConnectX-7 networking allows developers to cluster up to four units with NVIDIA Sync for a larger memory pool and more computing capacity. The announcement does not specify the resulting combined memory capacity or give a benchmark for clustered systems, so it does not establish how much a particular workload would benefit.
The company also points to a multi-node scientific-computing demonstration that connects research-hypothesis generation with simulation workflows. It says the demonstration combines NVIDIA Nemotron open models with the NVIDIA NemoClaw open agent blueprint. The release describes the setup as an example of potential support for scientific research, rather than reporting measured research outcomes.
Price and regional availability remain unspecified
GIGABYTE gives no price for the 64GB configuration. It says availability, sales channels and pricing may vary by region, and directs buyers to official announcements and authorized local distributors for details.
The announcement’s concrete addition is a second memory tier for a desktop positioned around on-premises AI work. The company says the platform is intended to support development from early proof-of-concept work through practical deployment, but publishes no pricing, performance comparisons or workload benchmarks for the new configuration. Buyers have a release date and a set of advertised software and networking features, but not enough information in the announcement to assess the 64GB model’s value against the 128GB version.