Running Qwen 3.8 27B tests, two clustered 64GB units delivered up to 1.7x performance compared to a single system. This local hardware removes cloud dependency for developers, researchers, and AI enthusiasts working on experiments and agentic workflows.
The 64GB platform provides a complete environment for inference, fine-tuning, and data science by combining the GB10 Grace Blackwell Superchip with NVIDIA DGX OS and the full NVIDIA AI software stack. This configuration supports models with up to 100 billion parameters directly on the device, offering a compact supercomputing solution that allows builders to keep their data private without relying on external cloud instances.
NVIDIA reports that when workloads demand more power, users can connect two units using the NVIDIA Sync Cluster Assistant without additional setup. This system is designed to scale alongside growing project requirements, offering a more accessible price point than the existing 128GB model while maintaining identical core software and architectural capabilities.
Major manufacturers including Acer, ASUS, Dell, Gigabyte, HP, and MSI are partnering with NVIDIA to offer the hardware. Specifically, Gigabyte announced the AI TOP ATOM 64GB version on October 2, 2026, while MSI has expanded its EdgeXpert lineup to include this new configuration. The hardware will be available to consumers starting this month.