NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI

NVIDIA’s DGX Spark, a compact AI supercomputer, will offer a new 64GB unified memory configuration starting this month, expanding options for local AI development. Manufactured by partners including Acer, ASUS, Dell, Gigabyte, HP and MSI, the system includes DGX OS and the full NVIDIA AI software stack preinstalled, allowing developers to run models privately without cloud dependency. The 64GB model retains the GB10 Grace Blackwell Superchip and supports models up to 100 billion parameters entirely on device, providing a cost-effective entry point for experimentation and edge development.

The 64GB configuration maintains the same software stack as the 128GB model, ensuring compatibility with NVIDIA’s AI tools and libraries. Developers can cluster two units via NVIDIA Sync Cluster Assistant, pooling memory to 128GB and supporting models up to 200 billion parameters. In testing, two clustered 64GB systems delivered up to 1.7x performance compared with a single unit, with the ability to scale further as workloads grow, making the platform suitable for larger projects.

DGX Spark ships ready for agent development, supporting tools like NVIDIA Agent Toolkit, CUDA-X AI libraries, Nemotron models, and runtimes such as Ollama, vLLM, and PyTorch with CUDA. Developers can begin running models within minutes of powering on the system. Major creator application Blender will offer a prebuilt installer for DGX Spark, enabling users to integrate the platform into their workflows seamlessly.

The new 64GB DGX Spark is priced at $4,999 and will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI on October 23. NVIDIA Sync Model Launcher, launching later this month, will simplify model deployment across single systems or clusters, including support for Qwen3.8 27B and OpenCode for browser-based coding.