NVIDIA launches four-node DGX Spark kit for local AI and scientific computing

What happened
NVIDIA introduced the DGX Spark HPC Developer Kit at the SC26 supercomputing conference in Chicago. The kit is a four-node desktop cluster that allows researchers to run large AI and scientific workloads locally. Each node is powered by the GB10 Grace Blackwell Superchip, and the system is designed to operate without a network switch. The kit is now available for purchase from several major OEM partners and through the NVIDIA Marketplace.
Key details
- The DGX Spark Developer Kit includes four interconnected nodes, each with a GB10 Grace Blackwell Superchip.
- The system provides 512GB of memory and 20TB of NVMe storage across the cluster.
- It supports local deployment of AI models with up to 700 billion parameters using NVFP4.
- The kit enables multi-node workloads without requiring a network switch, using NVIDIA ConnectX-7 networking.
- It is available from OEM partners such as Acer, ASUS, Dell Technologies, GIGABYTE, HP, Lenovo, and others.
Why it matters
Researchers, academic labs, and developers can now run large-scale AI and scientific simulations at their own desks, reducing dependence on shared supercomputing clusters and long queue times. This allows teams to develop, test, and debug applications locally, speeding up research cycles and keeping sensitive data on premises. Organizations gain more control over their scientific workflows and can make more efficient use of their compute resources.

