Dmytro Kharkov

NVIDIA launches DGX Spark developer kit for desktop scientific computing

NVIDIA launches DGX Spark developer kit for desktop scientific computing
NVIDIA debuts DGX Spark

NVIDIA is introducing a desktop-scale DGX Spark HPC Developer Kit at SC26 in Chicago, aimed at researchers, academics and developers building multi-node scientific and AI workloads. The package links four DGX Spark systems into a ring configuration, giving teams local access to up to 512GB of combined memory and 20TB of NVMe storage without relying on shared GPU cluster time.

Highlights

  • NVIDIA unveiled the DGX Spark developer kit, featuring four nodes connected via ConnectX-7 without a network switch for streamlined desktop scientific computing.
  • The kit, based on the GB10 Grace Blackwell Superchip, allows scaling up to four nodes and supports local deployment of AI models with up to 700B parameters.
  • An upcoming NVIDIA HPC SDK installation playbook will provide GPU-accelerated development tools to facilitate debugging and scaling multi-node applications before full data center deployment.

Desktop cluster design and software roadmap

As reported by NVIDIA Newsroom, the new kit combines four NVIDIA DGX Spark systems, each connected directly to two neighboring nodes over NVIDIA ConnectX-7 networking, eliminating the need for a separate network switch. NVIDIA says the setup lets researchers build, test and debug multi-node applications at their desks before moving workloads to larger data center infrastructure.

The system scales from one node to four and is built around the GB10 Grace Blackwell Superchip. NVIDIA says the four-node configuration supports local deployment of models with up to 700B parameters using NVFP4, including workloads such as DeepSeek V4.1, while also providing capacity for distributed scientific computing and AI agents.

An NVIDIA HPC SDK installation playbook is coming soon for the cluster. The company says the toolkit will bring GPU-accelerated C, C++ and Fortran compilers, libraries and message passing interface tools, allowing teams to examine data distribution, test communication across nodes and resolve issues before scaling up to production systems.

In a previous article, we explained how expanded CUDA-X libraries aim to speed up scientific computing and AI-driven research workflows.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.