NVIDIA has introduced the DGX Spark HPC Developer Kit for researchers to develop scientific applications and run AI models without waiting for access to shared computing infrastructure.

Unveiled at the SC26 supercomputing conference in Chicago, the four-system desktop cluster is available for sale, with an installation guide for NVIDIA’s scientific computing tools coming soon.

Built around the GB10 Grace Blackwell Superchip, the kit scales from one DGX Spark system to four. Each system connects directly to its two neighbours in a ring using NVIDIA ConnectX-7 networking, removing the need to buy or configure a separate network switch.

A four-node configuration provides 512GB of combined memory and 16TB of NVMe storage. It can deploy AI models with up to 700 billion parameters locally using NVFP4, a low-precision numerical format.

Developing locally

The kit is designed for a development environment that complements larger computing systems. Researchers can build, test and debug applications at their desks, reserving shared cluster time for production runs while keeping unpublished findings and proprietary datasets on premises.

Teams can develop and debug on their own, saving server hours for production runs and keeping unpublished results, and proprietary datasets on premises.

A forthcoming NVIDIA HPC SDK installation playbook will bring GPU-accelerated C, C++ and Fortran compilers, libraries and message-passing tools to the cluster.

Teams will be able to examine how workloads and data are distributed across four nodes and resolve communication problems before moving applications to larger systems.

DGX Spark systems and cables are available through NVIDIA Marketplace and hardware partners including Acer, Asus, Dell Technologies, Gigabyte, H3C, HP, Lenovo, MSI ,and xFusion.


Discover more from Entelechy Asia

Subscribe to get the latest posts sent to your email.

Leave a Reply