Nvidia launches DGX Spark HPC dev kit that runs 700B-parameter models locally

Nvidia announced the DGX Spark HPC development kit at the SC26 conference. It is built for researchers, academics and AI developers, and packs four DGX Spark units running on the GB10 Grace Blackwell superchip.

The four nodes link directly through Nvidia ConnectX-7 network adapters, so no extra network switch is needed. The kit offers 512GB of memory and 20TB of NVMe storage, letting teams spread models and science workloads across the cluster. It can run large AI models such as DeepSeek V4.1 locally, with up to 700 billion parameters.

Nvidia is also releasing an installation guide for the Nvidia HPC SDK, which brings GPU-accelerated C, C++ and Fortran compilers, libraries and message passing tools. Teams can build and test science apps on the four nodes and check how work is spread before scaling up. The kit is on show at SC26 and can be bought through OEM partners including Acer, Asus, Dell, Gigabyte, H3C, HP, Lenovo, MSI and xFusion, or as a direct reference model.

SponsoredDomoAI
AI Tools Mentioned

Comments

Loading comments…