Nvidia has launched the DGX Spark HPC Developer Kit, a four-node desk-side cluster tailored for developers, researchers, and academics. Unveiled at the SC26 supercomputing conference in Chicago, each system node incorporates Nvidia's GB10 Grace Blackwell Superchip. The developer kit is currently available for purchase through Nvidia Marketplace and computer manufacturing partners including Acer, Asus, Dell Technologies, Gigabyte, H3C, HP, Lenovo, MSI, and xFusion. [2][3]
The setup connects four DGX Spark units in a direct ring via Nvidia ConnectX-7 networking, eliminating the need to buy or configure an external network switch. Configured as a full four-node cluster, the system delivers 512GB of combined memory alongside 20TB of NVMe storage. Nvidia states that researchers using NVFP4 can locally deploy models featuring up to 700B parameters, such as DeepSeek V4.1. [2][3]
By shifting scientific simulations and multi-node artificial intelligence workloads onto local desktops, Nvidia aims to bypass the lengthy queue times typically encountered on shared institutional supercomputers. To assist multi-node development before teams scale up to full data center hardware, Nvidia plans to release an installation playbook for its HPC SDK, supplying GPU-accelerated C, C++, and Fortran compilers and communication tools. [3]