NVIDIA unveils DGX Spark HPC Developer Kit, a four-node desktop cluster for researchers
Announced at SC26, the kit links four DGX Spark systems in a switchless ring, letting researchers run multi-node scientific workloads and AI models at their desks instead of waiting for shared cluster time.
In plain words
NVIDIA has introduced a small computer cluster that fits on a researcher's desk. It connects four DGX Spark machines directly to each other, so scientists can test large AI models and simulations locally before requesting time on a big shared supercomputer.
What was announced
The kit connects four DGX Spark nodes over NVIDIA ConnectX-7 networking in a ring, with no network switch required. NVIDIA says the combined system offers 512GB of memory and 20TB of NVMe storage, and that NVFP4 quantization allows models with up to 700B parameters to run locally across the four nodes. An installation playbook for the NVIDIA HPC SDK, which brings GPU-accelerated C, C++ and Fortran compilers and messaging tools, is coming soon, according to the company.
NVIDIA demonstrated four-node workloads including LAMMPS with Kokkos and NAMD, while the wider DGX Spark community has validated GROMACS, OpenFOAM and WRF. Example users cited by NVIDIA include Kvantify, which ran quantum-chemistry circuit simulations on a single Spark, and UC Berkeley's Hartwig Group, which reported fitting a 120B-class model on one device and avoiding multi-day cluster queues. The kit is on display at SC26 and available through OEM partners including Dell, HP and Lenovo, and via NVIDIA Marketplace.
What it does not show
- All claims come from NVIDIA's own blog post; no independent testing or benchmark verification is included in the source.
For security and IT teams
- Access
- The developer kit is available for sale from OEM partners including Acer, ASUS, Dell, HP, Lenovo and others, and through NVIDIA Marketplace.
Sources
Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Updates and corrections
No corrections so far.
First published 17 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.