Skip to main content
Back to timeline
NVIDIA BlogSource publication:

NVIDIA puts a four-node DGX Spark cluster on the desk: 512GB of memory runs 700B-parameter models locally, no shared-cluster queue

Synopsis

At SC26, NVIDIA introduced the DGX Spark HPC Developer Kit, a four-node ring of DGX Spark systems for a researcher's desk. Multi-node scientific workloads and AI models have typically waited on shared GPU clusters; the kit links nodes directly over ConnectX-7 with no switch, combines 512GB of memory and 20TB of NVMe storage, and can deploy models up to 700B parameters locally under NVFP4. The kit is on sale, while the HPC SDK playbook is still to come.

Source-provided article image: NVIDIA DGX Spark Developer Kit Turns a Desk Into a Supercomputing Lab
NVIDIA Research

Interpretation

Researchers get a four-node cluster at the desk for multi-node scientific workloads, AI models and agents, instead of waiting for time on a shared GPU cluster. Such work has depended on shared-cluster allocation and queues, and teams also consumed server hours while developing and debugging. Four DGX Spark systems connect directly to their two neighbors over ConnectX-7 with no network switch to buy or set up, giving 512GB of combined memory and 20TB of NVMe storage.

Models and data stay on premises, so teams develop and debug themselves, save server hours for production runs and keep unpublished results and proprietary datasets local. The kit scales from one node to four, each carrying the GB10 Grace Blackwell Superchip. With NVFP4, the four-node cluster can deploy models with up to 700B parameters locally, such as DeepSeek V4.1.

Perspective

The kit targets researchers, academics and developers; demonstrated four-system workloads include LAMMPS with Kokkos for materials simulation and NAMD for molecular dynamics, while GROMACS, OpenFOAM, WRF and NIM microservices such as OpenFold3 and Boltz-2 were validated by the broader DGX Spark community. The kit is on display at SC26 and available for sale through OEM partners and NVIDIA Marketplace.

The HPC SDK build-and-run playbook is marked as coming soon and will only then give scientists a local platform for multi-node GPU applications and FP32 and mixed-precision algorithms. The 700B-parameter figure assumes NVFP4, and the four-system demonstrations are limited to the listed applications.