The Pulse
NVIDIA puts four-node DGX Spark developer kit on sale
The DGX Spark HPC Developer Kit links up to four systems for local scientific and AI workloads. NVIDIA says an installation playbook for its scientific computing software is coming soon.

NVIDIA's new DGX Spark HPC Developer Kit links up to four desktop systems for researchers developing scientific applications and AI workloads. The kit is on sale and on display at SC26 in Chicago, the company said in a November 16 post; an installation playbook for its scientific computing software is coming soon.
The kit connects DGX Spark systems in a ring, with each system linked directly to its two neighbors. It is aimed at researchers, academics and developers who want to run multi-node workloads locally.
Four systems, no switch
The configuration can scale from one system to four using NVIDIA ConnectX-7 networking, without a separate network switch. The systems use the GB10 Grace Blackwell Superchip.
In the four-node configuration, the kit provides 512GB of combined memory and 20TB of NVMe storage. NVIDIA says the four-node setup can deploy models with up to 700B parameters locally using NVFP4, naming DeepSeek V4.1 as an example.
Researchers can distribute models and scientific workloads across the cluster. NVIDIA says working at a desk can let teams develop and debug without waiting for every job to reach a shared GPU cluster, while keeping unpublished results and proprietary datasets on their own premises.
Scientific software to follow
The forthcoming HPC SDK playbook is intended to bring GPU-accelerated C, C++ and Fortran compilers, libraries and message passing interface tools to the cluster. Teams will be able to build and test applications across four nodes, inspect how computation and data are distributed, and debug communication before moving to larger systems.
NVIDIA says four-system workloads have been demonstrated with LAMMPS and Kokkos for materials simulation and NAMD for molecular dynamics. The wider DGX Spark community has also validated GROMACS for molecular dynamics, OpenFOAM for fluid dynamics and WRF for weather modeling, according to the company.
The community has validated NVIDIA NIM microservices including OpenFold3 and Boltz-2 for predicting biomolecular structures. NVIDIA also describes agent-assisted workflows in which AI prepares simulation inputs, launches runs and summarizes results, while scientists validate the setup and conclusions.
Examples from researchers
Several examples in the announcement use a single DGX Spark rather than the full kit. Kvantify reports that it used NVIDIA cuQuantum to run tensor-network simulations of quantum-chemistry circuits with up to 70 qubits and 975 gates on one system, benchmarking circuits generated for penicillin ground-state energy calculations.
The Rubin Lab at the University of California, Berkeley is using a DGX Spark and NVIDIA's Evo 2 model to predict bacterial traits from genome sequences. NVIDIA says results arrived 15x sooner on DGX Spark without an 11-hour cluster queue.
The Hartwig Group at UC Berkeley is using a single system to fine-tune Nemotron 3 Super 120B-A12B for chemistry machine learning. According to NVIDIA, the group fit a 120B-class model on the device using NVFP4 quantization and ran continuous eight- to 24-hour training fits back-to-back.
Buying the hardware
DGX Spark products and cables are available through NVIDIA Marketplace and original equipment manufacturer partners. The listed partners are Acer, ASUS, Dell Technologies, GIGABYTE, H3C, HP, Lenovo, MSI and xFusion.
Source
Companies in this story
Related stories
- NVIDIA plans Vera CPU racks for scientific computing in 2027
- NVIDIA adds AI workload inventory to GPU platform runtime
- NVIDIA’s 120B Nemotron 3.5 Super VL is available on SageMaker
- Nvidia’s private-company stakes reach $65.5 billion, TechCrunch reports
- NVIDIA reports Vera Rubin gains in MLPerf Training debut