DataCentreNews India - Specialist news for cloud & data centre decision-makers
India
Nvidia launches desk-side DGX Spark cluster for researchers

Nvidia launches desk-side DGX Spark cluster for researchers

Fri, 20th Nov 2026 (Today)
Mara Sugue
MARA SUGUE News Editor

Nvidia has introduced the DGX Spark HPC Developer Kit, aimed at researchers, academics and developers.

The kit is a four-node cluster of DGX Spark systems linked directly to neighbouring nodes in a ring using ConnectX-7 networking. The design removes the need for a separate network switch and lets users run multi-node workloads from a desk rather than relying on shared GPU clusters.

At full scale, the four-node setup provides 512GB of combined memory and 16TB of NVMe storage. It is intended for scientific workloads, AI models and agent-based software, with a playbook for installing the Nvidia HPC SDK due to follow.

The hardware uses the GB10 Grace Blackwell Superchip, extending the DGX Spark line from a single machine to a small clustered setup. Nvidia said the cluster can be used to build and test applications across four nodes, examine how data and compute are distributed, and identify communication issues before moving work to larger systems.

Local development

Nvidia is positioning the developer kit as a way to reduce waiting times for shared computing resources. That could appeal to research teams with fixed compute allocations or groups that want to keep unpublished results and proprietary datasets on local infrastructure.

With NVFP4, users can deploy models with up to 700 billion parameters locally across the four-node cluster, including DeepSeek V4.1, Nvidia said. The upcoming HPC SDK playbook will also bring GPU-accelerated C, C++ and Fortran compilers, libraries and message passing interface tools to the platform.

Nvidia also cited software already demonstrated or validated on DGX Spark systems, including LAMMPS with Kokkos for materials simulation, NAMD and GROMACS for molecular dynamics, OpenFOAM for fluid dynamics, WRF for weather modelling, and Nvidia NIM microservices including OpenFold3 and Boltz-2 for biomolecular structure prediction.

Research use

Nvidia outlined several examples of researchers using DGX Spark systems in practice, spanning quantum chemistry, geospatial image processing, structural biology and chemistry machine learning.

Kvantify is using Nvidia cuQuantum and reported running tensor-network simulations of quantum-chemistry circuits with up to 70 qubits and 975 gates on a single DGX Spark. Nvidia said the benchmarks were based on circuits generated for penicillin ground-state energy calculations.

CYRAN AI, a spin-out from IIT Delhi, is using DGX Spark and its PixelFlux library to speed up JPEG 2000 decoding for large satellite images. According to Nvidia, the work supports local geospatial analysis.

OpenFold has also shown how researchers can fine-tune OpenFold3 on DGX Spark to improve protein-ligand structure predictions for drug discovery. In another example, the Rubin Lab at the Innovative Genomics Institute at the University of California, Berkeley, is using a DGX Spark and the Nvidia Evo 2 model to predict bacterial traits directly from genome sequences.

Nvidia said the Rubin Lab compared embeddings from genomic and protein language models with classical baselines and produced results 15 times sooner by avoiding an 11-hour cluster queue. The Hartwig Group at the same university is using a DGX Spark for LoRA fine-tuning on the Nemotron 3 Super 120B-A12B model for chemistry machine learning, training a prediction head on experimental values.

According to Nvidia, the Hartwig Group was able to fit a 120B-class model on a single device using NVFP4 quantisation and avoid multi-day waits for shared cluster access. That allowed the lab to run training sessions lasting eight to 24 hours back to back on a local machine.

Desktop systems

The launch reflects a broader push to move some high-performance computing and AI development closer to individual research teams. Rather than replacing large shared supercomputers, the desktop cluster is intended to support development, debugging and smaller-scale runs before work is moved to larger data-centre infrastructure.

The system is available through original equipment manufacturer partners including Acer, Asus, Dell Technologies, Gigabyte, H3C, HP, Lenovo, MSI and xFusion, Nvidia said. It added that the forthcoming software playbook will give scientists a local platform for developing multi-node GPU applications and running simulation workloads, including LAMMPS with Kokkos, before scaling to larger systems.