Skip to content
Apixo
Blog
news· 3 min read· via NVIDIA blog

NVIDIA Introduces DGX Spark HPC Developer Kit for Local AI and Supercomputing Workloads

NVIDIA unveiled the DGX Spark HPC Developer Kit at SC26, giving researchers a four-node Grace Blackwell cluster to run scientific workloads and models locally.

NVIDIA Introduces DGX Spark HPC Developer Kit for Local AI and Supercomputing Workloads

At the SC26 supercomputing conference in Chicago, NVIDIA introduced the DGX Spark HPC Developer Kit, a compact multi-node cluster engineered to give researchers, academics, and developers local supercomputing capabilities. By enabling teams to run multi-node scientific simulations, large language models, and AI agents directly at their desks, the developer kit aims to eliminate the long queue times typical of shared cluster environments while keeping proprietary datasets securely on-premises.

Hardware Specifications and Network Architecture

The DGX Spark Developer Kit consists of a four-node cluster of NVIDIA DGX Spark systems. Each node is powered by NVIDIA's GB10 Grace Blackwell Superchip. To simplify local deployment, the four nodes connect directly to their two immediate neighbors in a ring topology using NVIDIA ConnectX-7 networking, removing the need to purchase or configure an external network switch.

In its full four-node configuration, the cluster provides 512GB of combined memory and 20TB of NVMe storage. This hardware footprint provides the capacity required to distribute large AI models and complex scientific workloads. Leveraging NVFP4 quantization, researchers can deploy models with up to 700 billion parameters locally across the cluster, including models such as DeepSeek V4.1.

Software Ecosystem and Early Benchmarks

NVIDIA announced that an installation playbook for the NVIDIA HPC SDK is coming soon for the DGX Spark cluster. This toolset brings GPU-accelerated C, C++, and Fortran compilers, optimized libraries, and Message Passing Interface (MPI) tools to the local setup, matching the development environments used on large-scale supercomputers.

The developer kit has demonstrated performance across several scientific computing workloads, including molecular dynamics with NAMD, materials simulation via LAMMPS with Kokkos, and fluid dynamics using OpenFOAM. Additional community-validated applications include GROMACS for molecular dynamics, WRF for weather modeling, and NVIDIA NIM microservices such as OpenFold3 and Boltz-2 for biomolecular structure prediction. NVIDIA also highlighted new open datasets and post-training recipes for Nemotron 3.5 Super VL to help adapt vision-language models for life sciences and chemistry.

Several research teams have published early results using the hardware:

  • Kvantify ran tensor-network simulations of quantum-chemistry circuits with up to 70 qubits and 975 gates on a single DGX Spark node while benchmarking penicillin ground-state energy calculations using NVIDIA cuQuantum.
  • CYRAN AI, an IIT Delhi spin-out, applied the DGX Spark and its PixelFlux library to accelerate local JPEG 2000 decoding for satellite image analysis.
  • Rubin Lab at UC Berkeley used a DGX Spark node running the NVIDIA Evo 2 model to predict bacterial traits from genome sequences, receiving results 15 times faster after avoiding 11-hour shared cluster queues.
  • Hartwig Group at UC Berkeley executed LoRA fine-tuning on the Nemotron 3 Super 120B-A12B model for chemistry machine learning using NVFP4 quantization on a single device, enabling continuous 8- to 24-hour training runs without queue delays.

What it means for developers

For developers and researchers, the DGX Spark HPC Developer Kit offers a dedicated sandbox to build, debug, and optimize multi-node code before deploying jobs to full-scale supercomputers. Teams can test how computational tasks and datasets distribute across nodes and resolve communication issues locally, saving cloud or data center compute allocations for final production runs.

While local physical hardware like the DGX Spark addresses on-premises development and private dataset requirements, developers looking for flexible cloud access can try top AI models cheaply through one API at https://apixoai.online.

The DGX Spark Developer Kit was showcased at SC26 and is available through the NVIDIA Marketplace as well as OEM partners, including Acer, ASUS, Dell Technologies, GIGABYTE, H3C, HP, Lenovo, MSI, and xFusion.


Source: NVIDIA DGX Spark Developer Kit Turns a Desk Into a Supercomputing Lab — NVIDIA blog. Written by the Apixo team from that report.

#ai-news#nvidia#hardware#supercomputing#ai#deepseek
Try it with your own tools

One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.

Get your API key

Keep reading