News
NVIDIA Unveils DGX Spark HPC Developer Kit at SC26
NVIDIA introduces the DGX Spark HPC Developer Kit, a four-node cluster enabling researchers to run AI and scientific workloads directly from their desks.
At SC26, the annual supercomputing conference held this week in Chicago, NVIDIA unveiled the DGX Spark HPC Developer Kit, a compact supercomputing solution designed for researchers, academics, and developers. The kit enables users to run multi-node scientific workloads and AI models directly from their desks, eliminating the need to wait for shared GPU cluster access.
Technical Specifications and Capabilities
The DGX Spark HPC Developer Kit consists of a four-node cluster of NVIDIA DGX Spark systems, each connected in a ring topology using NVIDIA ConnectX-7 networking. This configuration requires no additional network switches, simplifying setup. The cluster offers a combined 512GB of memory and 20TB of NVMe storage, providing ample capacity for distributing models and scientific workloads.
Powered by the GB10 Grace Blackwell Superchip, the system supports deploying AI models with up to 700 billion parameters using NVFP4 quantization. This capability allows researchers to run complex workloads such as DeepSeek V4.1 locally. Additionally, the forthcoming NVIDIA HPC SDK playbook will provide tools for GPU-accelerated development in C, C++, and Fortran, as well as message-passing interfaces for multi-node applications.
Applications in Research and Development
The DGX Spark Developer Kit has already demonstrated its utility across various scientific domains. For example, it supports molecular dynamics simulations using tools like LAMMPS with Kokkos and NAMD, as well as fluid dynamics modeling with OpenFOAM and weather simulations with WRF. In life sciences, researchers can adapt models like Nemotron 3.5 Super VL for secure workflows in chemistry and biology.
Specific use cases include the Rubin Lab at UC Berkeley, which used the kit to predict bacterial traits from genome sequences 15 times faster than traditional cluster-based methods. Similarly, the Hartwig Group at UC Berkeley employed the system for LoRA fine-tuning of chemistry machine learning models, bypassing multi-day cluster queues and enabling continuous training sessions.
Other notable applications include Kvantify's quantum-chemistry simulations with NVIDIA cuQuantum and CYRAN AI's geospatial analysis using accelerated JPEG 2000 decoding for satellite imagery.
Availability and Future Developments
The DGX Spark Developer Kit is currently on display at SC26 and available for purchase through NVIDIA Marketplace and OEM partners such as Dell Technologies, HP, Lenovo, and ASUS. NVIDIA plans to release the HPC SDK playbook soon, further enhancing the kit's capabilities for local scientific computing and AI development.