In brief
- The kit consists of four DGX Spark systems connected directly in a ring.
- Together, the four nodes provide 512 GB of memory and 20 TB of NVMe storage.
- NVIDIA says models with up to 700 billion parameters can run across the four systems using NVFP4.
- The kit is now available; a setup and usage guide for the NVIDIA HPC SDK is expected soon.
NVIDIA announced the DGX Spark HPC Developer Kit this week at SC26, the supercomputing conference in Chicago. Designed for researchers, academics and developers, the kit brings four DGX Spark systems together in a desktop cluster. NVIDIA says it showcased the kit at the conference and made it available for purchase.
How do four DGX Spark systems form a cluster?
Each system connects directly to two neighboring nodes to create a ring network. Using NVIDIA ConnectX-7 connectivity, the setup requires no separate network switch. The system can be used as a single node and expanded to up to four nodes as needed.
According to NVIDIA, the four systems together provide 512 GB of memory and 20 TB of NVMe storage. The company says models with up to 700 billion parameters can run locally on the cluster using NVFP4, citing DeepSeek V4.1 as an example. The source does not provide comparative performance test results against other hardware for this capacity.
What’s included for simulation and software?
The kit is positioned for scientific simulations as well as work with AI models and agents. Four-system use cases shown at SC26 included Kokkos with LAMMPS for materials simulation, and NAMD for molecular dynamics. NVIDIA also says software such as GROMACS, OpenFOAM and WRF has been validated in the DGX Spark community.
The setup and usage guide for the NVIDIA HPC SDK has not yet been published; the company says it will be available soon. The guide is intended to help bring GPU-accelerated C, C++ and Fortran tools, libraries and message-passing interface components to the cluster. The goal is to develop and test applications in a multi-node environment, and investigate deployment or communication issues before moving to larger systems.
What it means for architecture and visualization workflows
The announcement does not describe the kit as a solution specifically for architectural modeling or visualization software. It is therefore not possible to conclude that it is a direct rendering accelerator for architecture firms and archviz teams. However, research or design teams working with scientific simulations and local AI could assess the option of developing and testing their workloads on infrastructure under their own control.
Before using it in an architecture-related workflow, teams should separately verify whether their target software and computing tasks are suited to this multi-node setup. The source does not specify compatibility with architectural applications, hardware requirements or licensing terms. Teams should therefore clarify their use cases before comparing the kit with their current workstations and software licenses.
Sales channels and the upcoming guide
NVIDIA says the kit was showcased at SC26 and is now available. DGX Spark systems and cables can be purchased through the NVIDIA Marketplace and from partners including Acer, ASUS, Dell Technologies, GIGABYTE, H3C, HP, Lenovo, MSI and xFusion. The announcement does not include pricing.
Once the NVIDIA HPC SDK guide is published, teams will be able to find more details on developing and debugging scientific applications on a local cluster. In particular, seeing how the four nodes work with software and workloads will be important when evaluating a move to larger systems.
Sources
1 sourceSource texts are not republished; short quotes are marked, everything else is our own summary and commentary.
The DGX Spark HPC Developer Kit’s main appeal is its goal of enabling research and development without having to wait for access to a shared GPU cluster. Teams with simulation or local AI workloads could use it to experiment and debug before moving to larger infrastructure. However, the use cases in the announcement are not specific to architectural software, so it would be misleading to view the product as a direct alternative to a rendering workstation.
For offices in Turkey, total cost, software compatibility and licensing terms will be key factors in a purchasing decision. Since the source provides neither pricing nor compatibility details for architectural applications, it would be sensible to wait for these points to be clarified and for the SDK guide to be published before calculating the investment.
Frequently asked questions
How many systems make up the DGX Spark HPC Developer Kit?
The kit brings together four DGX Spark systems connected directly in a ring. It uses NVIDIA ConnectX-7 connectivity.
What are the memory and storage capacities of the DGX Spark HPC Developer Kit?
According to NVIDIA, the four-node setup provides a total of 512 GB of memory and 20 TB of NVMe storage.
How much does the DGX Spark HPC Developer Kit cost?
The source text of NVIDIA’s announcement does not include pricing. It says the kit is now available.



