
DGX Spark HPC Developer Kit: A 4-Node Cluster for Your Desk
NVIDIA's DGX Spark HPC Developer Kit links four Sparks in a ring for 512GB of memory and 20TB of storage, enough to run 700B-parameter models locally.
NVIDIA used the SC26 supercomputing conference in Chicago to announce the DGX Spark HPC Developer Kit, a four-node cluster of DGX Spark mini PCs aimed at researchers, academics and developers. Announced on November 16, 2026, the kit turns four of NVIDIA's palm-sized AI boxes into a desk-sized machine for multi-node scientific workloads, large local models and AI agents, without a trip to the shared cluster queue.
- Configuration: four DGX Spark systems, each wired directly to its two neighbors in a ring
- Networking: NVIDIA ConnectX-7 links, with no network switch to buy or configure
- Capacity: 512GB of combined memory and 20TB of NVMe storage at four nodes
- Model size: up to 700B parameters locally using NVFP4 quantization, according to NVIDIA
What Is in the DGX Spark HPC Developer Kit?
Each node is a DGX Spark built around the GB10 Grace Blackwell Superchip. The kit scales from one system to four, and the ring topology is what keeps it simple: every Spark connects straight to the next over ConnectX-7, so the cluster needs no switch.
At full size the four nodes pool 512GB of memory and 20TB of NVMe storage. NVIDIA says that is enough to deploy models of up to 700B parameters across the cluster with NVFP4, and names DeepSeek V4.1 as an example. For anyone running local LLM workloads, that moves the ceiling well past what a single mini PC can hold.
The software piece is still arriving. NVIDIA says an installation playbook for its HPC SDK is coming soon, bringing GPU-accelerated C, C++ and Fortran compilers, libraries and message passing interface tools to the cluster. Those are the same development tools used on NVIDIA-powered supercomputers.
Who Is the DGX Spark Cluster For?
The pitch is aimed at researchers who share a GPU cluster and spend much of their week waiting for it. With a four-node kit on the desk, a team can build and debug multi-node applications locally, check how computation and data are distributed, sort out communication issues, and then spend its allocated server hours on production runs. NVIDIA also notes that unpublished results and proprietary datasets stay on premises.
Demonstrated four-system workloads include LAMMPS with Kokkos for materials simulation and NAMD for molecular dynamics. NVIDIA says the wider DGX Spark community has validated GROMACS, OpenFOAM for fluid dynamics and WRF for weather modeling.
What Are Researchers Already Doing With DGX Spark?
NVIDIA's announcement lists several early projects, all reported by the company and its partners:
- Quantum chemistry: Kvantify reports tensor-network simulations of circuits with up to 70 qubits and 975 gates on a single DGX Spark
- Genomics: the Rubin Lab at UC Berkeley uses a DGX Spark with the Evo 2 model to predict bacterial traits from genome sequences, with results arriving 15x sooner than waiting through 11-hour cluster queues
- Chemistry: the Hartwig Group at UC Berkeley fit a 120B-class Nemotron model on one device for LoRA fine-tuning, running training jobs of eight to 24 hours back to back
- Satellite imagery: CYRAN AI, an IIT Delhi spin-out, accelerates JPEG 2000 decoding for large images to support local geospatial analysis
The common thread is iteration speed. None of these jobs needs a supercomputer for every step, but each one benefits from a local machine that is always available.
Price and Availability
The developer kit is on display at SC26 and available for sale. NVIDIA says DGX Spark systems and cables can be purchased from OEM partners including Acer, ASUS, Dell Technologies, GIGABYTE, H3C, HP, Lenovo, MSI and xFusion, as well as through NVIDIA Marketplace. The announcement does not list a bundle price for the four-node kit, so budget from the per-unit cost of the systems you choose.
If you are sizing up a single node first, our guide to who the 64GB DGX Spark is for covers the entry model, and our Mac Studio M5 Ultra vs DGX Spark comparison looks at local AI speed. There is more compact computing in our mini PC coverage.
Sources: NVIDIA Blog — November 16, 2026.
More Mini Computers Stories

Flipper One Kickstarter: $399 Linux Multi-Tool Ships 2027
Flipper One is now on Kickstarter at $399, with an RK3576, 8GB RAM, Wi-Fi 6E and LoRa, 5G and SDR add-ons. First units are estimated for July 2027.

RTX Spark Desktops Arrive: What $4,699 Buys for Local AI
RTX Spark desktops from ASUS, Dell and MSI are on sale with up to 128GB of unified memory. MSI's mini PC starts at $4,699. Here is who should consider one.

TINY 2.1: A DOS Transformer Chatbot That Fits on a Floppy
TINY 2.1 packs a 367,000-parameter transformer into 380KB of weights. It runs under DOS on a 486DX or Pentium, and you can retrain it with your own facts.
