Lifestyle
NVIDIA Launches DGX Spark HPC Developer Kit: Four-Node Ring Cluster, 512GB Memory, Now on Sale
At SC26, NVIDIA unveiled the DGX Spark HPC Developer Kit, built from four DGX Spark systems with 512GB of combined memory and 20TB of NVMe. The kit can be purchased now; the HPC SDK installation playbook is still to come.
News date: About 5 min read
Contents
NVIDIA introduced the DGX Spark HPC Developer Kit at the SC26 supercomputing conference in Chicago. HPC stands for high-performance computing, meaning the use of large amounts of compute to run heavy workloads such as scientific simulation. The kit is aimed at researchers, academics and developers. According to NVIDIA, the kit has been shown and can be purchased, but the HPC SDK installation playbook (the installation and operation guide) has not yet been released. NVIDIA's aim is to let researchers develop and debug multi-node scientific computing and AI workloads at their own desks, without queuing for a shared GPU cluster every time.
What exactly is the DGX Spark HPC Developer Kit?
NVIDIA says the kit is a cluster made of four NVIDIA DGX Spark systems. Each machine is called a "node," and each node connects directly to its two neighbors, forming a ring. The nodes are interconnected with NVIDIA ConnectX-7 networking, so there is no need to buy or configure a network switch. The kit can scale from one node to four, and its core chip is the GB10 Grace Blackwell Superchip.
| Item | As described by NVIDIA |
|---|---|
| Nodes | One to four nodes |
| Connectivity | ConnectX-7 direct links, ring topology, no switch |
| Four-node memory | 512GB combined |
| Four-node storage | 20TB NVMe (high-speed solid-state storage) |
| Model scale | With NVFP4, up to 700B parameters locally (e.g., DeepSeek V4.1) |
| HPC SDK playbook | Coming soon |
| Availability | Shown and available to purchase |
NVFP4 is a low-precision numerical format that reduces the memory a model occupies, so larger models fit on the same hardware. NVIDIA says that with this format, four DGX Spark systems can deploy models of up to 700B parameters locally, such as DeepSeek V4.1.
Is it a full launch, or are some features not yet available?
NVIDIA says the developer kit was shown at SC26 and is on sale. DGX Spark products and cables can be bought from OEM partners including Acer, ASUS, Dell Technologies, GIGABYTE, H3C, HP, Lenovo, MSI and xFusion, and are also available through the NVIDIA Marketplace. NVIDIA has not announced pricing or said anything about availability in specific regions.
The NVIDIA HPC SDK is NVIDIA's software development kit for scientific computing, and its playbook is currently marked as coming soon. NVIDIA says it will then provide GPU-accelerated C, C++ and Fortran compilers, libraries and MPI tools. MPI (message passing interface) is the standard tooling that lets multiple machines pass data to each other and compute together. With these tools, teams can build and test scientific applications on four nodes and debug communication issues before scaling up to larger systems.
What does it mean for research workflows?
NVIDIA notes that researchers with fixed compute and server allocations can develop and debug locally, save server hours for production runs, and keep unpublished results and proprietary datasets in their own environment. NVIDIA demonstrated two four-system workloads: LAMMPS (with Kokkos) for materials simulation, and NAMD for molecular dynamics. NVIDIA also says the DGX Spark community has validated GROMACS, OpenFOAM, WRF, and NIM microservices such as OpenFold3 and Boltz-2 (AI model services that NVIDIA packages for direct deployment).
NVIDIA also cites several user examples: - Kvantify says it used cuQuantum on a single DGX Spark to simulate quantum chemistry circuits of up to 70 qubits and 975 gates. - The Rubin Lab at UC Berkeley IGI says that by avoiding an 11-hour cluster queue, it received results 15 times sooner. - The Hartwig Group at UC Berkeley says it used NVFP4 on a single machine to LoRA fine-tune Nemotron 3 Super 120B-A12B, running continuous eight- to 24-hour training fits back-to-back. LoRA is a more resource-efficient fine-tuning method that adjusts only a small number of parameters. These are self-reported examples from NVIDIA's article and have not been independently verified.
Frequently asked questions
How do you build a four-node cluster without a network switch?
NVIDIA says each of the four DGX Spark systems connects directly to its two neighbors via ConnectX-7, forming a ring, so no switch needs to be bought or configured.
Can I buy it now?
NVIDIA says the kit was shown at SC26 and is on sale, with products and cables purchasable through OEM partners and the NVIDIA Marketplace. NVIDIA has not announced pricing or regional availability.
Can I use the HPC SDK now?
According to NVIDIA, the HPC SDK installation playbook is still "coming soon," and NVIDIA has not given a release date.
Who is this kit for?
NVIDIA positions it for researchers, academics and developers, to develop, debug and run scientific workloads and AI models at their desks before scaling up to data center systems.
Browse the latest news in this topic
Found this guide helpful? Share it with someone who needs it
Lifestyle
Amazon Bedrock Managed Knowledge Base Agentic Retrieve Adds AgentCore Memory: Restore Conversations in One Call and Filter Documents by User ACL
AWS explains that adding memoryConfiguration to AgenticRetrieveStream lets a Bedrock Managed Knowledge Base use AgentCore memory, with userContext applying document ACLs. The integration is opt-in, and the memory resource must be in the same account as the knowledge base.
Lifestyle
GitHub Copilot weekly update: Copilot app supports GitHub Enterprise Server (GA), plus Claude Fable 5.2 and Gemini 4 Carbon for Pro+, Max, Business and Enterprise
GitHub's Copilot update for the week of November 9: the Copilot app now supports GHES repositories (generally available), VS Code 1.146 adds an experimental Inbox mode, and two new models are available to Pro+, Max, Business and Enterprise.
Lifestyle
GitHub Copilot Retires Claude Opus 4.8, Opus 5 and GPT-5.6 Terra on Nov. 16: Switch to Opus 5.5 and Other Models
GitHub announced that from November 16, 2026, three models are deprecated in most Copilot experiences, and it lists replacement options. Individual annual-plan subscribers can still use Claude Opus 4.8, while Enterprise and Business admins should check the default enablement rules.
Lifestyle
Nature Sustainability Study: 403 US Hyperscale Data Centers Use About 82 TWh a Year, 1.8% of US Electricity
According to The Verge, a Harvard team published a study in Nature Sustainability estimating that US hyperscale data centers use about 82 TWh a year, with roughly 54% of that power possibly from natural gas and coal. The figures are estimates based on satellite imagery and grid data.
Sources
Compiled from the references and verification dates listed above; fares, schedules, product features, and terms can change, so please confirm the latest details on the relevant official website before travelling, booking, or purchasing.