Skip to content
0MESubscribe
AI16 Nov 20262 min read

Nvidia Launches Desktop Supercomputing Kit That Runs 700B AI Models

Nvidia unveiled a four-node desktop supercomputer kit at SC26, giving scientists 512GB of combined memory to run massive AI models without waiting in data center queues.

Listen to this storyAbout 3 min, read by your device
Picture made with AI

Nvidia introduced the DGX Spark HPC Developer Kit at the SC26 supercomputing conference in Chicago. The hardware package connects four compact computers directly together in a ring, placing multi-node supercomputing power directly on an office desk.

Supercomputing Without the Shared Queue

Researchers typically must wait in lines to secure time on large, shared GPUGPUA chip that does many small sums at once. First made for graphics, now the main chip for AI. Open in glossary clusters. The new kit aims to eliminate those delays by providing dedicated desktop hardware powered by the GB10 Grace Blackwell SuperchipSuperchipA single processor module combining central processing cores and graphics hardware closely together. Open in glossary. The four nodes link directly through ConnectX-7 networking, which removes the need to buy or configure a separate network switch.

Combined, the four-system setup provides 512GB of memory and 20TB of fast NVMeNVMeA fast hardware standard used for connecting solid-state storage to computers. Open in glossary storage. When using NVFP4 precision, researchers can run AI models with up to 700 billion parameters locally, such as DeepSeek V4.1. This allows teams to test, debug, and run workloads locally while keeping sensitive data and unpublished findings private.

Scientific Simulation and Early Results

The kit supports complex scientific simulations and AI tools, including molecular dynamics software like NAMD and GROMACS, fluid dynamics through OpenFOAM, and weather modeling via WRF. Nvidia stated that a setup guide for its HPC SDK software tools is coming soon, bringing compilers and messaging libraries used on full-scale supercomputers to the desktop.

Several research groups have already put the systems to work. The Rubin Lab at UC Berkeley used a single nodeNodeA single computer system within a larger connected group of computers. Open in glossary to predict bacterial traits from genetic sequences, cutting an 11-hour cluster wait time and finishing results 15 times faster. The Hartwig Group, also at UC Berkeley, applied NVFP4 quantization to fine-tune a 120-billion-parameter chemistry model locally, bypassing multi-day server queues.

The developer kit is available for purchase through Nvidia Marketplace and computer manufacturers including Dell Technologies, HP, Lenovo, ASUS, and Acer.

How does this story land?

One tap per reader.

Quick poll

Would your team benefit from desktop supercomputing clusters over cloud allocations?

One tap. No sign-up.

Ask this story

Still have a question?

Answers use only what is in this story. They are written by AI and can be wrong.

Sources

  1. - Archives Page 1 | NVIDIA Blog, NVIDIA DGX Spark Developer Kit Turns a Desk Into a Supercomputing Lab
Companies in this storyDell TechnologiesHPNvidiaWords explainedSuperchipGPUNVMeNode
Up next in AINvidia Vera Rubin delivers 10x throughput per watt in debut MLPerf AI tests2 min read

Keep reading

All AI stories