NVIDIA celebrates the launch of the world's smallest AI supercomputer! Jen-Hsun Huang personally delivered the first batch to Musk at SpaceX.

NVIDIA officially launched the world's smallest AI supercomputer DGX Spark this week, marking a new chapter in AI computing moving towards desktop integration. This revolutionary device integrates NVIDIA's most powerful AI technology stack in a compact desktop form, providing developers with the capability for local training and inference of large language models.

To celebrate the launch of the first batch of DGX Spark, NVIDIA founder and CEO Jensen Huang personally visited SpaceX headquarters in Texas to deliver the first batch of DGX Spark to SpaceX founder Elon Musk, recreating the historic moment when Huang personally delivered the world's first DGX-1 supercomputer to Musk in 2016, symbolizing the continuation of the AI revolution.

The world's smallest AI supercomputer, NVIDIA DGX Spark, is about to be released.

The rapid development and demand for AI have exceeded the workloads that millions of developers previously relied on PC desktops, workstations, and laptops to handle, forcing teams to shift their work to the cloud or local data centers. DGX Spark, as a new type of computer, offers AI performance with trillions of floating-point operations and 128GB of unified memory in a compact and sleek computer form factor, enabling developers to perform inference on AI models with up to 200 billion parameters locally and fine-tune models with up to 70 billion parameters. Furthermore, DGX Spark also allows developers to create AI agents locally to run advanced software stacks.

Jensen Huang personally delivered the DGX Spark to Musk at SpaceX.

Jensen Huang stated that in 2016, NVIDIA created the DGX-1, a dedicated supercomputer for AI researchers. He personally delivered the first system to Elon Musk, who was then starting OpenAI, which led to the birth of ChatGPT and initiated the AI revolution. The first generation of the supercomputer launched in 2016 weighed 134 pounds and cost $129,000, while the current weight is 2.65 pounds and the price is only $3,999. The GPU architecture has also evolved from NVIDIA Pascal to Blackwell.

Other early adopters of DGX Spark, including Anaconda, Cadence, ComfyUI, Docker, Google, Hugging Face, JetBrains, LM Studio, Meta, Microsoft, Ollama, and Roboflow, are testing the validation and optimization of their tools, software, and models for DGX Spark. Global research institutions, including NYU's Global Frontier Lab, have previewed DGX Spark to facilitate their AI development. Kyunghyun Cho, a professor of Computer and Data Science at NYU's Global Frontier Lab, stated that the new approach of DGX Spark allows for rapid prototyping and experimentation with advanced AI algorithms and models, even for privacy and security-sensitive applications, such as healthcare.

Application scenarios of DGX Spark

DGX Spark integrates the complete NVIDIA AI platform (including GPU, CPU, networking, CUDA® libraries, and NVIDIA AI software stack) into a compact system that is small enough to fit in a lab or office, yet powerful enough to accelerate both agent and physical AI development. By combining groundbreaking performance with the breadth of the NVIDIA ecosystem, DGX Spark transforms the desktop into an AI development platform.

The DGX Spark system offers up to 1 petaflop of AI performance, accelerated by the NVIDIA GB10 Grace Blackwell super chip, NVIDIA ConnectX®-7 200 Gb/s networking, and NVIDIA NVLink-C2C technology, providing 5 times the bandwidth of fifth-generation PCIe with memory and CPU consistency consistent with PCIe.

Pre-installed NVIDIA AI software stack enables developers to get started with AI projects out of the box. With DGX Spark, developers can access NVIDIA AI ecosystem tools, including models, libraries, and NVIDIA NIM microservices, to facilitate local workflows, such as customizing the Black Forest Labs FLUX.1 model for optimizing image generation, creating visual search and summary agents using the NVIDIA Cosmos Reason visual language model, or targeting the DGwen X3 summary robot version.

This article celebrates the launch of the world's smallest AI supercomputer by NVIDIA! Jensen Huang personally delivered the first batch to Musk at SpaceX. It first appeared on Chain News ABMedia.