Nvidia officially launches the world’s smallest AI supercomputer ‘DGX Spark’ from October 15. The company explained that this product packs data center-level performance into a desktop the size of a hand. It is designed so that companies and individual developers can develop and run their own AI models without expensive AI servers or rental data centers.
The DGX Spark includes Nvidia’s latest AI-dedicated semiconductor, the ‘GB10 Grace Blackwell’ superchip. It also features ultra-fast data transfer technology ‘ConnectX-7’ and Nvidia’s AI operating software ‘DGX Software Stack’. According to the company, this combination implements data center-level processing performance in a small system.
It supports a maximum storage capacity of 128GB, necessary for running AI programs with large computation requirements, such as Large Language Models (LLM) or image-generating AI. Additionally, connecting two Sparks can handle large-scale AI models with up to 405 billion parameters.
Nvidia explains that Spark is part of a new computer product line co-developed with partners like ASUS and Dell. Nvidia emphasized, “Small businesses, startups, and independent developers can now develop and experiment with AI locally without relying on the cloud.”
Jensen Huang, Nvidia’s CEO, stated, “In 2016, we created the DGX-1 to provide supercomputers to AI researchers, and the first system was handed directly to Elon Musk, leading to the creation of ChatGPT.” Huang also noted, “DGX Spark represents another leap by putting AI supercomputers in the hands of all developers.”
Nvidia introduces DGX Spark as the “world’s smallest AI supercomputer.” A higher model, the DGX Station, features a more powerful GB300 ‘Grace Blackwell Ultra’ superchip.
However, DGX Spark and Station are not products used like typical computers. They operate on ‘DGX OS’, an in-house developed Linux-based operating system by Nvidia, optimized solely for creating and training AI models. Standard video games or office programs cannot be run.
The DGX Spark takes on the role of a ‘development system’ within Nvidia’s AI ecosystem. Users can experiment with AI models or create prototypes using Spark, then expand their use on larger DGX servers.
Currently, Nvidia is a leading company in the AI semiconductor market, having built an ecosystem centered around high-performance GPUs and the AI software platform ‘CUDA.’ Recently, the company signed a $100 billion GPU supply contract with OpenAI, which plans to establish a 10 gigawatt (GW)-class AI infrastructure through this deal.
Additionally, cloud computing company CoreWeave has agreed to purchase $6.3 billion worth of GPUs from Nvidia. Nvidia also supplies GPUs to major global IT companies such as Elon Musk’s xAI, Amazon, Google, Meta, Microsoft, and Oracle.
With the release of DGX Spark, Nvidia expects it to broaden AI development accessibility. The company stated, “DGX Spark is a product that supports all developers, not just large corporations, to attempt AI innovation.”