Nvidia has announced two new AI supercomputers, the DGX Spark and DGX Station, during the GTC conference. The DGX Spark, previously known as “Digits,” is a compact yet powerful machine designed for intensive AI tasks. Priced at $3,000, it is being marketed as the “world’s smallest AI supercomputer” and is now available for preorder, with deliveries expected this summer.
This diminutive device is powered by Nvidia’s GB10 Blackwell Superchip, equipped with advanced GPU capabilities including fifth-generation Tensor Cores and FP4 support. Despite its small size, the Spark can perform up to 1,000 trillion operations per second, making it suitable for fine-tuning and inference of sophisticated AI models like the NVIDIA Cosmos Reason and the GR00T N1 robot foundation models. The unit boasts 128GB of unified memory and up to 4TB of NVMe SSD storage.
In contrast, the newly launched DGX Station is aimed at developers, researchers, and students. It features a larger form factor equipped with the powerful GB300 Blackwell Ultra super chip, which delivers an impressive 20 petaflops of AI performance and houses 784GB of unified system memory. While pricing for the DGX Station has not been disclosed, it reflects Nvidia’s commitment to providing robust tools for AI development.
Nvidia has also partnered with several OEM companies, including Asus, Dell, and HP, that will produce custom versions of both the DGX Spark and DGX Station. This strategic move provides greater accessibility for users who require high-performance computing for AI applications. With the increasing demand for GPUs capable of handling large language models, competitors like AMD and HP are also entering the market, signaling a growing trend in AI hardware development.
