<h1>Nvidia Unveils ‘Personal AI Supercomputers’ DGX Spark and DGX Station at GTC 2025</h1>
<p>Nvidia has officially announced two groundbreaking AI personal supercomputers, the DGX Spark and DGX Station, at the GTC 2025 conference. Powered by the Grace Blackwell chip platform, these innovative systems are designed to cater to developers, researchers, and data scientists, enabling them to prototype, fine-tune, and run AI models locally.</p>
<h2>DGX Spark</h2>
<p>The DGX Spark is equipped with the GB10 Grace Blackwell Superchip and offers remarkable performance, delivering up to 1,000 trillion operations per second (TOPS). This compact AI desktop includes 128 GB of memory and is targeted primarily at edge computing. The DGX Spark is available for reservations today, and its price tag is estimated to start around $3,000. Shipments are expected to commence in the summer of 2025.</p>
<h2>DGX Station</h2>
<p>The DGX Station, on the other hand, features the more powerful GB300 Grace Blackwell Ultra Desktop Superchip, boasting up to 20 petaFLOPS of AI compute and 784 GB of unified memory. This makes the DGX Station particularly suitable for tasks demanding intensive AI performance. While reservations for this model are anticipated later this year, availability will follow in late 2025.</p>
<h2>OEM Partnerships and Expansion</h2>
<p>Nvidia has partnered with prominent manufacturers such as Asus, Dell, and HP to produce and distribute these supercomputers. Noteworthy models include Asus’s Ascent GX10, HP’s ZGX Nano AI Station G1n, and Dell’s Pro Max series, each integrating the powerful Grace Blackwell platform.</p>
<h2>Focus on Inference and New Software</h2>
<p>Alongside the hardware announcements, Nvidia has shifted its focus towards advancing inference with the introduction of several new AI accelerators in the Blackwell Ultra family. Complementing this hardware, the open-source inference software Dynamo and the highly accurate reasoning model Llama Nemotron were also announced, emphasizing Nvidia’s commitment to enhancing machine learning capabilities on GPUs.</p>
<h2>RTX PRO 6000-Series GPUs</h2>
<p>Additionally, Nvidia unveiled the RTX PRO 6000-series GPUs, tailored for server and workstation applications. These GPUs promise significant improvements in floating-point performance, memory bandwidth, and throughput with the latest advancements in Nvidia’s Streaming Multiprocessor and RT Core technologies. They will be available starting April 2025 from providers like Dell, HP, and Lenovo.</p>
<p>These announcements underscore Nvidia’s strategic vision in AI computing. With the introduction of the DGX Spark and DGX Station, Nvidia aims to meet the growing computational demands of AI development and deployment, reinforcing its dominant position in the AI hardware market.</p>


Collections

Nvidia announced the DGX Spark and DGX Station, two powerful personal AI supercomputers, at GTC 2025. The DGX Spark is designed for edge computing with 128 GB of memory, while the DGX Station offers up to 20 petaFLOPS of AI compute and 784 GB of memory for high-performance tasks. Nvidia also introduced new AI accelerators, the open-source inference software Dynamo, and the highly accurate reasoning model Llama Nemotron. Additionally, the RTX PRO 6000-series GPUs were unveiled for server and workstation applications, emphasizing significant performance improvements.

Nvidia Introduces High-Performance AI Workstations: DGX Station and DGX Spark