Nvidia DGX Station: New AI Supercomputer with GB300 Chip Revealed | Price & Specs

Nvidia has begun selling the DGX Station, a desktop-class AI system representing a significant step toward bringing high-performance computing power typically found in data centers directly to individual users. Priced around $1.9 million Czech crowns (approximately $90,000 USD), the DGX Station isn’t aimed at the average consumer, but rather at developers and enterprises seeking to run demanding AI models locally, without relying on cloud infrastructure. This move addresses a growing need for data privacy, control, and reduced latency in AI development and deployment.

Bringing Data Center Performance to the Desktop

The DGX Station builds upon Nvidia’s earlier DGX Spark model, but with substantially increased capabilities. Both systems are designed for on-premises AI model execution, but the Station leverages the more powerful GB300 chip, compared to the Spark’s GB10. The core of the DGX Station is the GB300 Grace Blackwell Ultra Desktop Superchip, fusing a 72-core Arm Neoverse V2 CPU with a Blackwell Ultra GPU. This combination is linked by Nvidia’s NVLink-C2C interconnect, providing 1.8 terabytes per second of coherent bandwidth – a critical factor in enabling the CPU and GPU to share memory seamlessly and avoid performance bottlenecks. VentureBeat highlights this as a key advancement, potentially marking the most significant personal computing product since the original Mac Pro.

The system boasts 20,480 CUDA cores within the Blackwell Ultra GPU, paired with 252 GB of ultrarapid HBM3e memory (delivering 7.1 TB/s bandwidth). Adding to this, the DGX Station incorporates an additional 496 GB of LPDDR5x memory, accessible to the GPU, creating a unified memory pool of 748 GB. This substantial memory capacity allows the DGX Station to handle AI models with up to one trillion parameters – comparable to the scale of models like GPT-4 – without offloading to slower storage.

Specifications and Performance

Compared to the DGX Spark, the DGX Station represents a considerable performance leap. Here’s a breakdown of key specifications:

The DGX Station delivers up to 20 PFLOPS (petaflops) of FP4 (half-precision floating point) performance, a metric used in many AI workloads. It also achieves 10 PFLOPS in FP6/8, 80 TFLOPS in FP32, and 1.3 TFLOPS in FP64, making it suitable for scientific computing as well. The system supports integer operations at 330 TOPS (trillions of operations per second). Nvidia emphasizes that the DGX GB300, upon which the Station is built, is designed to overcome performance bottlenecks in AI infrastructure.

Implications for AI Development and Deployment

The DGX Station addresses a critical tension in the AI industry: the need for powerful infrastructure versus the desire for localized data and intellectual property. Currently, many AI models require massive data center resources, but developers increasingly want to keep their data, agents, and proprietary information secure and under their direct control. The DGX Station offers a solution by bringing that level of computing power to a single workstation. This is particularly relevant for organizations handling sensitive data, such as financial institutions, healthcare providers, and government agencies.

Beyond data security, the DGX Station’s local processing capabilities reduce latency, which is crucial for real-time applications like autonomous vehicles, robotics, and interactive AI agents. The ability to run trillion-parameter models on a desktop also democratizes access to advanced AI capabilities, potentially accelerating innovation in various fields. The system also offers flexibility with four NVMe SSD slots and the ability to connect up to three DGX Stations via a 400Gb/s ConnectX network.

Availability and Ecosystem

While Nvidia is directly selling the DGX Station, the company is also partnering with major system integrators like Asus, Dell, Gigabyte, HP, MSI, and Supermicro to offer pre-configured and customized solutions. This broader availability ensures that customers can choose a system tailored to their specific needs and receive support from established hardware vendors. The DGX Station is now available for pre-order, with initial shipments expected to begin shortly.

What Comes Next

The launch of the DGX Station marks a significant milestone in the evolution of AI infrastructure. However, several factors will shape its future adoption. Continued optimization of AI models to leverage the Station’s unique architecture will be crucial. The development of software tools and frameworks that simplify deployment and management of large-scale models on desktop systems will be essential. Finally, the broader industry response – including the emergence of competing desktop AI systems – will determine the long-term impact of Nvidia’s innovation.