Nvidia has launched its smallest workstation yet, the DGX Spark, designed for AI and robotics developers, data scientists, and machine learning researchers. This compact system features a Blackwell GPU capable of delivering up to 1 petaFLOP of compute power, paired with 128GB of unified system memory and high-speed networking.
The DGX Spark’s GB10 system-on-a-chip is powered by two compute dies connected at 600 GB/s via Nvidia’s proprietary NVLink chip-to-chip interconnect tech. This allows for efficient processing of large models up to 200 billion parameters in size, which are commonly used in AI workloads. The system ships with a custom spin of Ubuntu Linux and does not come with Windows.
With a starting price tag of around $3,000, the DGX Spark is a high-end workstation that caters to professionals who require powerful computing capabilities for their work. Its integrated networking card enables fast data transfer and connection to other systems, making it ideal for collaborative projects. The system will be available for purchase from various OEM partners starting October 15.
While the price may seem steep, the DGX Spark offers unmatched performance for its size, making it a valuable asset for those in AI-driven fields.
Source: https://www.theregister.com/2025/10/13/nvidia_gb10_spark