Microsoft and NVIDIA Announce New Windows Super-PCs

Microsoft has announced the release of the new DGX Spark and DGX Station personal supercomputers, developed together with NVIDIA. Based on Windows, these devices allow users to run complex AI models locally without relying on cloud services.

DGX Spark: Powerful Supercomputer Weighing Just 1 kg

The compact DGX Spark weighs slightly over 1 kg, consumes 240 W, and delivers performance up to 1 PFLOPS. This allows it to run local AI models with over 200 billion parameters (such as FLUX.1 or Qwen3).

The device is built on the NVIDIA GB10 Grace Blackwell superchip (20-core Arm CPU + Blackwell GPU) with 128 GB of unified memory. The system is priced at over 32 times less than the classic DGX-1.

DGX Station: Data-Center-Level Workstation

The flagship desktop workstation, DGX Station, weighs 40 kg and is powered by NVIDIA GB300 Grace Blackwell Ultra chips. It is designed to handle trillion-parameter models.

It shows performance up to 20 PFLOPS, has 784 GB of unified memory, a ConnectX-8 SuperNIC controller (up to 800 Gbps), direct liquid cooling, and a 1600 W Titanium-certified power supply.

NVIDIA RTX Spark Platform: AI Power for Laptops

At the presentation, an emphasis was also placed on the NVIDIA RTX Spark platform, which combines Grace compute cores, Blackwell RTX graphics with native CUDA support, and shared RAM. This platform already serves as the foundation for the new Microsoft Surface Laptop Ultra notebook.

Top-tier chip configurations (20 CPU cores, 6,144 GPU cores, up to 128 GB RAM) deliver up to 1 PFLOPS and process models with 120+ billion parameters locally. Beyond AI tasks, the devices are suitable for AAA gaming, with confirmed support for the Call of Duty series.