Microsoft is releasing a compact workstation designed to let developers run large AI models on their own desks instead of relying on cloud servers. The Surface RTX Spark Dev Box targets users who need heavy local compute power without the noise and energy costs of traditional data center hardware. This shift matters because it moves the bottleneck of AI development from expensive cloud subscriptions to a single, owned piece of equipment.
Compact workstation targets local AI development with Nvidia Blackwell hardware
The device combines two Nvidia technologies into one chassis. It uses an Arm-based Grace CPU paired with a Blackwell RTX GPU that contains 6,144 CUDA cores. This specific hardware pairing allows the system to handle complex tasks that usually require separate server-grade components.
Specifications
- CPU: Nvidia Grace (Arm-based)
- GPU: Nvidia Blackwell RTX (6,144 CUDA cores)
- Memory: 128 GB unified memory
- Storage: 2 TB
- AI Performance: 1 petaflop FP4
Microsoft allocated 128 GB of unified memory to this system. The memory pool sits between the processor and graphics hardware so both can access the same data quickly. The box also includes 2 TB of storage and ships with a custom Windows 11 Pro build optimized for developers. It comes with tools like WSL 2, CUDA, and VS Code pre-installed.
The system claims to deliver one petaflop of FP4 AI performance. Microsoft says this power allows users to run models with more than 120 billion parameters and context windows reaching one million tokens locally. This capability lets developers tackle model fine-tuning, inference, and long-running agentic workloads that would ordinarily require a cloud GPU instance. The device connects via two USB-C ports, one USB-A port, full-size HDMI, RJ45 Ethernet, and a 3.5 mm headphone jack.
Microsoft has not yet disclosed the USB, HDMI, or Ethernet link speeds. The Surface RTX Spark Dev Box costs $5,999 and will be available for purchase later this year.
Source: Microsoft, NotebookCheck
Discussion
0 comments