Microsoft opened preorders on 7 October for its $5,999 Surface RTX Spark Dev Box, a compact computer built for developers working with AI models locally, The Verge reported. The machine uses Nvidia’s Arm-based RTX Spark platform and has 128GB of memory shared between its processors.
Key points
- The Dev Box costs $5,999 and has 128GB of unified memory.
- It is optimised for local AI work involving models with more than 120 billion parameters.
- US preorders are exclusive to Microsoft, with shipping expected to begin in November.
The $5,999 Dev Box at a desk
The price puts the Dev Box above the DGX Spark mini PC Nvidia launched last year, according to The Verge, which also reported that shortages of RAM and other components have been pushing PC prices higher. The new machine is directed principally at developers: its listed uses include testing AI models, prototyping agents and running AI coding tools on the device itself.
Its enclosure has a job beyond holding the components. The flat, 3D-printed anodised aluminium chassis also acts as a heatsink, giving warmth from the hardware a route into the surrounding air. The Verge reported a 100-watt thermal envelope for the Dev Box, a specification that makes the chassis part of the machine’s cooling design rather than just its appearance.
Microsoft is positioning the desktop alongside the Surface Laptop Ultra. They use the same RTX Spark platform, but the Dev Box is intended for local AI work at a developer’s desk, including model evaluation and longer-running experiments, The Gadgeteer reported. Its preorders are available exclusively through Microsoft in the US.
Nvidia’s shared 128GB memory pool
RTX Spark places Nvidia’s Grace CPU and Blackwell RTX GPU around a shared pool of memory. In many computers, the main processor and graphics processor work from separate supplies, so work handed from one to the other can involve moving data between them. In this design, the processors draw on the same pool. The Dev Box has 128GB of unified memory for that arrangement.
That capacity matters for a model whose working data must be held close to the processors using it. A parameter is a numerical value used by a model as it processes a request. The Verge described the Dev Box as optimised for running models with more than 120 billion parameters locally, for work such as testing AI models, prototyping AI agents and running AI coding tools on the device.
Testing an AI model could take place locally rather than through a cloud service when the model and its workload fit the machine’s available memory. Keeping that work on the desktop would let a test use the shared pool described above; whether a particular model fits would depend on the memory it needs while running.
The Surface Laptop Ultra uses the same platform and can be configured with up to 128GB of unified memory, The Gadgeteer reported. For that laptop, the amount available to its GPU is lower than the system total and varies with the configuration and workload. The desktop’s stated 128GB is likewise a total for its shared pool, rather than a separate allotment for each processor.
Project Zenith brings Windows development tools
The Dev Box is among Microsoft’s Project Zenith devices and comes with Windows 11 Pro configured for development. Visual Studio Code, Git, GitHub CLI, GitHub Copilot, WSL, Python and Node are preinstalled. That gives developers a starting set of tools for writing code and working across Windows and Linux environments on the machine.
Microsoft’s accompanying Windows changes also address what AI software can do once it runs. Microsoft Execution Containers became generally available on Windows 11 on 7 October, The Gadgeteer reported. They let organisations set rules for the files and networks an agent may access and enforce those rules while it operates, provided the agent supports the controls.
Windows Search is gaining actions that can change settings or arrange windows, beginning with Windows Insiders in the Experimental channel. Separately, GitHub’s HydraFusion is intended to route work between local and cloud models in GitHub Copilot, Copilot CLI and Visual Studio Code. Those software plans extend the local-development approach beyond the Dev Box hardware.
Microsoft will begin shipping the Surface RTX Spark Dev Box in the US in November.
