Sintrones VBOX-3631: an in-vehicle edge AI computer on Intel Wildcat Lake

The Sintrones VBOX-3631 is a fanless in-vehicle edge AI computer built on Intel Core Series 3 processors, codenamed Wildcat Lake, with a built-in NPU rated up to 17 TOPS. It carries E-Mark approval, and we supply it for fleet management, intelligent transportation and defence logistics across automotive and railway environments.

The launch signals where in-vehicle compute is heading: AI inference is moving from add-in accelerator cards into the host processor itself. The VBOX-3631 carries no discrete GPU or separate accelerator module for its baseline AI capability; the NPU sits on the Wildcat Lake die, which changes the thermal, power and form-factor maths for a fanless box. For fleet and transit operators specifying perception or analytics at the vehicle, that integration is the headline.

The VBOX-3631 is among the first rugged in-vehicle computers announced on Intel Core Series 3. The Core 7 350 and Core 5 320 pair two performance cores with four low-power efficiency cores; the Core 3 304 uses one performance core with four. Processor base power is 28 W. That mix lets the system drop into a low-power mode for routine tasks and scale up to turbo for demanding edge workloads, which matters in a fanless box where the sustained thermal budget is finite.

The integrated NPU provides up to 17 TOPS for inference, giving three compute engines on one die (CPU, GPU and NPU) to share perception, analytics and application work. That is a mid-range inference budget, well suited to single-stream or modest multi-stream perception, in-cabin analytics, driver or operator monitoring and classification. It is not pitched at heavy multi-camera cascades that demand a discrete GPU or several dedicated accelerators.

A discrete accelerator, such as a Hailo M.2 module or a GPU, is separate silicon on PCIe. An on-die NPU avoids that cost, power and mechanical overhead, at the trade-off of a fixed inference budget that cannot be expanded by adding more modules. Our comparison of GPU and NPU for edge AI covers when each makes sense.

Heavy multi-camera perception, large-model inference or workloads needing tens to hundreds of TOPS still point towards edge AI computers with discrete GPUs or multiple dedicated accelerators. The VBOX-3631 is deliberately not that box. Within the same family, the VBOX-3650 steps up to Intel Core Ultra Series 3.

Because the NPU is on-die, the VBOX-3631 keeps a single-chip thermal and power profile, which is what lets the fanless design, the wide temperature range and a compact chassis coexist with real AI capability. With wide-voltage input, ignition control and M12 options, it is built to drop into in-vehicle and railway deployment without a ruggedisation programme around it.

For railway or other certified deployments, we confirm the current certificate status for the exact build before you order.