Korean AI silicon vendor DEEPX has rolled out the DX-AIPlayer, a palm-sized edge AI box that pairs an Intel Processor N97 with the company's DX-M1 neural accelerator on an M.2 card. Aimed at machine vision workloads in robotics, smart cities, and factory floors, the chassis squeezes a full x86 host and a dedicated NPU into a 95 x 95 x 55 mm (3.7 x 3.7 x 2.2 inches) enclosure weighing 450 grams (1.0 lbs).
The star of the show is the DX-M1 module, a 25 TOPS INT8 accelerator that drops into an M.2 2280 M-Key slot over PCIe Gen 3 x4. Crucially, the card carries its own 4GB of LPDDR5 memory, so multi-model inference does not eat into the host's RAM budget, and it pulls between 1 and 5 watts depending on load. That leaves the quad-core Alder Lake-N N97 (boosting to 3.6 GHz with Intel UHD graphics) free to handle the OS, I/O, and application logic at its 12W TDP.
The rest of the spec sheet is built for industrial deployment. The base configuration ships with 8GB of LPDDR5 and 64GB of eMMC, expandable to 16GB and 128GB respectively. Display output covers HDMI 2.0b and DisplayPort 1.2, networking is handled by dual Gigabit Ethernet with an optional Wi-Fi/Bluetooth M.2 2230 E-Key slot, and there are three USB 3.2 Gen 2 Type-A ports plus a pair of internal USB 2.0 headers. Two RS-232/422/485 COM ports, a TPM 2.0 chip, VESA mounting, a 12V locking barrel jack, and an operating range of 0°C to 60°C (32°F to 140°F) with 0.5 m/s airflow round out the package. DEEPX rates MTBF at 1,224,238 hours.
Software support spans Windows 10/11, Ubuntu 20.04/22.04/24.04 LTS, and Yocto Project 5.1. Developers get the DX-AllSuite toolkit for compiling, optimizing, and running models locally or in Docker, with the DX-RT inference engine exposing C++ and Python APIs. PyTorch, TensorFlow, ONNX, Keras, and Ultralytics YOLO are all on the supported list. The low-level side is covered by the open-source dx_rt_npu_linux_driver, a PCIe kernel module that handles NPU access via the standard Kbuild toolchain. The DX-M1 also implements the OAAX (Open AI Accelerator eXchange) interface, a Linux Foundation AI and Data project that defines a vendor-neutral runtime layer so compliant accelerators can be swapped in without framework-level integration work.
The fully assembled DX-AIBOX-M1-N97 with 8GB RAM is listed at $995 (€916), with shipments estimated to begin 2026-07-17. Further details are available on the DEEPX product page.
