Geniatech has rolled out the APC888, an edge AI box PC built around NXP's i.MX 95 applications processor that lets integrators pick their own neural accelerator through a standard M.2 socket. Rather than locking buyers into a single inference engine, the chassis accepts modules from Hailo, MemryX, DeepX, or Kinara (recently acquired by NXP), with options reaching up to 40 TOPS for heavier vision workloads.

The NXP i.MX 95 does the heavy lifting on the application side, combining up to six Arm Cortex-A55 cores at 1.8 GHz with a Cortex-M7 real-time core, a Cortex-M33 safety core, and an Arm Mali-G310 V2 GPU supporting OpenGL ES 3.2 and OpenCL 3.0. An integrated eIQ Neutron NPU handles 2 TOPS of inference on its own at 750 inferences per second, while the VPU manages 4Kp30 and 1080p60 H.265 and H.264 encode and decode. The base configuration ships with 4GB of LPDDR5 and 32GB of eMMC storage, with 64GB and 128GB flash options on the menu.

Connectivity leans toward industrial deployment. Two Gigabit Ethernet ports sit alongside a pair of USB 3.0 ports, a USB Type-C connector, and optional WiFi, 4G LTE, and GNSS radios. The expansion header exposes MIPI-DSI/CSI, PCIe, CAN, UART, SPI, I²C, PWM, and GPIO lines, and the metal enclosure measures 20.3 x 17.4 x 6.7 cm (8.0 x 6.9 x 2.6 inches) at 1.25 kg (2.8 lbs). Power runs from a 12V/1.5A locking DC jack, and buyers can choose commercial (0°C to 60°C / 32°F to 140°F) or industrial (-40°C to 85°C / -40°F to 185°F) temperature grades.

Geniatech is targeting multi-camera video analytics, machine vision, robotics perception, and retail intelligence workloads, with Linux support delivered through Yocto 5.0 Scarthgap. NXP maintains the meta-imx Yocto BSP layer publicly on GitHub and folds eIQ, its machine learning software environment with TensorFlow Lite integration, directly into the standard BSP image for use alongside the on-die Neutron NPU. The M.2 accelerator slot benefits from a notably open ecosystem as well: Hailo publishes HailoRT under the MIT license (with a GPL v2 PCIe driver), and MemryX releases its MX3 Linux kernel driver and C++ runtime library openly on GitHub. Pricing and availability have not been announced, but full specifications and a press release are posted on the official product page.