MIB-742-AT AI Inference Systems

Introduction

Advantech’s MIB-742-AT family is a wall-mountable Edge AI inference single-board computer platform based on NVIDIA Jetson Thor modules. Available with either the Jetson Thor T5000 or T4000, the systems combine Arm-based processing, NVIDIA Blackwell-architecture GPUs, and fifth-generation Tensor Cores in a compact 180 mm × 180 mm format.

The platform is notable for combining high-bandwidth networking, multi-camera connectivity, local storage expansion, and industrial-oriented interfaces in a single system. NVIDIA JetPack 7.0 support provides the software environment for Jetson Thor-based development and deployment.

What it does

MIB-742-AT systems are intended for edge-resident AI inference and sensor-processing workloads, particularly where multiple cameras or high-throughput network connections must be integrated locally.

The T5000 configuration uses a 14-core Arm Neoverse V3AE 64-bit CPU, a 2560-core NVIDIA Blackwell GPU, 96 fifth-generation Tensor Cores, and 128 GB LPDDR5X memory. The T4000 option provides a 12-core Arm Neoverse V3AE CPU, a 1536-core Blackwell GPU, 64 Tensor Cores, and 64 GB LPDDR5X memory. The T5000 configuration is specified for up to 2070 FP4 TFLOPS of AI performance.

For connectivity, the family includes a multi-rate RJ45 Ethernet port supporting speeds from 10 Mbps through 5 Gbps, plus one QSFP28 interface. Depending on the selected Jetson module, the platform supports up to four 25GbE connections with the T5000 or up to three with the T4000.

Four four-lane MIPI interfaces support camera connectivity, while an optional eight-channel GMSL2 camera arrangement is available through mini-FAKRA connectors. The system also provides HDMI output up to 3840 × 2160 at 60 Hz, USB 3.2 Gen 2 ports, CAN FD interfaces on T5000 configurations, I²C, audio, SATA, and multiple M.2 expansion options.

Applications

Potential uses that follow from the available interfaces and compute options include:

• Multi-camera computer-vision systems using MIPI or optional GMSL2 cameras
• Edge AI inference appliances with high-bandwidth Ethernet connectivity
• Robotics platforms requiring local AI processing, USB peripherals, and CAN FD connectivity
• Industrial AI systems integrating cameras, sensors, storage, and networked equipment
• Sensor-processing systems using QSFP28 and up to 25GbE network links
• Vision systems with local HDMI display output and NVMe storage expansion

Design considerations

System selection begins with the Jetson Thor module option. The T5000 and T4000 versions differ in CPU core count, GPU core count, Tensor Core count, memory capacity, and supported number of 25GbE connections. Designs requiring four CAN FD interfaces should also use the T5000 configuration.

Camera and network requirements should be defined early. The board provides four four-lane MIPI interfaces, while eight-channel GMSL2 support is optional and uses mini-FAKRA connectors. The QSFP28 interface and 25GbE capability may be relevant where high-throughput sensor or network data must be handled.

The system requires a 19 V to 36 V input supply and is specified for operation from -10 °C to +60 °C with 0.7 m/s airflow. It is wall-mountable, weighs 1 kg, and is rated for 3 Grms vibration from 5 Hz to 500 Hz for one hour per axis. Storage and peripheral integration can use two SATA connectors, an M.2 2280 M-key slot with PCIe x4 NVMe support, and additional E-key and B-key M.2 slots.

Availability and further information

The source/distributor and datasheet links on this page provide further information on the Advantech MIB-742-AT AI Inference Systems, including configuration details and supporting documentation.