Specifications
Module
Jetson T5000 / T4000 (Thor)
AI (T5000)
2,070 FP4 TFLOPS
GPU
2560 CUDA cores, 96 Tensor Cores (5th Gen)
CPU
14-core Arm Neoverse V3AE 64-bit
Memory
128GB LPDDR5X (T5000) / 64GB (T4000)
Networking
QSFP28 (4× 25GbE / 3× 25GbE), 1× 5GbE RJ-45
I/O
CANBus, HDMI, USB 3.2 Gen 2 × 4
Storage
M.2 NVMe × 2, 1TB drive included
Expansion
M.2 B-Key 4G/5G, M.2 E-Key WiFi/BT
Edge AI
Real-time LLM/VLM inference, multi-model serving
Power
External adapter (XMIC-FSP230)
OS
JetPack 7.0+
Overview
The Advantech MIC-743-AT is a hardened edge AI inference system built on the NVIDIA Jetson Thor platform (T5000/T4000 modules). With up to 2,070 FP4 TFLOPS and 128GB LPDDR5X, it is designed for real-time Large Language Model (LLM) and Vision Language Model (VLM) inference directly at the edge — eliminating cloud dependency for latency-sensitive industrial applications. 100GbE-class QSFP28 connectivity, CANBus fieldbus, and Advantech's industrial-grade reliability make the MIC-743-AT the go-to platform for next-generation AI agents in factory, logistics, and transportation environments.
Key Benefits
Edge LLM Serving: Deploy Llama 3.1 8B at 150.8 tokens/sec, Llama 3.3 70B at 12.64 tokens/sec — 1.34× to 1.71× faster than Jetson AGX Orin. Dual Module Support: Compatible with both T5000 (2,070 TFLOPS, 128GB) and T4000 (1,200 TFLOPS, 64GB) for flexible price-performance scaling. Industrial Hardened: Designed for harsh edge environments with Advantech's validated thermal and mechanical design, enabling continuous 24/7 AI inference in factories, warehouses, and roadside cabinets.
Request a Quote — Advantech MIC-743-AT
QS Compute — NVIDIA Goldendisk Partner. Global B2B supply, 15-day sample lead time.
Get Your Quote →