Description
The IOTRouter LQ50 AI Accelerator Card accelerates local inference for large language models (LLMs) and other AI models on compatible PCs and embedded systems.
It delivers up to 160 TOPS of AI performance and 100 TFLOPS at bFP16. Its 24 GB LPDDR5/LPDDR5X memory provides 153.6 GB/s bandwidth for AI inference. Typical power consumption is 13 W and can drop to 6 W in simultaneous interpretation applications.
The compact M.2 Key M card connects through PCIe Gen4 ×4, with peak bandwidth of up to 8 GB/s. Two UART interfaces and one I²C controller provide connectivity for peripherals and control devices, making it suitable for industrial automation and other edge AI deployments.
Highlights At A Glance
160 TOPS Computing
Delivers up to 160 TOPS of AI acceleration for demanding edge workloads and local inference
M.2 Form Factor
The compact M.2 design supports integration into compatible AI PCs and embedded systems
Local LLM Support
Built for local LLM inference applications, helping move AI processing closer to the device
HOUMO AI Platform
Features the HOUMO AI accelerator solution in the MOMAGIC 50 card design for edge AI deployment
Product Specifications
See the table below for details.
| FEATURE | SPECIFICATION |
|---|---|
| Physical dimensions | 22 mm × 80 mm × 3.3 mm |
| AI processor | 1 × HOUMO M50 chip 2 integrated IPU cores |
| AI computing performance | Up to 160 TOPS of flexible compute Up to 100 TFLOPS @ bFP16 |
| Memory | 24 GB LPDDR5 / LPDDR5X Memory bandwidth: 153.6 GB/s |
| Storage | 1 × 32 MB NOR Flash |
| Form factor / interface | M.2 Key M |
| PCIe interface | 1 × PCIe Gen4 interface; backward compatible with PCIe Gen3 / Gen2 / Gen1 Up to x4 lanes; compatible with x2 / x1 Peak bandwidth: up to 8 GB/s per interface |
| UART | 2 × UART interfaces Supports standard baud rates up to 921,600 bps |
| I²C | 1 × I²C controller Supported speeds: 100 kHz and 400 kHz |
| Weight | 9 g (±10 g) |
| Operating temperature | −20 °C to +60 °C |
| Storage temperature | −20 °C to +65 °C |
| Operating humidity | 10% RH to 90% RH (non-condensing) |
| Storage humidity | 5% RH to 95% RH (non-condensing) |
| Supply voltage | 3.3 V ± 5% |
| Peak current | 10 A |
| Typical power consumption | 13 W |
Need help integrating the LQ50? Talk to our team about system compatibility, AI applications, and pricing.
Typical Use Cases of the LQ50 AI Accelerator Card
1. Industrial Automation
Integrate the LQ50 into a compatible industrial PC to accelerate AI models for visual inspection, defect detection, and production monitoring. Local inference helps manufacturers process data closer to the production line.
2. Security Analytics
The LQ50 supports AI inference for video analytics applications such as object detection and event monitoring. It can be integrated into compatible edge devices to process video data locally.
3. Smart Meeting Systems
With the LQ50, compatible meeting terminals can accelerate AI models for speech transcription, translation, and meeting summaries. Local processing supports meeting applications where data needs to stay on the device.
