Algorithm Appliance · MES20-16A
Edge AI MES20-16A, 24 TOPS on Tap
8-core RISC-V CPU, 24 TOPS AI engine, 16-channel HD decoding, 4 independent networks — a single unit handles multi-stream video analytics and 3B/7B LLM inference, industrial-grade -20°C ~ +60°C full-scenario deployment.
- AI Performance
- 24 TOPS
- Video Decode
- 16 Ch
- Typical Power
- < 20W
Product Highlights
One Edge Compute Base, Covering All Edge AI Scenarios
Powerful Compute · 8-Core RISC-V
EA6530 8-core RISC-V @2.0GHz, performance comparable to ARM A73, with vector instruction set support.
24 TOPS AI Engine
Up to 24 TOPS (INT8) / 12 TFLOPS (FP16), supporting INT4/FP8 and Transformer LLM acceleration. 3B/7B models run locally.
Multi-Channel HD Codec
16-channel 1080P HD video decoding + 4-channel HD video encoding, ready for multi-stream video analytics.
Four Independent Networks
4 Gigabit Ethernet ports supporting 4 independent subnets, flexibly adapting to complex on-site topologies.
Hardware Encryption · Remote Guard
Hardware VPN encryption and dedicated remote management channel, ensuring long-term maintainability after deployment.
LCD Status Display
Front-panel LCD displays IP and system status in real time, making deployment and inspection more intuitive.
Full Coverage of Major Frameworks
SDK compatible with FFMPEG/OpenCV, supports Caffe / PyTorch / TensorFlow. Examples and documentation included.
Rich Interfaces
USB 3.0 / HDMI 2.0 / RS-485 / RS-232 / CAN / Audio / Custom I/O, integrating multiple field buses in one device.
Specifications
Specifications
| System | CPU | EA6530, 8-core RISC-V @2.0GHz (comparable to ARM A73), vector instruction set support |
| AI Performance | Supports INT4 / INT8 / INT16 / FP16 / BF16, up to 24 TOPS (INT8) / 12 TFLOPS (FP16) | |
| Memory | 16GB LPDDR4X (64-bit) | |
| Storage | 32GB eMMC, 1 × M.2 SSD slot, 1 × TF card slot | |
| Media Processing | Video Decode | H.264 / H.265, 1080P@480fps (4K@120fps) |
| Video Encode | H.264, 1080P@120fps | |
| Image Decode | 480 images/sec @1920×1080, max 32768×32768 | |
| Image Encode | 120 images/sec @1920×1080, max 8192×8192 | |
| Interfaces | Network | 4 × 10/100/1000Mbps adaptive Ethernet; optional 4G/5G, WiFi/BT; 6 × SMA antenna connectors |
| I/O Interfaces | 2 × USB 3.0; 1 × HDMI 2.0; 3.5mm headphone / microphone; 2 × RS485 / 1 × RS232 / 1 × CAN / custom IO | |
| Power | DC 12V, Phoenix 2-pin terminal (with locking screws) | |
| Software | OS Support | Linux / Ubuntu |
| AI Toolchain | NPU Runtime on-device inference + PC-side NPU toolchain (model import, quantization, simulation, precision optimization, performance analysis) | |
| Frameworks & Libraries | Caffe / PyTorch / TensorFlow; FFMPEG & OpenCV hardware-accelerated libraries; CV / multimedia system libraries | |
| Demo & Deployment | Built-in ~20 mainstream models; AIBOX Pipeline sample (stream→decode→infer→draw→display→forward); Docker container management | |
| Environment / Physical | Typical Power | < 20W |
| Operating Temperature | -20°C ~ +60°C | |
| Operating Humidity | 10% ~ 90%, non-condensing | |
| Protection Rating | IP40 | |
| Dimensions | 220 × 200 × 55 mm, weight 2.18 kg | |
| Accessories | Power Adapter | 60W industrial-grade; 100–240V AC 50/60Hz 2A, output 12V DC 5A |
| Mounting Brackets | 2 pcs, 200 × 15 × 10.5 mm | |
| Antennas (Optional) | 4 pcs |
Interface Overview
Rich I/O, Plug and Play on Site
Front and rear panels integrate network, video, serial, audio, and antenna interfaces, paired with LCD status display for more efficient deployment, debugging, and maintenance.
Industrial Wide Temperature
-20°C ~ +60°C, 10–90% RH non-condensing
Compact Form Factor
220 × 200 × 55 mm, 2.18 kg
Flexible Storage
32GB eMMC + M.2 SSD + TF expansion
Want a Custom MES20-16A Solution for Your Scenario?
Supports OEM/ODM customization, bulk delivery, and on-site algorithm porting. Our engineering team responds within 48 hours.