M5-182X-A1 AI Edge Box

Shimeta's M5-182X-A1 V1.0 AI computing development board / edge box combines the RK182X series AI coprocessor with the RK3588 host SoC, stacking the general-purpose compute of the RK3588 (native 6 TOPS) on top of the high-performance NPU of the RK182X (20 TOPS @INT8) for a combined 26 TOPS of total compute.

Widely applicable to: video AI terminal products, industrial automation terminals, and computing terminals — such as smart self-service kiosks, O2O smart devices, AI industrial-control hosts, and robotic equipment.

1. What is it?

RoleWhat it isKey capabilities
Host SoCRK3588 (8nm)4×A76 + 4×A55 @ 2.4GHz, Mali-G610 MP4 GPU, 6 TOPS on-board NPU
AI coprocessorRK182X (22nm)3× RISC-V control cores, 20 TOPS @INT8 NPU, 3D-stacked DRAM
Module interface260pin SO-DIMMThe whole system works once the RK182X module is seated in the SO-DIMM slot
InterconnectPCIe 3.0 x4 laneRK3588 ↔ RK182X communicate over PCIe

Model note: RK182X is the Rockchip AI coprocessor series name, covering two concrete models: RK1820 (2.5GB DRAM, 3B LLMs) and RK1828 (5GB DRAM, 7B LLMs).

2. Application scenarios

ScenarioTypical applications
Video AI terminalsSmart self-service kiosks, O2O smart devices
Industrial automationAI industrial-control hosts, robotic equipment
Computing terminal productAI Box, edge inference servers

3. What does it look like?

Product front view

Product left side

Product right side

3.1 Board front and back

Board isometric front view

Board front view

Board back view

4. Key advantages

  • High integration: HDMI / MIPI / 2× Ethernet / WiFi5 / BT5.0 integrated in one compact, clean-looking unit
  • Scalable compute: native 6 TOPS + 20 TOPS from the module = 26 TOPS total
  • Multi-model support: 0.5B ~ 7B LLM / VLM, one-click conversion with the RKNN3 SDK
  • Open ecosystem: supports mainstream Agent frameworks such as OpenClaw and Hermes Agent
  • Strong expandability: 2× USB3.0, 40pin IO (Raspberry Pi compatible), MIPI DSI/CAM, SO-DIMM compute expansion

5. System block diagram

┌─────────────────────────────────────────┐
│      M5-182X-A1 AI Edge Box (V1.0)      │
│   128(L)×95(W)×70(H)mm  DC 12V/5A       │
├─────────────────────────────────────────┤
│  Host SoC: RK3588 (8nm, 6 TOPS NPU)     │
│  CPU: 4×A76 + 4×A55 @ 2.4GHz            │
│  GPU: Mali-G610 MP4 (8K codec)          │
│  DDR: 4GB / eMMC: 32GB / OS: Linux      │
├─────────────────────────────────────────┤
│  PCIe 3.0 x4 lane (internal link)       │
├─────────────────────────────────────────┤
│  AI coprocessor: RK182X (22nm, 20 TOPS) │
│  3× RISC-V 64-bit control cores         │
│  (plugs in via SO-DIMM)                 │
│  8-core NPU; 2.5GB (RK1820) / 5GB       │
│  (RK1828)                               │
│  PCIe 2.1 / USB 3.0 interfaces          │
└─────────────────────────────────────────┘

6. Typical use cases

6.1 Deployed cases

  1. Smart QQ assistant: deploy OpenClaw for QQ chat + image recognition + scheduled reminders, one-click deployment script →
  2. Daily good-morning briefing: automatic weather + schedule lookup, pushed to QQ at a fixed time every day, see details →
  3. Real-time meeting monitoring: speech-to-text + keyword detection, see details →

6.2 More scenarios

  • Smart-home Agents
  • Industrial robots (from "executing commands" to "understanding intent")
  • Security / video analytics

7. Contact

  • Company: Shimeta
  • Website: www.shimeta.com.cn
  • Phone: 0755-66890088
  • Email: sales@shimeta.com.cn

8. Next steps