Component guide · Advanced · 3 min read
NPU AI Boards

D-Robotics RDK X3 (5 TOPS), RDK X5 and RDK S100 (80 TOPS) with Waveshare RDK X3 Audio Board, Avnet RZBoard V2L DRP-AI, SiMa.ai MLSoC DevKit 2.0, Luckfox Omni3576 (RK3576), ODROID-M2 (RK3588S2) and Arduino Portenta X8
By Geekobot Learning Team · Updated 29 Oct 2026
On this page
These boards run AI models on a built-in NPU (neural processing unit) instead of a GPU, keeping power and price low. D-Robotics RDK boards are aimed at ROS 2 robots, from the 5 TOPS RDK X3 to the 80 TOPS S100. RZBoard V2L uses Renesas DRP-AI, SiMa.ai MLSoC targets high-efficiency vision, and the Rockchip RK3576 and RK3588 boards give 6 TOPS with good Linux support. Portenta X8 is an industrial Linux module.
Key specs at a glance
- RDK X3
- 4x A53, 5 TOPS BPU, 4GB
- RDK X5
- 8x A55, 10 TOPS, 4GB
- RDK S100
- 80 TOPS BPU, robot controller MCU
- RK3576 / RK3588S2
- 6 TOPS NPU, LPDDR4/5
- RZBoard V2L
- Cortex-A55 with DRP-AI accelerator
- Portenta X8
- i.MX 8M Mini Linux + STM32H7, no NPU
Models in this guide
These are alternatives. Pick the one that fits your project; you only need one.
Option 1D-Robotics RDK X3 Dev Board – Cortex-A53, 5 TOPS AI, 4GB RAM₹14,699
Option 2D-Robotics RDK X5 Development Board₹21,889
Option 3D-Robotics RDK S100 Robot Dev Kit w/ 80 TOPS BPU₹79,799
Option 4Waveshare Add-On Board for RDK X3₹2,000
Option 5Avnet RZBoard V2L Vision AI SBC Cortex-A55 DRP-AI₹18,159
Option 6SiMa.ai MLSoC Development Kit₹2,08,949
What the D-Robotics RDK X3, X5 and S100, Avnet RZBoard V2L, SiMa.ai MLSoC, Luckfox Omni3576, ODROID-M2 and Portenta X8 is
RDK and other NPU AI boards, such as the D-Robotics RDK X5, Rockchip RK3588 boards and Sophgo boards, are Linux computers with built-in neural processing units for edge AI and robotics.
Every NPU needs models converted with its own toolchain: Horizon OpenExplorer for RDK, RKNN-Toolkit2 for Rockchip, DRP-AI TVM for RZBoard and Palette for SiMa.ai. Check that your model's layers are supported before you commit to a board.
RDK boards ship with TogetheROS (ROS 2) and many ready robot examples, which makes them a quick start for AI robots. For general Linux use with some AI, an RK3576 or RK3588 board is the most flexible.
D-Robotics RDK X3, X5 and S100, Avnet RZBoard V2L, SiMa.ai MLSoC, Luckfox Omni3576, ODROID-M2 and Portenta X8 pinout
The pins on the D-Robotics RDK X3, X5 and S100, Avnet RZBoard V2L, SiMa.ai MLSoC, Luckfox Omni3576, ODROID-M2 and Portenta X8 and what each one does.
| Pin | Function |
|---|---|
| 40-pin header | GPIO, I2C, SPI, UART on most |
| MIPI CSI | Cameras |
| Gigabit Ethernet | Network |
| USB 3 | Peripherals |
| M.2 | NVMe or Wi-Fi on some |
- 40-pin header
- GPIO, I2C, SPI, UART on most
- MIPI CSI
- Cameras
- Gigabit Ethernet
- Network
- USB 3
- Peripherals
- M.2
- NVMe or Wi-Fi on some
Wiring the D-Robotics RDK X3, X5 and S100, Avnet RZBoard V2L, SiMa.ai MLSoC, Luckfox Omni3576, ODROID-M2 and Portenta X8
Use the maker's recommended camera modules; MIPI cameras are not interchangeable across boards.
| D-Robotics RDK X3, X5 and S100, Avnet RZBoard V2L, SiMa.ai MLSoC, Luckfox Omni3576, ODROID-M2 and Portenta X8 pin | Connection | Note |
|---|---|---|
| USB-C PD | 5V/3A or more | Check each board |
| MIPI camera | CSI connector | Board-specific cameras |
| Audio board | RDK X3 header | Waveshare add-on |
- USB-C PD
- Connection:
- 5V/3A or more
- Note:
- Check each board
- MIPI camera
- Connection:
- CSI connector
- Note:
- Board-specific cameras
- Audio board
- Connection:
- RDK X3 header
- Note:
- Waveshare add-on
Example code
Use target_platform='rk3588' for ODROID-M2; RDK and RZBoard use their own toolchains.
pip install rknn-toolkit2
python3 - <<'EOF'
from rknn.api import RKNN
r = RKNN()
r.config(mean_values=[[0,0,0]], std_values=[[255,255,255]], target_platform='rk3576')
r.load_onnx(model='yolov8n.onnx')
r.build(do_quantization=True, dataset='calib.txt') # list of sample images
r.export_rknn('yolov8n.rknn')
EOF
# Copy yolov8n.rknn to the board and run it with rknn-toolkit-lite2Setup and tuning
flash the vendor's Ubuntu or Debian image with their NPU drivers.
convert models to the vendor format (RKNN for Rockchip, BPU for RDK) with their toolkit.
RDK boards ship with ROS 2 packages for robots.
use the specified USB-C PD or 12 V supply.
fit a heatsink and fan for sustained AI loads.
What you need
Shop this list10 of 10 found in our store · ₹4,06,863
D-Robotics RDK X3 Dev Board – Cortex-A53, 5 TOPS AI, 4GB RAM: the part this guide coversD-Robotics RDK X3 Dev Board – Cortex-A53, 5 TOPS AI, 4GB RAM₹14,699
D-Robotics RDK X5 Development Board: another version this guide coversD-Robotics RDK X5 Development Board₹21,889
D-Robotics RDK S100 Robot Dev Kit w/ 80 TOPS BPU: another version this guide coversD-Robotics RDK S100 Robot Dev Kit w/ 80 TOPS BPU₹79,799
Waveshare Add-On Board for RDK X3: another version this guide coversWaveshare Add-On Board for RDK X3₹2,000
Avnet RZBoard V2L Vision AI SBC Cortex-A55 DRP-AI: another version this guide coversAvnet RZBoard V2L Vision AI SBC Cortex-A55 DRP-AI₹18,159
SiMa.ai MLSoC Development Kit: another version this guide coversSiMa.ai MLSoC Development Kit₹2,08,949
Luckfox Omni3576 Edge Computing Board (RK3576): another version this guide coversLuckfox Omni3576 Edge Computing Board (RK3576)₹22,850
ODROID-M2 SBC – RK3588S2, 8GB LPDDR5, 64GB eMMC: another version this guide coversODROID-M2 SBC – RK3588S2, 8GB LPDDR5, 64GB eMMC₹20,449
Arduino Portenta X8 SOM ABX00049: another version this guide coversArduino Portenta X8 SOM ABX00049₹17,984
USB to TTL serial converter: to talk to the module from a PCUSB to TTL Serial Converter (CH340G / CP2102 / FT232RL / PL2303)₹85Closest match, check the specs
Alternatives and when to choose them
- Edge AI boards: all AI boards.
- Alternative SBCs: Orange Pi and Radxa RK3588 boards.
- Jetson modules: GPU-based.
Common mistakes to avoid
- Running unconverted models.
- Using images without NPU drivers.
- Weak power supply.
- No cooling.
Frequently asked questions
What is an NPU?
A neural processing unit that runs AI models faster and with less power than the CPU.
Can I run YOLO on an RK3588 board?
Yes. Convert the model to RKNN format with Rockchip's toolkit, then run it on the NPU.
Why is my model slow on the NPU?
It may be running on the CPU; check conversion and runtime logs.
Does the RDK X5 support ROS 2?
Yes. The RDK X5 ships with ROS 2 and ready robot packages.
Which is easier: Jetson or NPU boards?
Jetson has a bigger ecosystem; NPU boards are cheaper but need model conversion.
Parts for this guide
Keep learning
Sources
Written by the Geekobot team from manufacturer datasheets. Always check the datasheet for the exact part and variant you buy.
