Specifications
SoC
Rockchip RK3576
CPU
4x Cortex-A72 at up to 2.2 GHz plus 4x Cortex-A53 at up to 2.0 GHz
GPU
Arm Mali-G52 MC3
Integrated NPU
Up to 6 TOPS INT8
Optional AI Accelerator
Rockchip RK1820, up to 20 TOPS INT8, 2.5 GB on-chip DRAM
Accelerator Software
Rockchip RKNN3 stack; optimised for Qwen, DeepSeek and YOLO models
Memory Options
4 GB LPDDR5 + 32 GB eMMC, or 8 GB LPDDR5 + 64 GB eMMC
Module Connector
200-pin board-to-board to carrier / IO board
Storage Expansion
1x M.2 M-Key 2280 over PCIe 2.1 x1, plus microSD slot
Networking
1x Gigabit Ethernet, Wi-Fi 6, Bluetooth 5.4
PoE
PoE PD support (requires an additional PoE module)
Display
HDMI, DisplayPort over USB Type-C, 1x 4-lane MIPI-DSI
Camera
1x 4-lane and 1x 2-lane 22-pin MIPI-CSI
USB
1x USB 3.0 Type-A, 2x USB 2.0 Type-A, 1x USB Type-C (OTG + DisplayPort)
Expansion Header
40-pin header exposing UART, CAN, I2C, SPI, PWM and GPIO
Video Decode
H.265, VP9, AV1, AVS2 up to 8K30; H.264 up to 4K60
Video Encode
H.265 and H.264 hardware encoding up to 4K60
Benchmark
YOLO11 at 77.9 fps with 640 x 640 input on the integrated NPU
Power Input
9V to 19V DC
Operating Temperature
0C to 65C
Dev Kit Dimensions
164 x 95 x 31.4 mm
Operating System
Armbian preinstalled; Debian, Ubuntu, Android 14+ and balenaOS supported
Dev Kit Contents
RK3576 module, IO board, 12V/3A adapter, 32 GB microSD, active cooling, wireless antennas
Pricing
Module $130 (4GB/32GB) or $170 (8GB/64GB); IO board $18; standard Dev Kit $219; RK1820 Dev Kit $379
Availability
Late September through October 2026
Overview
Seeed Studio reComputer RK3576 is a modular edge AI development platform built around the Rockchip RK3576 SoC, offered as a bare compute module, as a module plus IO board, or as a complete Dev Kit with the RK1820 AI accelerator fitted.
The RK3576 supplies heterogeneous octa-core CPU performance, a Mali-G52 MC3 GPU and a 6 TOPS INT8 NPU that already handles practical vision workloads - Seeed quotes YOLO11 at 77.9 fps with 640 x 640 input. Adding the RK1820 M.2 accelerator raises local AI throughput to 20 TOPS INT8 backed by 2.5 GB of on-chip DRAM, which is what lets the platform run distilled LLMs and VLMs such as DeepSeek-R1-Distill-Qwen 7B or Qwen2.5-VL 3B at the edge.
One tradeoff is worth noting before design-in: the RK1820 occupies the board's only M.2 M-Key slot, so NVMe storage cannot be fitted in the same slot when the accelerator is installed. Teams needing both should plan on microSD or eMMC boot storage. The platform is fully documented through the reComputer AI Lab toolchain, covering model conversion, deployment and evaluation with the RKNN3 software stack.
Key Benefits
Modular path from a $130 compute module up to a $379 accelerator-equipped Dev Kit; 6 TOPS integrated NPU plus an upgrade path to 20 TOPS INT8 via the RK1820; 4-8 GB LPDDR5 and up to 64 GB eMMC; Wi-Fi 6, Bluetooth 5.4 and Gigabit Ethernet; 40-pin expansion header with CAN, UART, I2C, SPI and PWM; RKNN3 software stack with pre-optimised Qwen, DeepSeek and YOLO models.
Applications
Local LLM and VLM prototyping, computer vision and smart camera development, robotics and drone perception, industrial inspection, education and research platforms, AI gateway development.
Request a Quote — SEEED RECOMPUTER RK3576 — EDGE AI DEV KIT WITH RK1820 20-TOPS ACCELERATOR
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →