Specifications
Model
Sipeed MaixCAM2
Product Type
Modular AI camera for makers, researchers and embedded vision development
SoC
Axera AX630
CPU
Dual-core Arm Cortex-A53 @ 1.2 GHz running Ubuntu Linux plus a 32-bit RISC-V E907 core running RT-Thread
NPU
12.8 TOPS @ INT4, 3.2 TOPS @ INT8; supports convolution and Transformer models including YOLO, LLM and VLM
Measured Inference
YOLO11n at 640 x 640 reaches up to 113 FPS
VPU
H.264 / H.265 / MJPEG — 4K @ 30 fps encoding, 1080p @ 60 fps decoding
System Memory
1 GB or 4 GB LPDDR4
Storage
32 GB eMMC flash plus microSD card slot
Built-in Display
2.4-inch HD 640 x 480 IPS capacitive touchscreen
Display Output
4-lane MIPI DSI, 31-pin interface, up to 1080p60; 6-pin capacitive touchscreen connector
Image Sensor
1/1.8-inch sensor, 8 MP, 4Kp30 capture — up to 80 % larger than a Raspberry Pi Camera sensor
Lens Mount
M12, convertible to CS-mount for industrial camera lenses
Mounting
Extra screw fixing holes accept a 1.25-inch telescope adaptor or a 23.2 mm microscope adaptor
Camera Input
4-lane MIPI CSI, 22-pin interface
Audio Out
Onboard PA amplifier with 1 W speaker
Audio In
2x analog silicon microphones for direct audio capture
Ethernet
6-pin FPC Ethernet interface with external FPC-to-RJ45 module
USB
1x USB 2.0 Type-C (Device and Host modes), supports USB cameras
Sensors
Onboard 6-axis IMU (3-axis accelerometer plus 3-axis gyroscope)
Expansion
2x 2.54 mm PMOD interfaces (20 IOs plus Vsys / 3.3 V / GND) and a 6-pin 1.25 mm connector
Expansion Protocols
I2C, SPI, UART, ADC, PWM, watchdog and other common peripherals
Indication
Power, user and illumination LEDs
Overview
The Sipeed MaixCAM2 is a modular AI camera built around the Axera AX630 SoC. Its NPU is rated at 3.2 TOPS at INT8 and 12.8 TOPS at INT4, and Sipeed claims 10 to 20 times the performance of OpenMV-N6 and Raspberry Pi 5 class hardware on the same workloads — measured output is YOLO11n at 640 x 640 running at up to 113 FPS. Transformer models are supported alongside convolutional ones, which is what allows local LLM and VLM work on the device.
The camera ships with 1 GB or 4 GB of LPDDR4, 32 GB of eMMC plus a microSD slot, and an integrated 2.4-inch 640 x 480 IPS capacitive touchscreen for standalone operation. The imager is a 1/1.8-inch 8 MP sensor capturing 4K at 30 fps — physically larger than the sensor in typical maker cameras — behind an M12 lens mount that converts to CS-mount for industrial lenses. Screw fixing holes accept a 1.25-inch telescope adaptor or a 23.2 mm microscope adaptor.
I/O covers a 4-lane MIPI CSI camera input, 4-lane MIPI DSI output up to 1080p60, FPC Ethernet, USB 2.0 Type-C in both device and host modes, an onboard 6-axis IMU, two analog microphones and a 1 W speaker. Two 2.54 mm PMOD interfaces expose 20 IOs plus power for I2C, SPI, UART, ADC and PWM peripherals — the expansion route that makes it usable as a vision front-end on a larger robot rather than a closed camera.
Key Benefits
12.8 TOPS INT4 NPU is far above typical maker-camera silicon and runs YOLO11n at 640 x 640 at up to 113 FPS. Transformer support enables local LLM and VLM inference on the camera itself. 1/1.8-inch 8 MP sensor with M12-to-CS mount conversion allows genuine industrial lenses and telescope or microscope adaptors. Built-in touchscreen, IMU, microphones, speaker and FPC Ethernet make it a complete standalone device, and two PMOD expansion headers let it act as a vision module inside a larger system.
Applications
Embedded vision prototyping and research, robotics perception front-ends, local YOLO and transformer inference at the edge, smart-camera product development, machine-vision experiments needing CS-mount optics, educational and maker AI vision projects.
Request a Quote — SIPEED MAIXCAM2 — MODULAR 4K AI CAMERA ON AXERA AX630 WITH 12.8 TOPS INT4
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →