Core-186JD4 Core Board

7.2 TOPS
Computing Power

Private Deployment
of Large Models

Supports Multiple Deep
Learning Frameworks

Supports 16-Channel
Video Decoding

AI Processor CV186AH

Core-186JD4 equipped with the SOPHON CV186AH hexa-core chip, featuring six ARM Cortex-A53 cores, it delivers 7.2 TOPS of
computing power and is widely used in AI inference and computer vision applications.

Equipped with the aBMC management system

Supports the private deployment of expansive parameter models based on the Transformer framework, including substantial
language models like Gemma-2B, ChatGLM3-6B, Qwen1.5-1.8B, and Llama2-7B.

Multi-Channel
Video Decoding

Supports full-process handling of 16 channels of 1080P
high-definition video (decoding + AI analysis), meeting the
needs of various AI application scenarios such as video
stream facial recognition, license plate recognition, and
smart city initiatives.

Diverse and Functional
Algorithms

It supports multiple algorithms porting such as "person/
vehicle/object" recognition, video structuring, and
trajectory behavior, with high security and reliability. It
can be flexibly applied to various product development

Easy and Convenient
to Develop

Provides a one-stop deep learning SDK with a complete
toolkit of software utilities, encompassing driver
environments, compilers, and deployment tools for inference.

Comprehensive Expansion Interfaces

Specifications

Core-186JD4
Basic Specifications
SOC

SOPHON CV186AH

CPU

Hexa-core 64-bit ARM Cortex-A53 @ 1.6GHz

TPU

Built-in SOPHGO neural network acceleration engine TPU, computing power up to 7.2T@INT8, 12T@INT4, 1.5T@FP16/BF16

ISP

Time-sharing multiplexing for up to 6 sensor input videos, with maximum widths of 4608 (non-tile mode) and 8192 (tile mode) Supports Sensor self-band dynamic and 2-frame wide dynamic range, and the maximum performance supports: 12M@30 HDR or 8K@15 SDR or 16M@30 SDR Support RGB-IR, AI ISP interface, 3A (AE/AWB/AF, 3A control user adjustable) Support fixed mode noise removal, dead pixel correction, shadow correction, lens distortion correction, purple edge correction, Bayer noise reduction, 3D denoising, image edge enhancement, dehazing, dynamic contrast enhancement, image video Mirror, Flip and other functions

VPU

Video decoding: H.265/H.264 decoding (maximum performance: 1920×1080@480fps or 8192×4320@30fps) Video encoding: H.265/H.264 encoding (maximum performance: 1920×1080@300fps or 8192×4320@15fps) Image codec: JPEG/MJPEG Baseline codec (JPEG codec: 1080P@480fps, maximum resolution 32768×32768)

RAM

4GB LPDDR4(4GB/8GB optional)

Storage

32GB eMMC(32GB/64GB/128GB/256GB optional)

OS

Linux OS(Ubuntu)

Software Support

・ The private deployment of ultra-large-scale parameter models under the Transformer architecture, including large   language models such as Gemma-2B, LlaMa2-7B, ChatGLM3-6B, Qwen1.5-1.8B ・ Traditional network architectures such as CNN, RNN, and LSTM; a variety of deep learning frameworks, including   TensorFlow, Pytorch, PaddlePaddle, Caffe and ONNX, as well as custom operator development ・ Docker container management technology

Power

5V (voltage tolerance ± 5%)

Power consumption

Normal: 5.5W (5V/1100mA),Max: 20W(5V/4000mA)

Interface

Goldfinger (260 PIN, SODIMM, 0.5mm pitch)

Size

69.6mm × 45.0mm

Weight

≈20g

Environment

Operating Temperature: -20℃~60℃, Storage Temperature: -20℃~70℃, Storage Humidity: 10%~90%RH(non-condensing)

Interface Specifications
Internet

2 × Gigabit Ethernet (MDI interface is provided, the core board has an onboard Ethernet PHY chip), WiFi/Bluetooth can be expanded through SDIO, and 5G/4G wireless network can be expanded through USB3.0/USB2.0

Video input

It supports up to 6 video inputs, including 1 channel 8L, two channels 8L+4L, three channels 4L, six channels 2L, and other combinations, and maximum resolution is 8192×3840 Supports MIPI-CSI/subLVDS Support Bayer RGB (RAW8/RAW10/RAW12/RAW16) parallel interface, maximum resolution of 1920×1080 Support BT (BT.601/ BT.656/ BT.1120), maximum resolution of 1920×1080 Support high-definition CMOS sensors such as SONY, OnSemi, OmniVision, etc Supports high dynamic range (up to 2 frame HDR) inputs, mainstream CMOS-level thermal imaging sensors

Video output

1 × HDMI 2.0: Maximum output capacity 4 (4096×2160)@60fps 1 × MIPI DSI: 2.5G bps/lane, up to 4 lanes, maximum output capacity 1440P(2560 ×1440)@60fps 1 × LVDS: Bit Rate: 1.6G bps/lane, maximum 1 link, maximum output capacity 720P(1280×720)@60fps

Audio

Integrated Audio CODEC supports 16-bit source/voice input and output Support two dual-channel microphone inputs, support two-channel two-channel microphone output (external amplifier is required to push the speaker) It supports connecting to an external Audio CODEC via I2S/PCM/TDM interface, I2S mode supports up to 4-channel stereo input, and the built-in Audio PLL supports MCLK output

SATA/PCIe

2 × PCIe 3.0/SATA Gen3 high-speed interface: 1) Configurable as 2-Lane PCIe3.0 RC/EP + 2-Lane PCIe3.0 RC/EP 2) Configurable as 2-Lane PCIe3.0 RC/EP + 2 x SATA Gen3

USB

2 × USB3.1/USB2.0 Host/Device

SDIO

2 × SD3.0/SDIO3.0 (support 1-bit/4-bit mode)

I2C

11 × I2C (1 no die domain)

UART

9 × UART (1 no die domain)

CAN

2 × CAN 2.0 (Support CAN2.0 A/B ISO11898 and CAN FD 1.1)

PWM

20 × PWM (support up to 50MHz (100MHz/2) output, down to about 0.093Hz (100MHz/(2^30-1)))

ADC

5 × single-ended ADC (2 no die domain)

Other interfaces

4 × SPI, 3 × I2S, 209 × GPIO (17 no die domain), IR