SOPHGO

VISION Computing Architecture

CV184x

Name: SOPHGO CV184x
Brand: SOPHGO
SKU: CV184x

1.5T Dual-Core Vision SoC with ISP V4.0 & Smart Encoding

Product Brief

The CV184x series is a high-performance visual processor designed for consumer, carrier, and professional security markets. It features a dual-core architecture (ARM Cortex-A53 @ 1.1GHz + RISC-V C906 @ 800MHz) and a self-developed TPU delivering 1.5 TOPS (INT8) with BF16 mixed precision support.

CV184x introduces the new ISP V4.0 engine, offering comprehensive image quality enhancements including starlight-grade 3DNR, digital wide dynamic range (DRC), dehaze, and purple fringe removal (PFR). The improved MCTF algorithm significantly reduces motion smear and spatial noise.

For efficient storage and transmission, the chip supports a self-developed Smart Encoding engine with H.265/H.264/MJPEG support. The unique Intra4 mode improves compression rates and texture retention in complex motion scenes compared to standard encoding.

CV184x provides SDK support for edge-side large-model workflows and is compatible with the OpenCLIP multimodal model, enabling generic object detection, image-text retrieval, and zero-shot inference. It comes with a rich library of turnkey algorithms including face detection, person/vehicle tracking, and helmet detection.

Designed for seamless migration, the CV184x series is pin-to-pin compatible with the CV181 series, sharing the same SDK, ISP tuning tools, and deep learning environment.

ISP V4.0 Engine

Features MCTF-based 3DNR, Digital Wide Dynamic Range (DRC), and Purple Fringe Removal (PFR) for superior low-light and backlit imaging.

Smart Encoding + Intra4

Smart Encoding with Intra4 retains complex textures while reducing bitrate and storage cost.

OpenCLIP + Edge LLM SDK

Supports image-text matching, zero-shot inference, and SDK workflows for edge-side large models.

High Performance TPU

1.5 TOPS INT8 performance with BF16 support. Outperforms predecessors in ResNet50 and YOLOv5s benchmarks.

Rich Algorithm Library

Includes Face/Person/Vehicle detection, Helmet check, Gesture recognition, Baby cry detection, and Semantic segmentation.

Seamless Migration

Hardware pin-compatible with CV181. Shared SDK and toolchain ensure low-cost upgrades.

System Architecture

Key Innovation 01

Dual-Core: ARM A53 (1.1GHz) + RISC-V C906 (800MHz)

Key Innovation 02

TPU: 1.5 TOPS @ INT8 / BF16 Mixed Precision

Key Innovation 03

ISP V4.0: 3DNR, DRC, PFR, Dehaze, Distortion Correction

Key Innovation 04

Smart Encoding + Intra4: Higher compression, better texture retention

Key Innovation 05

OpenCLIP + Edge LLM SDK: Zero-shot, Image-Text Retrieval

Key Innovation 06

Pin-to-Pin Compatible with CV181 Series

High Performance Compute

AI Performance

The self-developed TPU delivers significant performance gains over previous generations, supporting complex models like ResNet, YOLO, and MobileNet.

Benchmark based on CV1842C-P (1.5 TOPS).

ResNet50 v2: +40% vs CV181.

YOLO v5s: +81% vs CV181.

MobileNet v2: +51% vs CV181.

Supports BF16 mixed precision for higher accuracy.

Benchmarks & Metrics

ResNet50 v238.5 FPS

YOLO v5s28.3 FPS

MobileNet v2261.9 FPS

OpenCLIP4.07 FPS

Advanced ISP Pipeline

Imaging Pipeline

A completely overhauled imaging engine focusing on noise reduction, dynamic range, and color accuracy.

MODULE-01

3DNR (MCTF)

Reduces motion smear and fine temporal noise. Superior low-light performance (0.1lux).

MODULE-02

DRC (Dynamic Range Compression)

Brightens shadows without overexposing highlights. Better than competitors in backlit scenes.

MODULE-03

PFR (Purple Fringe Removal)

Completely removes purple fringing at object edges without color loss.

MODULE-04

Edge Enhancement

Frequency-based sharpening distinguishes between fine textures and strong edges.

MODULE-05

Color Noise Suppression

Eliminates low-frequency color noise and edge spill.

AI Model Library

Turnkey solutions for consumer and security scenarios.

Face

Human

Vehicle

Behavior

Face: Detection, Recognition, Capture, Keypoints (5 points).

Human/Body: Person detection, Head detection, Skeleton keypoints.

Vehicle: Motor/Non-motor vehicle detection, License plate recognition.

Behavior: Gesture recognition, Fall detection, Passenger flow counting.

Security: Helmet detection, Intrusion detection (Tripwire).

Audio: Baby cry detection, Voice command recognition (e.g., 'Call', 'Screen On').

Other: Cat/Dog detection, Semantic Segmentation.

Variant Comparison

Feature	CV1841C-P	CV1842C-P	CV1842H-P/CV1842HP	CV1843H-P
Processor	CA53 1.1GHz + C906 800MHz	CA53 1.1GHz + C906 800MHz	CA53 1.1GHz + C906 800MHz	CA53 1.1GHz + C906 800MHz
DDR3 SiP	1Gbit	2Gbit	2Gbit	4Gbit
TPU (INT8)	~1.5 TOPS	~1.5 TOPS	~1.5 TOPS	~1.5 TOPS
Package	QFN-88 9x9	BGA 10x10	BGA 10x10	BGA 10x10
Video Input	MIPI 2L+1L or 4L	MIPI 2L+2L+DVP or 4L	MIPI 2L+2L+DVP or 4L	MIPI 2L+2L+DVP or 4L
Video Output	MIPI DSI(2L)/BT/RGB	MIPI DSI(4L)/BT/LVDS/RGB	MIPI DSI(4L)/BT/LVDS/RGB	MIPI DSI(4L)/BT/LVDS/RGB
Ethernet	10/100 MAC PHY	10/100 RMII + MAC PHY	10/100 RMII + MAC PHY	10/100 RMII + MAC PHY

Commercial Deployment Scenarios

Smart Cameras

Access Control

Robotics

4K IPC

Drones

Dash Cam

Datasheet

Technical Reference

Processor
CPU	Dual-Core ARM Cortex-A53 + RISC-V C906
Frequency	Up to 1.1GHz (A53) / 800MHz (C906)
AI
TPU	1.5 TOPS @ INT8 / BF16
Memory
Integrated	DDR3 512Mb / 1Gb / 2Gb / 4Gb
Video
ISP	ISP V4.0 (4K, 3DNR, DRC, PFR)
Encoding	H.265/H.264/MJPEG Smart Encoding + Intra4
Resolution	Max 8M@25fps + 2x 720P@25fps
Interface
Camera	2x MIPI CSI (2L/4L/DVP)
Display	1x MIPI DSI (2L/4L), LVDS, RGB, BT.656/1120
Peripherals
Connectivity	USB 2.0, SD 3.0, Ethernet (MAC PHY/RMII)
Low Speed	UART, PWM, GPIO, I2C, SPI, RTC
System
Features	Fast Boot, Secure Boot, Low Power

Flight Controller

All-In-One FC

FPV/UAV

FPV/UAV Solutions

Consumer

Consumer Solutions