Rockchip has built a complete product matrix covering 0.5 T to 6 T computing power in the visual chip field, from entry-level smart cameras to high-end industrial vision systems, almost covering all mainstream vision-computing scenarios. The most representative is the RV1126 series. RV1126B and RV1126B-P are widely used in: intelligent security (IPC, smart doorbells, dashcams); industrial vision (industrial cameras, inspection equipment); intelligent automotive and smart home; smart city, edge AI and other diversified AIoT scenarios. The RV1126B version, with its 3 TOPS strong computing power, customized AI-ISP architecture, dynamic stitching, anti-shake, super-encoding and hardware-level security features, provides these AIoT terminals with an efficient solution that goes from “seeing clearly” to “understanding what is seen”.
RV1126B-P – Simplified Package and Interface Version
RV1126B-P is the “simplified package and interface version” of RV1126B. The core computing power (CPU/GPU/VPU/NPU) is unchanged; cost and package complexity are reduced mainly by decreasing pin count, trimming high-speed interfaces (USB 3.0) and auxiliary interfaces (multiple CAM_CLK/SARADC channels). Removing USB 3.0 targets solutions with lower storage and bandwidth requirements; reducing CAM_CLK and ADC channels focuses on lightweight AI vision devices such as dashcams and DMS. RV1126B-P is the “compatible upgrade version” within the RV1126 series. It achieves Pin-to-Pin compatibility with the previous RV1126, maintaining the same core performance in computing power, ISP, and AI architecture as RV1126B. The biggest highlight is that it can replace the old chip directly without redesigning hardware, helping manufacturers upgrade products at minimal cost and significantly shortening development cycles. This means that if you already have a mature product based on RV1126, a lightweight software adaptation can give you the same level of computing power and image quality experience as RV1126B.
Processor Architecture
RV1126B uses a quad-core Cortex-A53 (1.5 GHz) architecture, integrates Rockchip’s self-developed 3 TOPS NPU, supports W4A16/W8A16 mixed-precision quantization and Transformer-optimized acceleration, and can efficiently run large models (≤ 2 B parameters) and multimodal models on-device. In image processing, it is equipped with an independent AI-ISP engine, combined with AI Remosaic technology to achieve “day-night dual-mode adaptation”, still producing clear images in extremely dark environments of 0.01 Lux, solving night-time monitoring noise issues. Coupled with 6-DOF digital anti-shake and dual/quad-camera dynamic stitching, the picture remains stable and wide-angle even during motion. At the system level, RV1126B’s AOV3.0 low-power architecture reduces standby power to 1 mW and supports 24 × 7 abnormal sound-source wake-up (dog bark, glass break, gunshot, etc.), achieving both energy saving and real-time early warning; it also integrates a super-encoding engine that can save 50 % of bitstream at the same clarity, reducing transmission and storage costs; in terms of security, it supports national cryptography SM2/SM3/SM4 algorithms, TrustZone isolation and a key-management system, providing end-to-end security from capture to inference.
Processor Architecture Details
RV1126B and RV1126B-P core configuration: quad-core ARM Cortex-A53, 64-bit architecture, supporting ARM v8-A instruction set. Cache specifications: each core has 32 KB L1 I-Cache + 32 KB L1 D-Cache, sharing a 512 KB L2 Cache. Extended features: integrated Neon Advanced SIMD and FPU, supporting TrustZone technology.
RV1126 core configuration: quad-core ARM Cortex-A7 32-bit architecture, supporting ARM v7-A instruction set, integrated Neon Advanced SIMD and FPU. Cache specifications: each core has 32 KB L1 I-Cache + 32 KB L1 D-Cache, sharing a 512 KB unified L2 Cache.
NPU Performance
RV1126B and RV1126B-P compute specifications: INT8 compute power 3.0 TOPS (supporting sparsity optimization), compatible with INT4/INT8/INT16/FP16 operations. Framework support: compatible with TensorFlow, Caffe, Tflite, Pytorch, Onnx NN, Android NN and other mainstream deep-learning frameworks.
RV1126 compute and precision: INT8/INT16 mixed operations, compute power up to 2.0 TOPS, supporting integer 8/16-bit convolution operations. Framework compatibility: compatible with TensorFlow, TF-lite, Pytorch, Caffe, ONNX, MXNet, Keras, Darknet and other mainstream deep-learning frameworks, supporting OpenVX API.
ISP Features
RV1126B & RV1126B-P, 2D Graphics Engine (RGA). Data format support: input includes ARGB/RGB/YUV series formats (including TILE4X4 packaged format), output supports ARGB/RGB/YUV420/422 and other 8-bit formats. Core functions: support 1/16 ~ 16× arbitrary non-integer scaling (down-sampling uses average/bilinear filtering, up-sampling uses bicubic filtering), 0/90/180/270° rotation and mirroring, alpha blending, color fill, OSD overlay, etc. Resolution limits: maximum input 8192 × 8192, maximum output 4096 × 4096.
Video Encoding/Decoding
Both RV1126 and RV1126B support H.265/H.264 video encoding, efficiently handling 4K ultra-HD video compression, storage and transmission. RV1126B explicitly supports multi-stream encoding, with an integrated intelligent encoding engine, supporting up to 8 MP @ 45 FPS ultra-HD encoding. Combined with dynamic bitrate control technology, it can save up to 50 % of the bitstream compared with traditional CBR mode.
RV1126B and RV1126 both have relatively rich audio, storage and other peripheral interfaces, supporting high-performance external DRAM, meeting basic multimedia and peripheral expansion needs. They integrate RTC, POR, RMI Ethernet PHY and audio codec modules; RV1126 supports 12 Kbit space OTP, providing unique identification and secure storage capability. RV1126B specifically introduces AI-ISP, supporting cooperation with the NPU; RV1126 does not support AI-ISP. RV1126B & RV1126B-P support 2-lane CANFD, suitable for automotive and industrial control scenarios; RV1126B supports USB 3.0 (RV1126B-P / RV1126 only support USB 2.0).
Leveraging the powerful 3 TOPS AI compute and AI-ISP architecture, the RV1126B series has achieved large-scale application deployment in multiple vertical industries, widely used in IPC, smart doorbells, AI PTZ cameras, and with high integration and low power advantages, realizing the upgrade from “video recording” to “intelligent analysis”.