RV1103/1106/1109/1126&RK3562/3568/3576/3588 Параметры AI

This is a very detailed description of the functions and features of Rockchip series SoC processors, including RV1103, RV1106, RV1109, RV1126 as well as RK3562, RK3568, RK3576, RK3588. These chips are designed for use in IPC (Internet Protocol Cameras), CVR (Video Recorders) and AI-related applications, and their features are as follows:

RV1103 & RV1106

RV1103 is a highly integrated SoC for IPC vision processors, especially suitable for AI-related applications. It is based on a single-core ARM Cortex-A7 32-bit core with integrated NEON and FPU, featuring 32KB I-cache, 32KB D-cache and 128KB unified L2 cache.

RV1106 is also a highly integrated SoC for IPC vision processors, particularly applicable to AI-related applications. It is based on a single-core ARM Cortex-A7 32-bit core with integrated NEON and FPU, equipped with 32KB I-cache, 32KB D-cache and 128KB unified L2 cache.

Core: Single-core ARM Cortex-A7, supporting NEON and FPU.
NPU: Supports INT4/INT8/INT16, with a computing power of 0.5 TOPs.
ISP: RV1103 supports 4 megapixels, while RV1106 is upgraded to support 5 megapixels.

RV1103：Supports two MIPI CSI or LVDS.

It also supports multi-stream encoding. With this function, the video from the camera can be encoded at a higher resolution and stored in the local memory, while another video with a lower resolution is transmitted to the cloud storage. To accelerate video processing, a smart video engine with 22 computing units is also embedded. RV1103 has built-in 16-bit DRAM DDR2 and can withstand demanding memory bandwidth. It also integrates a built-in Power-On Reset (POR), an audio codec and a Media Access Control Physical Layer (MAC PHY).

RV1106: Supports two MIPI CSI and one DVP.

The video encoder embedded in RV1106 supports H.265/H.264 encoding. It also supports multi-stream encoding. With this function, the video from the camera can be encoded at a higher resolution and stored in the local memory, while another video with a lower resolution is transmitted to the cloud storage. To accelerate video processing, a smart video engine with 22 computing units is also embedded.

RV1106 has a built-in 16-bit DRAM DDR3L, which can withstand demanding memory bandwidth. It also integrates a built-in Real-Time Clock (RTC), Power-On Reset (POR), audio codec and Media Access Control Physical Layer (MAC PHY). Video encoding: It supports H.265/H.264 and multi-stream encoding.
RV1109 is a high-performance AI vision processor SoC for IPC (Internet Protocol Cameras) or other intelligent vision applications. It is based on a dual-core ARM Cortex-A7 32-bit core with integrated NEON and FPU. Each core has 32 KB I-cache and 32 KB D-cache, as well as 512 KB unified L2 cache. The built-in NPU supports INT8/INT16 mixed operations, and its computing power is as high as 1.2 TOPs. In addition, with its powerful compatibility, it can easily convert network models based on a series of frameworks such as TensorFlow, MXNet, PyTorch and Caffe.

RV1109 also introduces a new generation of fully hardware-based 5-megapixel ISP (Image Signal Processor) and post-processor. It implements many algorithm accelerators that are commonly used in IPCs (Internet Protocol Cameras) and CVRs (Video Recorders), such as HDR, 3A functions (AE - Auto Exposure, AF - Auto Focus, AWB - Auto White Balance), LSC (Lens Shading Correction), 3DNR (3D Noise Reduction), 2DNR (2D Noise Reduction), sharpening, defogging, fisheye correction, gamma correction, feature point detection, etc. All of these are processed in real time. With two MIPI CSI and one DVP (BT.601/BT.656/BT.1120) interfaces, users can build a system that simultaneously receives video data from three camera sensors. The video encoder embedded in RV1109 supports 5M H.265/H.264 encoding. It also supports multi-stream encoding, and can simultaneously handle one 5M30FPS and one 720P30 at most. With this function, the video captured by the camera can be encoded at a higher resolution and stored in the local memory, while another video with a lower resolution is transmitted to the cloud storage.

RV1109 & RV1126

Core:

RV1109: Dual-core Cortex-A7.

RV1126: Quad-core Cortex-A7.

NPU:

RV1109: 1.2 TOPs.

RV1126: 2.0 TOPs.

ISP:

RV1109: 5 megapixels.

RV1126: 14 megapixels, supporting more advanced real-time image processing algorithms.

RV1126 is a high-performance vision processor SoC for IPC (Internet Protocol Cameras) / CVR (Video Recorders), especially suitable for AI-related applications. It is based on a quad-core ARM Cortex-A7 32-bit core with integrated NEON and FPU. Each core has 32 KB I-cache and 32 KB D-cache, as well as 512 KB unified L2 cache.
RK3562 includes an embedded 3D GPU, which ensures full compatibility with OpenGL ES 1.1/2.0/3.2, OpenCL 2.0 and Vulkan 1.1. In addition, it also includes a special 2D hardware engine that can maximize display performance and ensure smooth operation. The built-in NPU supports mixed operations with INT4/INT8/INT16/FP16 data types. 。

In addition, it also has strong compatibility with a series of frameworks such as TensorFlow, MXNet, PyTorch and Caffe, and can easily convert network models.

RK3562 has a high-performance external memory interface (DDR3/DDR3L/DDR4/LPDDR3/LPDDR4/LPDDR4X) and can withstand demanding memory bandwidth. RK3568 is a high-performance, low-power quad-core application processor, specially designed for personal mobile Internet devices and AIoT devices. It provides many embedded powerful hardware engines to optimize the performance of high-end applications.

RK3568 supports almost all formats of H.264 decoder at 4K@60fps, H.265 decoder at 4K@60fps, and also supports H.264/H.265 encoder at 1080p@60fps, as well as high-quality JPEG encoder/decoder. The embedded 3D GPU makes RK3568 fully compatible with OpenGL ES 1.1/2.0/3.2, OpenCL 2.0 and Vulkan 1.1. The special 2D hardware engine will maximize display performance and provide a very smooth operation. The built-in NPU supports INT8/INT16 mixed operations. In addition, with its powerful compatibility, it can easily convert network models based on a series of frameworks such as TensorFlow, MXNet, PyTorch and Caffe. RK3568 has a high-performance external memory interface (DDR3 / DDR3L / DDR4 / LPDDR3 / LPDDR4 / LPDDR4X) and can withstand demanding memory bandwidth.

RK3576 is a low-power, high-performance processor, suitable for ARM-based PCs and edge computing devices, personal mobile Internet devices and other digital multimedia applications. It integrates quad-core Cortex-A72 and quad-core Cortex-A53, both with NEON coprocessors.

The video decoder of RK3576 supports H.264, H.265, VP9, AV1 and AVS2, etc., with a maximum resolution of 8K@30fps or 4K@120fps. The video encoder supports H.264 and H.265, with a maximum resolution of 4K@60fps. The high-quality JPEG encoder/decoder can reach a maximum resolution of 4K@60fps. The embedded 3D GPU makes RK3576 fully compatible with OpenGL ES 1.1, 2.0 and 3.2, OpenCL up to 2.0 and Vulkan 1.1. The dedicated 2D hardware engine with MMU (Memory Management Unit) will maximize display performance and provide a very smooth operation.
RK3576 introduces a new generation of 16-megapixel ISP (Image Signal Processor). It implements many algorithm accelerators, such as HDR (High Dynamic Range), 3A (Auto Exposure, Auto Focus, Auto White Balance), CAC (Chromatic Aberration Correction), 3DNR (3D Noise Reduction), 2DNR (2D Noise Reduction), sharpening, defogging, enhancement, Debayer, small-angle lens distortion correction and so on. The built-in NPU supports INT4 / INT8 / INT16 / FP16 / BF16 / TF32 mixed operations. In addition, thanks to its powerful compatibility, network models based on a series of frameworks such as TensorFlow, MXNet, PyTorch and Caffe can be easily converted.
RK3588 is a low-power, high-performance processor, which is applicable to ARM-based PCs, edge computing devices, personal mobile Internet devices and other digital multimedia applications. It integrates quad-core Cortex-A76 and quad-core Cortex-A55, both of which are equipped with NEON coprocessors respectively.

Many embedded powerful hardware engines provide optimized performance for high-end applications. RK3588 supports H.265 and VP9 decoders at 8K@60fps, H.264 decoder at 8K@30fps and AV1 decoder at 4K@60fps. It also supports H.264 and H.265 encoders at 8K@30fps, high-quality JPEG encoder/decoder, dedicated image pre-processors and post-processors. The embedded 3D GPU makes RK3588 fully compatible with OpenGL ES 1.1, 2.0 and 3.2, OpenCL up to 2.2 and Vulkan 1.2. The dedicated 2D hardware engine with MMU (Memory Management Unit) will maximize display performance and provide a very smooth operation.

RK3588 introduces a new generation of fully hardware-based ISP (Image Signal Processor) with a maximum of 480,000 pixels. It implements many algorithm accelerators, such as HDR (High Dynamic Range), 3A (Auto Exposure, Auto Focus, Auto White Balance), LSC (Lens Shading Correction), 3DNR (3D Noise Reduction), 2DNR (2D Noise Reduction), sharpening, defogging, fisheye correction, gamma correction and so on.

The built-in NPU supports mixed operations of INT4 / INT8 / INT16 / FP16, and its computing power is as high as 6 TOPs. In addition, thanks to its powerful compatibility, it can easily convert network models based on a series of frameworks such as TensorFlow, MXNet, PyTorch and Caffe.

RK3588 has a high-performance four-channel external memory interface (LPDDR4 / LPDDR4X / LPDDR5), which can withstand demanding memory bandwidth. It also provides a complete set of peripheral interfaces to support highly flexible applications.