Rockchip has been deeply engaged in the AI market since 2019, with its flagship AI chips RK3399Pro and RK1808 widely applied in smart homes, intelligent security, vision, gesture recognition, and smart voice fields. In the global AI chip ranking released by market research company Compass Intelligence in 2018, Rockchip ranked in the Top 20 as one of the two domestic companies.
< High-Performance AI Flagship Chips RK3399Pro & RK1808>
RK3399Pro adopts a dual-core Cortex-A72 + quad-core Cortex-A53 big.LITTLE CPU architecture, making the chip technically advanced in overall performance and power consumption. The chip integrates a quad-core ARM high-end GPU Mali-T860, further enhancing its graphics processing capabilities.
RK3399Pro has a built-in performance of up to 3.0 Tops, integrating Rockchip’s years of experience in machine vision, voice processing, and deep learning to create an NPU, allowing typical deep neural network models like Inception V3, ResNet34, VGG16 to perform excellently on it, significantly enhancing performance. RK3399Pro is an AI chip designed with a CPU + GPU + NPU hardware structure, featuring three major characteristics:
1. High AI Hardware Performance
With proprietary AI hardware design, the NPU achieves a computing performance of up to 3.0 Tops, significantly leading in both high performance and low power consumption: It outperforms similar NPU chips by 150%; Its power consumption is less than 1% compared to large chip solutions that use GPU as the AI computing unit.
2. Strong Platform Compatibility
The NPU supports 8-bit and 16-bit operations, making it compatible with various AI software frameworks. Existing AI interfaces support OpenVX and TensorFlowLite/AndroidNN API, and AI software tools support the import and mapping, optimization of Caffe/TensorFlow models.
3. Complete Solution Easy to Develop
Rockchip provides a one-stop AI solution based on the RK3399Pro chip, including hardware reference designs and software SDK, greatly improving the AI product development speed for global developers and significantly shortening product time-to-market.
These performances allow RK3399Pro to be quickly applied in various AI application fields such as smart driving, image recognition, security monitoring, drones, and voice recognition.
RK1808 is an AIoT chip built using 22nm FD-SOI technology, featuring high computing power and low power consumption. At the same performance level, RK1808’s power consumption is reduced by about 30% compared to mainstream 28nm products; It has 2MB of system-level SRAM for always-on devices without DDR operation; It features hardware VAD functionality, supporting low-power listening for far-field wake-up.
The built-in NPU of RK1808 also supports INT8/INT16/FP16 mixed operations, maximizing performance, power consumption, and computing accuracy; it supports model conversion for a series of frameworks including TensorFlow/MXNet/PyTorch/Caffe, ensuring strong compatibility.
At the same time, RK1808’s rich peripheral interfaces facilitate application expansion, supporting video input via MIPI/CIF/BT1120 and display output via MIPI/RGB; It has a series of sensor input/output interfaces like PWM/I2C/SPI/UART; It includes high-speed device interfaces such as USB3.0/USB2.0/PCIE, supporting Gigabit Ethernet and external WiFi/BT modules; Audio supports microphone array input and audio output.
RK1808’s high computing power, low power consumption features, and targeted AIoT interface design allow it to better achieve functions such as voice wake-up, voice recognition, face detection and attribute analysis, posture analysis, target detection and recognition, and image processing, making it widely applicable in security, education, cleaning, automotive, wearable devices, home appliances, storage, and various other scenarios.
To help partners quickly implement AI applications, Rockchip has released the RKNN-Toolkit and the Rock-X SDK. Developers can use RKNN-Toolkit to convert, quantize, profile performance, and analyze memory consumption of models like TensorFlow, Caffe, PyTorch, MXNet, rapidly deploying AI models onto embedded hardware platforms; Rock-X SDK is a set of quick AI component libraries based on RK3399Pro/RK1808, with the initial version including: face detection, liveness detection, face attribute analysis, facial feature points, human skeletal key points, finger key points, and object detection, allowing developers to use these features offline in embedded products with just a few API calls without worrying about AI model deployment details, greatly accelerating product prototyping and development deployment.
< Rockchip Collaborates with Partners to Achieve Rich AI Scene Implementation >
Rockchip continues to innovate technologically, developing superior performance chips, but more importantly, it partners with others for rapid implementation.
At the beginning of this year at MWC 2019, leading company in the 3D intelligent vision field, AURORA Technology, launched an ALL IN ONE integrated 3D intelligent vision development platform based on Rockchip RK3399Pro + RK1608. The platform includes 3D structured light (front) and TOF (rear) modules, deeply integrating AURORA Technology’s UNRE 3D SENZ engine, achieving precise scanning, 3D reconstruction, and object measurement for faces/objects, providing TURNKEY one-stop hardware and software 3D intelligent vision solutions for the 3D vision application industry, supporting diverse application scenarios.
Global leading gesture recognition technology provider uSens has launched a gesture recognition solution based on the Rockchip TB-RK3399Pro development platform for smart home, security, and automotive fields, combining uSens’s infrared binocular hardware FINGO with RK3399Pro, equipped with two infrared cameras, allowing gestures to have depth information, enabling tracking of three-dimensional gestures, greatly expanding the application range to meet scenarios requiring three-dimensional gesture recognition.
Developers and partners have implemented functions like customer flow statistics, object recognition, and abnormal behavior analysis based on RK1808, deploying them in shopping malls, public places such as subways/malls/train stations/communities, and smart home appliances. RK1808’s edge AI capabilities allow developers to directly process, store, and analyze large amounts of raw data locally, achieving intelligent optimization based on edge-cloud collaboration, ensuring real-time, intelligent, secure, and private operations.
Based on RK1808, Rockchip has also launched the RK1808 AI computing stick with powerful deep learning inference capabilities (purchase link: https://dwz.cn/7lULM8li). It supports PC, industrial computers, robots, etc., and its learning capabilities can be used for object detection/recognition, natural language understanding, etc., conveniently providing developers with powerful computing power and deep learning inference capabilities across various scenarios such as home appliances, robotics, new retail, industrial vision, virtual reality, augmented reality, security, education, automotive, wearables, logistics.
How to Find Me
1. Click the blue text at the top Shenzhen Automotive Electronics Industry Association to follow;
2. Search for the WeChat public account: “Shenzhen Automotive Electronics Industry Association” or “qidianxiehui”;
3. Scan the QR code to follow
