The RPP architecture, independently developed by XDL Technologies Inc. is the world's first chip architecture designed for parallel computing, realizing an all-around balance of low cost, low power consumption, low latency, high performance, rapid deployment and wide range of applications, and bringing a new solution to the AI computing field.The RPP architecture combines the CGRA and CUDA languages, overcoming the problem of reconfigurable array programming, and significantly improving the efficiency of parallel computing. The RPP architecture combines CGRA and CUDA language to overcome the challenge of reconfigurable array programming and significantly improve parallel computing efficiency. The RPP architecture combines CGRA and CUDA language to overcome the challenge of reconfigurable array programming, significantly improving the efficiency of parallel computing.

Compared with mainstream technologies in the market (such as NVIDIA's GP-GPU), the RPP architecture shows significant advantages in terms of computational density and power efficiency. With the same chip area and power consumption, RPP is able to output more computing power. the RPP-R8 chip is measured to process 1,500 images per second in the ResNet-50 test, with a performance far exceeding that of NVIDIA Xavier, and with lower power consumption.

The AE7100 chip, designed based on RPP architecture, has an outstanding performance, with a total area of 1/3 of Nvidia AGX chip under the same computing power, integrating 1024 computing cores, and a compact size of 17mmX17mm, which makes it suitable for embedded and edge computing.The AE7100 supports a variety of programming languages to simplify the development of applications, and has a high energy-efficiency ratio of up to 5 times that of Nvidia Jetson Xavier.The AE7100 supports multiple programming languages, simplifying application development. The AE7100 chip also offers powerful programming flexibility. It supports RPP assembly and CUDA C/C++, simplifying the application development process.
The chip products have been successfully adapted to Llama3-8B, Stable Diffusion, Tongyi Qianqian and other advanced models, and the application areas have covered the edge of large models, industrial automation, security monitoring, content filtering, medical imaging and signal processing.

This project is expected to play an important role in the deployment of big model end-side devices, promote the development and application of AI technology, and have a positive impact on China's electronics industry.