Inspur Information releases the first intelligent domain controller EIS400 to provide super computing power for intelligent driving

On May 25, Inspur Information's "Intelligent Computing Opens a New Game·Innovation Machine" national tour exhibition was held in Hangzhou. With the theme of new technologies and new applications in the era of intelligent computing, it gathered industry guests from government departments, industry organizations, well-known enterprises, etc. Talk about new trends and opportunities in the intelligent computing industry.

At the conference, Inspur Information's edge computing product family welcomed a new member and launched the first intelligent domain controller EIS400, which can provide data center-level powerful, safe and efficient vehicle-mounted edge computing power for intelligent driving. EIS400 supports the largest number of cameras, radars, and inertial navigation interfaces in the industry, covering the computing power needs of various autonomous driving scenarios such as autonomous taxis, autonomous buses, autonomous trucks, and unmanned cargo vehicles. At the same time, Inspur Information also released the first autonomous driving computing framework AutoDRRT (Autonomous Driving Distributed Robust Real-Time). Based on the highly open EIS400, it can provide users with autonomous driving perception, planning and decision-making, control algorithm API interfaces, computing parallelism and acceleration Modules and development tools accelerate the rapid development of the intelligent driving industry.

The "intelligence" of automobiles is accelerating, with computing power taking the lead

In recent years, with the rapid development of intelligent and autonomous driving technology, intelligent driving functions have become increasingly rich, including driving, parking, and vehicle safety. The application scenarios of various functions are also more complex. For example, the driving function involves lane centering control, automatic lane change assistance, traffic jam assistance, high-speed pilot driving assistance and other scenarios. Complex application scenarios place higher demands on computing performance, reliability, and software and hardware collaboration.

In this process, the domain controller, the core of in-vehicle computing, needs to process data from various sensing devices including lidar, cameras, millimeter-wave radar, etc. The underlying chip architecture and upper-layer application systems are becoming more and more complex, making technology development more difficult. Come bigger. Currently, in order to allow vehicles to more accurately perceive the interior and road environment, domain controllers need to access more and more devices, the number of cameras has increased significantly, and the detection distance and pixels have also been continuously improved. Judging from the needs of car users in the industry, currently a car’s 14 5-megapixel cameras require at least 1,000 TOPS of computing power for deep learning calculations. At the same time, in order to ensure the security of application operation, very high requirements are also placed on the data processing time and communication transmission delay of the domain controller.

In addition, as the level of autonomous driving increases, the demand for computing power will increase exponentially. A domain controller usually needs to match multiple processor chips, which not only brings high power consumption, heat dissipation and reliability challenges to the system, but also poses challenges to the system. Reducing the latency of distributed computing puts forward stringent requirements.

EIS400 provides data center-level on-board computing power for on-board computing.

Inspur Information's first intelligent domain controller EIS400 relies on Inspur Information's strong data center-level product research and development capabilities to extend data center-level safe, reliable, and efficient computing power to the edge of the automobile, and targets the harsh environment of in-vehicle computing. It has been optimized for various computing needs and has extremely strong computing power, extremely high security and efficient heat dissipation capabilities.

Due to the significant increase in the number of various types of vehicle sensors, including lidar, cameras, millimeter-wave radar, etc., the detection pixels, frame rates, and distance accuracy are also constantly improving, requiring higher on-board computing power. EIS400 has been optimized in terms of computing performance and interface scalability, and supports the industry's widest range of terminal interfaces. It can support up to 16 vehicle cameras, 8 lidars, 4 millimeter wave radars, and 1 inertial navigation access. It covers the high computing power and low latency requirements of various self-driving scenarios such as self-driving taxis, self-driving buses, self-driving trucks, and unmanned cargo vehicles.

In addition, in order to ensure the safety of intelligent driving and vehicle privacy, the EIS400 motherboard is embedded with a car-grade functional safety chip, which can perform key tasks such as vehicle control and system status management, and has multiple redundant designs to improve the security of the entire system. and usability to ensure road safety and vehicle privacy.

In view of the environmental characteristics of high temperature, high cold, high altitude, rainy fog, strong electromagnetic interference and complex geological conditions in mines and highways where some self-driving trucks are located, EIS400 fully considers these environmental factors, designs an efficient heat dissipation system, and supports liquid cooling. Heat dissipation not only provides good performance of computing power, but also ensures high stability, high reliability of the entire system and high continuity of vehicle operation.

Support the autonomous driving computing framework AutoDRRT to create an open ecosystem for in-vehicle computing

In order to promote the rapid development of the intelligent driving industry, at this conference, Inspur Information also released the first autonomous driving computing framework AutoDRRT. AutoDRRT is a highly open autonomous driving distributed high fault-tolerant and low-latency computing framework. It is based on the EIS400 heterogeneous distributed architecture design and provides algorithm API interfaces for perception, decision planning, control and other algorithms upward for the autonomous driving application layer, as well as parallel computing and Acceleration modules and development tools; downwardly compatible with open source middleware and OS for the underlying system layer, supporting users to develop their own software, allowing users to select appropriate algorithms for different intelligent driving application scenarios and achieve software and hardware collaboration, thereby quickly building and deploying smart devices Driving applications.

AutoDRRT has three innovative computing functions: distributed, high fault tolerance, and low latency. Among them, the distributed computing function is oriented to users’ needs for rapid migration of autonomous driving applications and supports distributed parallel computing from a single computing engine to multiple computing engines. Users can implement hundreds of different algorithms on 5 computing engines without code development. distributed computing. In order to solve the problem of safe operation of the application layer, AutoDRRT also designed a highly fault-tolerant computing function to achieve redundancy in computing, communication, and IO. When a certain automatic driving algorithm fails due to system failure, it can switch to the redundant algorithm in real time, and the switching delay time as low as 1ms to ensure system security. In addition, the low-latency computing function adopts software and hardware collaborative optimization technology to solve the low-latency challenge of application operation and realize the end-to-end delay from perception to control of autonomous driving applications as low as 60ms, which is shorter than the average application running time in the industry. The delay is reduced by 40%, which can better meet the real-time requirements of autonomous driving.

With the current development of automobile intelligent technology, automobile manufacturers and technology companies have invested heavily in promoting the research and development of automobile intelligence and autonomous driving technology. Sun Bo, general manager of Inspur Information Edge Product Department, said that the "intelligentization" of automobiles is accelerating, and computing power takes precedence. With the rapid development of automobile intelligence and autonomous driving technology, various scenarios have a huge impact on computing power, security, software and hardware collaboration, etc. Putting forward higher requirements. Intelligent computing power needs to be continuously extended from cloud data centers to edge vehicle computing to promote the development of intelligent driving applications. "With the release of Inspur Information's first intelligent domain controller EIS400 and autonomous driving computing framework AutoDRRT. , providing data center-level super computing power for all types of on-board computing, as well as a more convenient, efficient and secure algorithm development platform, to help the rapid development of the intelligent driving market.