Facing the Challenge of HCP in Vehicles, Enjoy Move Tech Lauches a HPC Cluster: XCG
- 2022-07-25 18:56:52
- author:
- click0
In the upcoming era of intelligent electric vehicles, rapid technological iterations demand that hardware and software evolve synergistically to deliver the ultimate user experience.
Enjoy Move Tech specializes in the research, development, and services of high-performance computing software platforms and autonomous driving software products for intelligent vehicles.
1 Recently, Enjoy Move Tech has unveiled the XCG (X-Computing Grouplet) Gen1, a versatile vehicle-grade high-performance computing cluster built upon Horizon Journey 5 AI chips and high-performance processors. Through deep collaboration with chip manufacturers, proactive hardware development, and a rich software ecosystem, Enjoy Move Tech is accelerating product deployment to embrace the era of high-computing power in vehicles.
The Rising Tide of In-Vehicle Computing: Opportunities and Challenges
With the advent of intelligent vehicles, in-vehicle computing has become a focal point.
The evolution of CPU computing power in vehicles has been a gradual process. Traditionally viewed as mere transportation, vehicles had relatively modest computational requirements. However, in recent years, vehicles have transformed into "intelligent mobile spaces," leading to a surge in diverse usage scenarios and expectations. Large and multiple screens have become standard features in connected vehicles. Chipmakers like Qualcomm and Intel have even prioritized automotive applications, bringing their high-end chips directly to vehicles.
All these advancements are fueled by the continuous breakthroughs and improvements in in-vehicle computing power. Currently, CPU computing power is experiencing exponential growth, moving from a plateau to a rapid expansion phase. By 2025, CPU computing power is projected to reach 500kDMIPS, while in-vehicle AI computing power is expected to soar to 1000TOPS.
However, while computing power is rapidly increasing, software development has often lagged behind. In-vehicle software has historically been constrained by the limitations of computing capabilities. There have been periods where computing power was insufficient, followed by periods of excess. Although automakers have implemented OTA (Over-the-Air) updates, it typically takes 1-2 years to fully utilize the excess computing capacity.
The explosive growth of software in vehicles is driving up development costs for automotive chips and controllers, creating a dual challenge of performance and cost. This has spurred the development of new chip technologies to strike a balance between cost and capability. Chiplets, consisting of modular chips with different processes integrated through packaging technology, offer a solution by enhancing performance while achieving lower costs and higher yields. The recent introduction of the universal chiplet interconnect standard, UCle, has further fueled the momentum behind chiplets.
In the development of future zonal controllers, there is a need to control overall development costs while inheriting legacy software and IP assets, and simultaneously achieving rapid improvements in core performance. To address the diverse and demanding computational requirements of zonal control, it is necessary to connect and combine different computing modules. This context has given rise to high-performance computing clusters, or grouplets.
Comprehensive Solutions for Multi-Scenario Applications
While high-performance chips and controllers offer increased computational efficiency, they also pose significant thermal management challenges. Each high-performance controller generates substantial heat, necessitating centralized cooling to ensure vehicle safety and performance. This has led to the concept of central computing.
Enjoy Move Tech has introduced a new high-performance computing cluster solution – X-Computing Grouplet (XCG) Gen1. This is a versatile solution designed for intelligent driving, intelligent central computing control, and intelligent cockpits.
To deliver high-performance computing, the XCG Gen1 employs the popular Horizon Journey 5 chip as its primary chip, complemented by NXP S32G 399 and TI TDA4 as high-performance logic computing units. The system boasts up to 20 CPUs, 4 DSPs, 2 CV engines, and a GPU, achieving powerful AI computing capabilities.
As a computational hub, the XCG Gen1 offers comprehensive and high-speed interfaces. Thanks to increased bandwidth and backend processing capabilities, this high-performance computing cluster can support up to 12 cameras, all with 8-megapixel resolution. Additionally, it provides 10 Ethernet interfaces and can accommodate 4D radar, lidar, and ultrasonic sensors.
The XCG Gen1 offers a highly flexible hardware and software configuration, enabling rapid development and customization for mass production solutions. This high-performance computing cluster provides a flexible hardware configuration, allowing for 1-2 Horizon Journey 5 chips to be selected on the controller to address diverse intelligent driving and intelligent cockpit applications. Additionally, different MPUs and MCUs can be chosen to create customized hybrid computing cluster platforms.
A robust platform is essential for any solution. This solution is built upon Enjoy Move Tech's EMOS high-performance computing software platform, which has been extensively validated in production and ensures deterministic scheduling and communication. It also seamlessly integrates with mature AUTOSAR platforms from Elektrobit and ETAS and incorporates the RTI DDS communication protocol stack. Autonomous driving places stringent requirements on scheduling and communication latency, and EMOS effectively reduces latency from the sensor end to the actuator, providing a safe and reliable software platform and application ecosystem for autonomous driving.
A mature platform relies on a comprehensive toolchain. The XCG Gen1 comes with Enjoy Move Tech's efficient toolchain, including EM Studio, GCase, and calibration tools. Moreover, the solution adopts multiple security measures and achieves an ASIL-D functional safety level to facilitate efficient development and rapid deployment for users.
Software-First Approach: Bridging the Gap Between Chips and Autonomous Driving
In recent years, the rapid development of automotive perception and AI computing power has accelerated the realization of autonomous driving. Concurrently, the trend of software-defined vehicles has become increasingly evident: barriers between different vehicle domains are being broken down, with domain controllers and central computing platforms enabling centralized computing. The traditional model of tightly coupled hardware and software is gradually transitioning to a decoupled approach, and software decoupling is also underway. Under this trend, competition among software companies is undoubtedly intensifying.
As the core main control chip of the XCG Gen1, the Journey 5 chip boasts an impressive AI computing power of 128 TOPS thanks to its dual-core BPU. Moreover, it features 8 A55 DynamIQ CPU clusters that can be effectively allocated to various autonomous driving applications. The S32G 399, another critical chip in the system, offers robust computing power that complements the Journey 5. With the integration of these powerful chips, the computing cluster solution undoubtedly has a solid foundation in terms of computing power.
While hardware integration has reduced costs, it has also increased the complexity and quantity of software. As a partner of Horizon Robotics and NXP, Enjoy Move Tech has made numerous optimizations in both hardware and software, aiming to collaborate with partners in the domain controller and high-performance computing cluster fields.
In the era of software-defined vehicles, automotive product development faces a myriad of challenges, including managing massive amounts of code, complex integration, the need for rapid delivery and iteration, and extended development cycles. To address these challenges in the realm of basic software, Enjoy Move Tech provides not only near-production-ready software packages but also customized software packages tailored to specific computing clusters. Through XCG technology products and services, Tier 1 suppliers and OEMs can rapidly develop mass production solutions.
Furthermore, to meet the high performance and reliability requirements of autonomous driving while maintaining flexibility and efficiency in deployment, Enjoy Move Tech's solutions adopt a Service-Oriented Architecture (SOA) framework. Built upon the high-performance computing software platform EMOS and complemented by tools like EMOS Studio and GCase, this solution creates a comprehensive software ecosystem that bridges the gap between chips and autonomous driving technology.
The transformation of the automotive industry presents both challenges and unprecedented opportunities for the supply chain. Openness, integration, and co-creation will be the inevitable trends in the future automotive industry. Leveraging its extensive experience and capabilities in middleware and underlying software development, Enjoy Move Tech will collaborate with partners to accelerate the mass production and deployment of intelligent driving applications, continuously creating value for the new energy vehicle industry.
The above content is from a keynote speech titled "Research and Development of High-Performance Computing Clusters" delivered by Zhao Jianhong, Vice President of Products at Enjoy Move Tech, at the 2nd Intelligent Vehicle Domain Controller Innovation Cloud Forum hosted by Gasgoo on June 29, 2022.
Disclaimer
Part of the content of this website comes from content authors or automatic crawling from the Internet. The content of the relevant text only represents the views of the author or publisher of this article and does not represent the views or positions of Enjoy Move Tech. Enjoy Move Tech strives for the objectivity and impartiality of the content and opinions described in this information but does not guarantee the accuracy and completeness of the content, nor does it guarantee that the content will not be changed in the future. If the authors and editors of the content displayed on this website think that their works are not suitable for browsing on the Internet or should not be used free of charge, please inform us by email or telephone in time. Enjoy Move Tech will take reasonable measures in timely to avoid unnecessary economic losses to both parties.
Email us: contact@enjoymove.com
CALL US
FOLLOW US
沪公网安备 31010402009313号 | 沪ICP备19020085号