In a recent technological communication meeting, Fu Hongwei, the CEO of Smart Software Technology Co., expressed the constraints that come with the increasing complexity and demands of automotive technologyHe remarked, “At the beginning, I had the freedom to freely add capabilities, but as we progressed, I had to squeeze our CPU usage to get by.” This observation starkly reflects the ongoing transformation within the automotive industry, where the shift has moved from an era of electrification to one of intelligent automotive systems, primarily fueled by the integration of advanced AI technologies.

As automotive manufacturers like NIO and Smart eagerly anticipate the latest chipsets from tech giants such as NVIDIA, Qualcomm, and AMD—similar to how Chinese smartphone manufacturers once chased the Snapdragon chip launches—there is a rising trend for some manufacturers to invest in their in-house chip development as a viable alternative

Pursuit of High-Performance Computing Power

During the same conference, Fu cited, “This CPU is currently the best in the industry, nearly equal to two of the 8295 in terms of processing power.” He was referring to the V2000 chip, which is featured in Smart’s latest SUV model, the Spirit 5. This chip is essential for intelligent cabin control.

The V2000 is developed by AMD and has found its way not only into Smart vehicles but is also in use by Tesla; meanwhile, the 8295 series, created by Qualcomm, powers models from luxury names like Mercedes-Benz and Zeekr

This shift emphasizes the increasing demand for high-performance chips that can support a range of functionalities in electric vehicles.

The interiors admired by electric car owners, reminiscent of plush items such as refrigerators and large sofas, indeed enhance the overall in-car experienceThe software-defined aspect of vehicles has deeply intertwined with their operations, signaling a major transformation from pre-defined, unchanging fuel vehicles to adaptable, self-learning electric vehicles with ever-expanding functionalities, particularly thanks to Over-The-Air (OTA) updates.

Several onboard features like multiple displays and electrically adjustable seating rely heavily on advanced computational support, thereby pushing manufacturers to chase high-performance computing chips

For instance, when Smart was developing its Spirit 1 model, the company continuously added more functionalities, resulting in no less than nine iterations of OTA updates.

Fu noted, “Many customers have requested functionalities like NetEase Cloud Music for the next update.” As automotive displays burgeon with applications—ranging from music streaming to video platforms—the burden on CPU processing power gradually intensifies, leading him to become circumspect about resource allocation“Initially, I could add features freely, but eventually, I had to meticulously manage CPU power, making a high-performance CPU essential for future expansions.”

The advent of automotive intelligence is driven by two key lines: the smart interior system, which enhances digital engagement and involves smooth interaction through natural language, and the automation of driving, which presents its own complexities and potential for disruption.

In the context of cabin technologies, Qualcomm and AMD have emerged as primary contenders, with Qualcomm holding a larger market share, closely followed by AMD and others like Renesas and Intel

alefox

Meanwhile, in the domain of intelligent driving chips, voluntary participation for companies like Tesla, NVIDIA, Horizon Robotics, and Mobileye further fuels the competitive landscape.

In previous years, automakers took three years or more to develop a single model; however, the development cycle for electric vehicles has shrunk to as little as two years or even eighteen monthsThis rapid pace of development has forced both automobile manufacturers and chip producers to pursue high-performance computing chips aggressively.

“In the automotive industry, the speed of iterations is a crucially important focus,” observed Nakul Duggal, General Manager of Automotive, Industry Solutions, and Cloud Group at Qualcomm during the 2024 Snapdragon Summit

He emphasized the need to design products capable of remaining relevant throughout the vehicle's lifecycle while also being adaptable to consumer demands for more frequent feature upgrades“Among our diverse clientele, some require faster update cycles and expect a variety of product options.”

Unifying Intelligent Driving and Cabin Technologies

The automotive industry is navigating through a transition phase as it moves from a software-defined approach towards an AI-defined framework.

As NVIDIA’s Global Vice President Wu Xinzhao noted earlier this year, “With the widespread advancements of generative AI, an AI-defined automotive future is undoubtedly on the horizon.” He believes that this progression in generative AI will significantly elevate the limits of autonomous driving technologies.

Wu classified the evolution of autonomous driving technologies into three stages: the initial phase, which was built entirely on rigid rules; the second phase where AI began to replace traditional engineering parameters; and the third phase, which incorporates an end-to-end modeling approach utilizing large-scale models.

Chip manufacturers are keen on unifying both intelligent driving and cabin systems

NVIDIA has gained substantial market recognition in the autonomous driving realm and is now extending its capabilities into cabin control, while Qualcomm, dominating in smart cabin technology, aspires to penetrate deeper into smart driving.

NVIDIA’s Orin chips stand out as a top choice for automakersFor instance, NIO's ET7 is equipped with four Orin chips, while Smart’s Spirit 5 features two Orin chips, accomplishing an end-to-end, map-free autonomous driving solution.

Moreover, NVIDIA has high aspirations for its upcoming Thor chip, envisioning a future where a single AI computing platform in vehicles will facilitate both intelligent driving and smart cabin functionalities

“NIO, Xpeng, Li Auto, and BYD have confirmed that their next-gen autonomous driving platforms will be built using the DRIVE Thor architecture,” Wu remarked.

In October 2024, Qualcomm unveiled the Snapdragon Cockpit Platform and Snapdragon Ride Platform, featuring a threefold increase in CPU speed, up to twelve times enhanced AI performance, and a threefold boost in GPU capabilitiesAutomotive manufacturers can seamlessly run both digital cockpit and intelligent driving functionalities on the same SoC within this new platform, with companies like Li Auto and Mercedes-Benz announcing their plans to adopt it in future mass-produced models.

The need for automakers is clear—they seek a central processing unit that can integrate functionalities for intelligent driving and smart cabin controls.

With the rising number of chips installed in electric vehicles, where the most expensive components are typically those related to intelligent driving and cabin control, automotive manufacturers are recognizing the importance of flexibility within a unified platform that can run basic ADAS and infotainment functions

For example, Smart’s Spirit 5 utilizes AMD for cabin controls while depending on NVIDIA for intelligent drivingIf a single chip supplier can deliver a comprehensive solution, automakers could significantly reduce negotiation and deployment costs.

Furthermore, the demand for computational power extends beyond the vehicle itself, often relying on cloud infrastructureTraining large-scale models necessitates an array of high-performance chips, predominantly powered by cloud computing platforms.

“We anticipate that over the next three years, the scale of models will increase thirteenfold and data storage needs will expand by seventeen times

Reflecting on earlier ventures, models built on Transfomer architecture require approximately 3,000 server nodes, translating to around 24,000 GPUsTo reach the standards of GPT-4, the necessary infrastructure would surpass tens of thousands of server nodes,” Wu elaborated.

The consumer demand for intelligent driving features and smart cabin technologies is unidirectionalAs vehicles become increasingly intelligent, once individuals acclimate to natural voice interactions, reverting to traditional buttons or touch screens becomes unlikely; moreover, once accustomed to adaptive cruise control technology, drivers would be disinclined to return to old driving habits on highways.

This steadfast technological evolution propels automakers and chip manufacturers to relentlessly pursue greater computational power, ensuring they remain at the forefront of a rapidly transforming automotive landscape.