From dependence to autonomy: the rise and future of China's intelligent driving chips

Are domestic intelligent driving chips really going to stand up?

In the past two months, domestic car companies have been experiencing "frequent major events" in the field of intelligent driving chips. At the end of July, at the "NIO IN 2024 NIO Innovation Technology Day", NIO announced the successful tape-out of its first automotive-grade 5nm intelligent driving chip "Shenji NX9031".

According to the parameters announced by NIO, the Shenji NX9031 has more than 50 billion transistors. The computing power is said to reach 4 Nvidia Orin X chips (254 Tops each), which is about 1,000 Tops.

Similarly, on the evening of August 27, He Xiaopeng, chairman and CEO of Xpeng Motors, announced that Xpeng's independently designed intelligent driving chip Turing chip was successfully taped out on August 23. The Turing chip can be used for L4 autonomous driving.

According to information released by Xiaopeng, this Turing chip has a 40-core CPU, two NPUs, and two independent image ISPs. It can also support large models with 30 billion parameters running on the terminal side.

He Xiaopeng claimed that when used in L4 autonomous driving cars, one Turing chip is comparable to the current three mainstream smart driving chips.

In addition, there are reports that Ideal has also begun to focus on self-developed intelligent driving chips. Its self-developed intelligent driving chip project is code-named "Schumacher" and will be completed within the year.

So, with so many strong players emerging, a reader asked Director Xinghai: Does this mean that domestic car companies are really going to overthrow the traditional overlords in the field of intelligent driving?

In today's article, I will talk to you about domestic substitution in the field of intelligent driving chips.

01 The cake of intelligent driving chips

In the circle of intelligent driving chips, NVIDIA is definitely the boss.

According to data from Gasgoo Automotive Research Institute, in the ranking of installed capacity of intelligent driving chips in the Chinese market in 2023, Tesla's FSD chip ranked first, with sales of approximately 1.208 million chips, accounting for 37% of the market share; Nvidia's Orin-X chip followed closely behind, with sales of 1.095 million chips, accounting for 33.5%;

The domestically produced Horizon Journey 5 chip ranked third, with 200,000 units sold and a market share of 6.1%.

This shows that it is indeed not an easy task to grab the cake from the table of giants.

Thanks to the Orin-X's single-chip computing power of 254TOPS, NVIDIA has been unrivaled in the field of intelligent driving chips for a long time.

Such a situation means that any car company that wants to enter the field of intelligent driving has to look to Nvidia for support.

For example, data from the Gaogong Intelligent Automobile Research Institute shows that in 2023, NIO, Xpeng and Li Auto alone will contribute nearly 90% of Nvidia Orin chip's original equipment market share in China.

So the question is, why is Nvidia so powerful in the field of intelligent driving chips?

In Laoju's opinion, in addition to its super computing power as its hard power, its biggest advantage is its constantly updated products and complete software and hardware ecosystem.

To be more specific, Nvidia not only sells chips, but also provides a complete set of autonomous driving solutions - Drive AGX platform, which includes hardware, software, algorithms and development tools. The Drive AGX platform also includes Drive OS operating system, CUDA and TensorRT and other development frameworks.

In this way, car companies and developers can easily build and optimize intelligent driving algorithms and shorten the development cycle.

Simply put, it uses NVIDIA chips, which is convenient and flexible.

This is a one-stop service for hardware and software, from chips to development platforms. Nvidia directly provides car companies with a set of "idiot-proof" tools, which can be used as they are. The development cycle? It is shortened in an instant.

What’s even more amazing is the modular design. L2 and L3 levels can be handled with one chip; L4 and L5 can also be solved with dual or multi-chips, which means that car companies can use the same hardware platform to develop different levels of autonomous driving systems. Whether it’s low-end or high-end, you can play it however you want.

Weilai ES6-NIO is equipped with 4 Orin-X chips

More importantly, this dual-chip or multi-chip configuration is not just a simple hardware stack, but is intelligently managed through software to schedule task allocation and data communication between multiple chips.

To sum it up in one sentence: NVIDIA's strength lies not only in its hardware, but also in the fact that through its complete ecosystem, it has paved the way for car companies to take and the obstacles they have to overcome in the field of intelligent driving.

Such experience and effects make all car companies feel very smooth.

In the field of intelligent driving, Tesla and domestic new forces initially chose Mobileye as their chip supplier of choice, which entered the market earlier. However, Mobileye's ecosystem is relatively closed compared to NVIDIA, and it is not as flexible and scalable as NVIDIA in terms of overall performance. Later, it adopted the "black box model" of traditional suppliers, that is, what is inside the box and how it works are all trade secrets.

This is undoubtedly difficult to accept for some large car companies.

Therefore, even when the computing power is similar, many high-end intelligent driving models will still choose NVIDIA as their supplier.

Since it works well, why do domestic car companies have to take on this difficult task and pursue independent research and development?

02 The necessity of self-research

Nvidia's chips are indeed useful in the field of intelligent driving and have a high market penetration rate, but the price is so painful.

In 2022, Nvidia Orin chips will be available at a price between $400 and $500 (about RMB 3,000). If a car uses two Nvidia Orin-X chips, the cost of the main computing chip of its intelligent driving system will reach more than RMB 6,000.

What's even more painful is that for every extra Orin-X chip used by car companies, the price they sell to consumers will increase by 20,000 yuan! Domestic car companies and consumers are being fleeced by Nvidia!

Moreover, the new energy vehicle industry is in a state of internal competition! Car companies require that the price of L2+ autonomous driving solutions should be further reduced in the future, and chip companies should have stronger cost control capabilities.

On the other hand, the increasing complexity of autonomous driving algorithms will require chip computing power to continue to rise. In summary, a chip that can support the efficient operation of Transformer and has an advantageous price is a real talent!

And, more importantly, if car companies do not have the ability to independently develop software and hardware, it means that the "product definition rights" of the car companies are still in the hands of chip suppliers.

Such "right to define" depends largely on whether the car company has the ability to deeply customize and fully control itself.

In the field of intelligent driving chips, players who have already achieved self-research, such as Tesla and Huawei, have all put in a lot of effort in software, algorithm and hardware, forming their own unique advantages.

Although NVIDIA's NVIDIA DRIVE platform provides an open software ecosystem, it is ultimately a relatively general solution.

Take Tesla for example. Its FSD autonomous driving software is already a leader in the industry. However, without vertical integration from hardware to software, the specific algorithm requirements in FSD, such as end-to-end neural network reasoning, cannot be tailored to its own workload, let alone optimizing performance and energy efficiency.

By the same token, Huawei's MDC platform is not just an autonomous driving solution. It also integrates Huawei's technological advantages in communications, cloud computing, vehicle-road collaboration (V2X) and operating systems (such as HarmonyOS), thus forming a truly all-round competitive barrier.

This highly integrated and integrated feature of software and hardware largely defines and distinguishes the brand status of the two among many automobile companies today.

More importantly, although self-developed chips may seem like reinventing the wheel, today’s autonomous driving is a multidisciplinary and highly complex technology. There is no single solution that can fully adapt to all situations.

For example, Tesla's FSD is a purely visual perception system, and its AI algorithm relies on the processing and fusion of a large amount of camera data. Therefore, its FSD chip must also be developed in the direction of optimizing data transmission and processing paths to reduce latency.

Huawei's MDC platform focuses on multi-sensor fusion solutions, which use multiple sensing methods such as cameras, radars, LiDAR, high-precision maps, and V2X (vehicle-to-road collaboration) to achieve higher levels of autonomous driving. Such a solution requires the chip to support more complex sensor interfaces, data processing pipelines, and fusion algorithms.

Different intelligent driving solutions determine different algorithms, and different algorithms create different chips.

By developing their own chips, automakers can explore different technical routes and architectures, and continuously verify and optimize them in market competition and practical applications, so that the optimal solutions and standards gradually emerge.