news

By Karakush

Last week, at NIO IN, NIO held its annual technology conference, it unveiled its self-developed 5nm intelligent driving chip Shenji NX9031, vehicle-wide operating system SkyOS Tianshu, Banyan 3.0, intelligent driving world model NWM, and the second-generation Nio Phone - as well as a noisy battle for the weekly charts.

To some extent, the weekly rankings are overheated because the core knowledge is too hard. People only criticize those who can criticize. Even the masses who listen carefully are like raw eggs that resolutely fall on stones, breaking all over the floor in confusion.

The last time I felt my IQ was so humiliated was when Huawei released Pangu Big Model 5.0, Ideal released end-to-end + VLM, Jiyue AI Day, Xiaopeng 1024, or last quarter when NIO IN released 12 full-stack self-developed technology layouts.

Large-scale automotive technology launches are becoming abstract.

The reasons for abstraction are very complex, such as limited audience knowledge and ruthless commentary style. Putting aside such interactive factors, from the enterprise side, these technical books can still find some commonalities:

First of all, AI is an unavoidable core word. Almost all Tianshu technologies are oriented to AI and are based on the underlying architecture/solutions developed by AI. They are the inevitable result of the new technology tree. AI brings many challenges: How to share centralized computing power? How to deploy and call it flexibly across domains? How to achieve seamless connection between end and cloud? How to enable AI to empower each business function...

For example, NIO's SkyOS, the global operating system for the entire vehicle, is a new system that is "AI-oriented" and brings about an order of magnitude increase in data volume, latency, computing power, cross-domain linkage, security, and reliability. It redesigns the architecture, interface, algorithm, and tool chain, decouples layers, and can perform multi-core processing and multi-layer scheduling, opening up all resources in the system.(Including vehicle control, smart driving, smart cabin, mobile phone, etc.), and make unified coordination and management.

The NioRadio podcast mentioned a useful analogy: the system software of NIO mobile phones is named SkyUI. It is a UI, not an OS, because the mobile phone system is customized based on Android.

SkyOS is equivalent to Android for cars, providing the underlying layer for different brands such as NIO, Ledao, and Firefly, and then the brands define and build the intelligent application systems on top. For example, NIO’s “Tree” system(Banyan, Aspen, Alder)It is said that Ledao will be named after "fruit", a kind of fruit that is abundant in Hainan, bears fruits on trees and can be eaten. It is estimated to be called Coconut.

Such basic technology occurs on the fifth floor underground, while the application experience that people can access is on the first floor. Without direct perception, it is difficult to understand.

Second, the narrative is highly professional, and some key indicators used to illustrate the awesomeness even require further explanation.

For example, in SkyOS, NIO designed a layer of SkyOS-H as a Hypervisor(Virtual Machine Monitor), used to manage and dispatch the heterogeneous large computing power of the whole vehicle. NIO's standard configuration is more than 1000T computing power, including four Orin X, 8155/8295, etc.; and then the self-developed Shenji chip has 50 billion transistors, which can replace four Orin X. In short, it is a lot. Cabin-driver integration is the mainstream trend in the industry. As the top computing power, NIO can release a much larger resource ceiling.

The role of Hypervisor is to support multiple operating systems running on the same hardware platform, such as infotainment and driver assistance on the car. There is a mature open source solution KVM, and there are also many embedded solutions based on KVM in China.

Compared with KVM, SkyOS-H reduces the latency of real-time tasks by 50%, increases multi-thread throughput by 40%, and increases virtual disk throughput by 15%. Latency is easy to understand; throughput is also critical. Without data accumulation, it is meaningless to increase processing power. NIO also made a comparison, and SkyOS-H's read and write performance increased by more than 20%.

These improvements are designed to provide a high-performance virtualized hardware resource pool for high-computing, low-latency applications, while also providing a secure and isolated operating environment for each system.

As an ordinary spectator, these numbers are very impressive. After all, the optimization of basic technology, its emotional part is the pursuit of several decimal places, and only professional expression can live up to the accuracy of the results.

Third, it still takes time. Even if you get on the bus, it does not mean that the miracle of generational difference in experience will appear immediately.

SkyOS and Shenji chips will be available in ET9 in the first quarter of next year. For example, there are still many difficulties from successful tape-out to mass production. Apart from this, Li Bin said that "it takes time to put new architecture and new chips into the car and experience it". Expectations must be well managed.

In other words, the advantage of these technologies is that they bring higher critical performance ceilings, but it will take some time to truly experience applications pushing their ceilings.

In general, the underlying, basic, and long-term factors are the reasons why technology conferences are becoming increasingly difficult to understand.

This is a good thing. It shows that the competition among the top technology players is heading into the deep waters, driving smart electric vehicles to a more fundamental and thorough AI transformation. From the bottom up, these smart electric vehicles have undergone a qualitative change from ordinary smart electric vehicles. They are not only competing for the position on the track, but also defining the direction of the track. This is the real right to speak in technology.

We all agree that China's smart electric vehicles are strong, but in many cases, they are very similar to the way we used to think that China's Internet was strong. Its prosperity is based on commercial application innovation based on market scale, not basic technology. This is definitely not just my narrow-mindedness.

At last night's AI smart driving technology conference, He Xiaopeng shared his concerns about visiting the US business in June:

“If you go to Silicon Valley, you will see American venture capital investing in chips, large models, and even robots; but when you return to China, you will see that Chinese venture capital is discussing how to make money with them. The key is to make money.

In the U.S,(Technology Company)We are all thinking about how to make the end-to-end stronger.AutopilotTechnology; but everyone in China is arguing about the weekly charts, what are our technology companies doing? "

I think this is definitely not a difference in the pattern, but rather a dilemma of being caught in the competition of low-quality volumes.

Although NIO is making a lot of noise, it does not prevent it from working hard to break through this predicament. This is why Li Bin's confidence is stronger than knowledge in NIO IN.

He said that NIO's foundation is technology and research and development.

He said that our R&D efficiency is very high, and those who say that NIO is inefficient are just trying to set the pace.

He said,(Basic R&D)Not every company is willing to do this, not every company has the ability to do this, and not every company can survive to see the day when results are achieved.

Of course, he also made a lot of "outrageous remarks" that attracted a lot of traffic. But I think these technical statements are more worthy of being recorded. Such tough remarks will become the new normal for NIO's technology to move forward. Last year, they decided to hold NIO IN with the goal of showing everyone the real NIO, a big company that is bad at technology.

What they actually do is much more substantial than what they boast and shout about - the key to getting rid of low-quality coils is never to call on low-quality coils to give way, or to kill the leaders of the coils. The coils will never stop, especially in a market with a structural imbalance between supply and demand, coils will always find a way to enslave people; the key is how to use high-quality coils to make low-quality coils lose their competitive advantage and eliminate them.

Returning to the essence, high-quality volume is still the capability that companies need to win long-term competition.

The importance of basic technology is becoming more prominent. In the past, there was a model that gave people the illusion that car companies could become assembly plants that only defined products. This model is completely unfeasible in the current high-intensity competition. Many companies that have died in the past two or three years held such rash ideas and ended up being unsustainable. Today, Huawei has become a company that cannot be avoided in discussions about cars, and its foundation lies in its basic research and development. This will be the trend of the future technology leaders. Leading is not for the sake of leading, but because it is necessary and must be leading.

First, it can make the user experience better and safer, which is always the core demand of automobiles and the core development goal of automobile companies. For example, the underlying operating system, SkyOS can bring the advantages of high bandwidth, low latency, large computing power and heterogeneous hardware, cross-domain integration, flexible and continuous evolution, high reliability, and information security.

For example, another highlight of SkyOS is its SkyOS-L module, which is the first domestic alternative to Autosar and commercialized on a large scale. This is the embedded software architecture originally adopted by the industry, which is aimed at parts manufacturers to align with the vehicle development standards, thereby realizing on-board electronic control. However, with the high degree of electronic and electrical integration of vehicles, the architecture has also shifted from distributed to central integration. The design of Autosar for bulk architecture and simple applications is somewhat outdated. Even the later developed Autosar-adaptive still lacks consideration for concurrency and high-bandwidth data processing. Therefore, SkyOS-L is not a replacement, but an upgrade replacement, and at the same time solves the problem of poor adaptability of domestic chips.

Every time a wheel is reinvented, there is a reason to start over.

Almost all domestic technology leaders recognize the importance of operating systems. For example, Huawei has iDVP, Xiaomi has HyperOS, and He Xiaopeng revealed that he is also working on OS, and said, "In the OS system, I think every company should do a good job in R&D like NIO."

Of course, it is not necessary to compare the advantages and disadvantages at this stage. Each strategy is somewhat different, depending on different architectural designs, but also on the company's genes and overall strategic thinking.

NIO is car-centric and gives more consideration to aspects such as stability, safety, big data throughput, bandwidth, and latency. It will also expand to panoramic interconnected terminals, such as NIO Phone, to connect AI capabilities and information flow.

Xiaopeng is said to have developed the car's OS and the robot's OS together, with the robot's frame rate reaching 1000Hz.

Some people may suggest that this is a waste of resources and the industry will one day be unified, just like when car companies, parts suppliers, semiconductor and software system companies jointly established Autosar in 2003. However, any car company that wants to survive until then and wants to have a say will not give up.

Second, technology can be exchanged for gross profit.

Basic research and development will certainly prolong the profit cycle. But when the cycle is too long, as long as the realization of communism, it will raise questions in business. For example, according to statistics, Weilai has invested more than 43 billion yuan in research and development. This number is often used to imply that Weilai is "spending money lavishly" and "inefficient."

However, Li Bin pointed out that 60%-70% are basic research and development costs, not for a specific model or brand, but can be shared by all brands and models. This is a feature of smart electric vehicles. The automotive industry also had platform-level technology in the past, but it was an order of magnitude lower than today's platform-level technology.

"When doing basic research and development, you have to do the math. There has to be a return. You can't do research and development for the sake of research and development," said Li Bin. "We calculated at the time that we would be worth it if we sold as many cars as possible." For example, SkyOS can become the foundation for all car models. The larger the amortization base, the more efficient the R&D investment. Even if the sales volume reaches 10 million units a year, the investment in basic research and development will not increase much.

NIO has always liked to calculate this way, converting time returns into the model.

An example we are more familiar with is batteries. Because of battery replacement, battery packs and cars have been completely decoupled and universalized. One or two generations of models can use only one specification of battery pack. Li Bin said that when the battery form and high-voltage platform are stabilized, four or five generations of models can use universal battery packs, just like everyone uses No. 5 and No. 7 batteries at home, and there will not be much change. This will actually save adaptation costs.

In the long run, SkyOS system plus tool chain will be very competitive in the automotive development system. After it is solid, it is not ruled out that it can be opened to suppliers and car companies for use to earn technology empowerment fees. It has the foundation for an open industry, but it is not Weilai's priority at the moment.

Of course, building such underlying platform capabilities requires very stringent conditions.

First, endure the shitty process.

The investment is indeed longer. For example, SkyOS started internal discussions in 2018 and was officially launched in 2020. A lot of investment has been made, and it is still increasing every month. Looking at the entire R&D, 3 billion yuan of R&D expenses are invested every quarter, more than 10 billion yuan a year, and it has been two consecutive years. Not many companies are willing to do this.

In addition to capital, one must also be mentally prepared to resist all kinds of mental attacks.

"We often use buildings as an example. It makes a big difference whether you lay the foundation for five floors or a hundred floors. We want to lay the foundation for a hundred floors."

Four years have passed since then, and the foundation has emerged from the ground, with more work being done on upper-level applications.

Of course, the point is, "As long as it's above the ground, each floor can be built very quickly."

Second, you need to have a clear enough foresight about the technology, which of course is very difficult - or you need to do what Li Bin did and invest enough redundancy in advance.

Li Bin said that in the past four years, many applications have been running ahead of the operating system, such as large computing power and sensors. The application ceiling is relatively high, which in turn will provide a good input to the operating system. Referring to the ET7 delivered in 2022, it is still at the computing power ceiling today, which brings possibilities for algorithms and data closed loops such as the world model.

"R&D can bring in gross profit, but the effect is slow. However, it is definitely the most important direction for increasing gross profit, rather than simply engaging in price wars."

In Li Bin's opinion, simply measuring R&D efficiency by current sales is too short-sighted. The NIO IN has a chip and soul, which is also intended to show that only by hard work and efficient execution can NIO qualify for the finals.

Of course, NIO also reflects on its sales volume.

Originally, Li Bin always believed that "making a car bigger is just a matter of adding ten centimeters and a thousand dollars; but how much does it cost to add a chip, which can only be useful in five or ten years." This is a chain of contempt.

But now Li Bin said that at the product definition level, some functions and experiences with high user perception and high value should be spent more energy. In the past, Weilai did not think comprehensively enough and did not have a deep understanding, so it began to rectify the situation - learning from excellent peers in visible areas. For example, Ledao L60 is mainly about listening to advice, with refrigerators, color TVs, large sofas, safe, comfortable and large spaces, all arranged.

“This does not conflict with investment in basic technology.”

"We are running a marathon, and we still have a long way to go," said Li Bin. "Even in the matter of battery swapping today, I dare not say that we have a moat. The entire industry still has a long way to go in terms of competition. We can only go all out and resolutely do it according to the logic and model we understand. That's the only way."

It's like climbing a mountain.

When the mountain was low, there were a lot of people, and they were only a few steps apart; as they walked, most people fell further and further behind, and the higher they climbed, only a few were left, all of whom were originally carrying heavy loads and huge equipment.