If something happens to Nvidia Orin, who can replace the domestically produced autonomous driving chip?

Is there a risk of autonomous driving being "cut off from under the cauldron"?

"Cauldron" refers to the iteration of AI technology and the evolution of smart car capabilities.

"Salary" refers to the most basic high-performance AI chip at the bottom layer.

To put it more bluntly, it is Orin, which is currently dominating the autonomous driving chip market by Nvidia.

After the changes in high-end GPUs, "who can replace Nvidia Orin" has become an issue that must be paid attention to at the moment.

Nvidia Orin, is it really that critical?

At present, NVIDIA Orin is indeed unique in terms of technological advancement, performance indicators, and mass production and delivery capabilities.

In terms of performance, Orin uses a 7nm process and consists of an Ampere architecture GPU, an ARM Hercules CPU, a second-generation deep learning accelerator DLA, a second-generation visual accelerator PVA, a video codec, and a wide dynamic range ISP.

At the same time, the automotive-grade Safety Island design was introduced.

Orin supports 204GB/s memory bandwidth and up to 64GB of DRAM. The high-speed I/O interface is compatible with the interface of the previous generation Xavier SoC. It can achieve 275TOPS of INT8 computing power, which is 7 times that of Xavier, and consumes 55W of power.

From a technical point of view, the biggest difference between the hardware configuration used by Orin and its predecessor is the introduction of Tensor Core, which supports sparse computing (170 TOPS), a fine-grained computing structure that can double the throughput and reduce memory usage.

In addition, indicators such as deep learning acceleration, memory and communication, and CPU performance are all 1-2 times better than the previous generation.

In terms of software, NVIDIA provides a software development kit SDK (Software Development Kit).

It is mainly the board support package (BSP), which includes the bootloader Bootloader, Linux kernel, driver Driver, tool chain Tool chain and Ubuntu-based reference file system. BSP also supports various security features (secure boot, trusted execution environment, disk and memory encryption, etc.).

On top of BSP, there are multiple user-level libraries for accelerating applications, including deep learning acceleration libraries (CUDA, CuDNN, Tensor RT), accelerated computing libraries (cuBLAS, cuFTT), computer vision and image processing libraries (VPI), multimedia and camera libraries (libArgus and v4l2).

Companies with the ability to conduct independent research and development may not necessarily use the development tools provided by NVIDIA, but this illustrates the highly open and customizable nature of NVIDIA Orin.

In terms of mass production and delivery, NVIDIA Orin has begun stable delivery to OEMs and autonomous driving companies.

Currently on the market, models equipped with Orin chips include NIO ET7 and ES7 (equipped with 4 NVIDIA Orin chips), and will subsequently include almost all models aimed at advanced assisted driving.

For example, Xiaopeng G9, Jidu, WM Motor, Zhiji, Ideal L9, etc.

Autonomous driving companies, especially those focusing on L4, almost cannot avoid the underlying computing platform supported by NVIDIA Orin, whether it is testing and development or commercial implementation.

Therefore, the importance of NVIDIA Orin is almost irreplaceable at present: among all mass-produced autonomous driving chips, Orin has the highest single-chip computing power, the most advanced technology, and the fastest mass production pace.

The strongest on the planet, car companies are scrambling to buy it.

What if something happens to Orin?

Who can replace Nvidia Orin?

Nvidia dominates the autonomous driving chip market, in addition to its deep technical strength in the GPU field, another major reason is that it entered the market early.

In 2015, NVIDIA launched its first chip PX for autonomous driving.

But latecomers are not without opportunities. In fact, some options to replace Nvidia have gradually surfaced.

Possible alternative, but not entirely possible

The main references here are to two other foreign manufacturers: Qualcomm and Mobileye.

Among them, Mobileye is a veteran in autonomous driving. Its integrated hardware and software autonomous driving system was once the only choice for mass production of passenger cars.

The EyeQ 5 chip, which is currently widely mass-produced, has a computing power of 24TOPS, which is at the same level as Nvidia's previous generation Xavier.

The largest user in China is Geely Auto, and almost all of its models with intelligent assisted driving functions use Mobileye solutions.

But this is also the limitation of Mobileye. Its "black box" model that does not open data permissions to OEMs, and its "bundled sales" that require both software and hardware to be purchased as a set, make its path narrower and narrower.

It has been abandoned one after another by manufacturers such as NIO and Ideal.

Although the next-generation EyeQ 6 chip is expected to catch up with or even surpass Nvidia in terms of technical indicators, the Mobileye solution can only be said to be an alternative but not the best choice.

Qualcomm, which has already dominated the smart cockpit chip market, has come up with products that surpass Nvidia in the field of autonomous driving.

SnapDragon Ride chip, 7nm process, 360TOPS computing power at INT8 precision, and overall power consumption of 65w.

Its performance exceeds that of NVIDIA Orin, and it has been mass-produced and put into use in vehicles.

The latest version of Great Wall's Wei-brand Mocha DHT-PHEV is the first mass-produced Qualcomm SnapDragon Ride and will be delivered by the end of the year. It is officially claimed that it can achieve urban pilot assistance functions that exceed ordinary L2.

Qualcomm is certainly an alternative to Nvidia, but as a foreign manufacturer, Qualcomm faces the same risks as Nvidia.

Who can choose autonomous replacement?

The loudest voice comes from the domestic manufacturer Horizon and its upcoming mass-production Journey 5.

Horizon Journey 5 is built based on TSMC's 16nm process, and its AI computing power can reach 128TOPS.

In terms of core architecture, the CPU part of the Horizon Journey 5 chip uses an 8-core ARM Cortex A55, and the AI ​​computing unit uses a dual-core Horizon Bayesian architecture BPU (Brain Processor Unit).

At the same time, the Journey 5 chip also has 2 ISP cores, a computer vision engine, 2 DSP cores, and a video encoding and decoding unit.

In terms of mass production progress, Journey 5 has been delivered to the OEM for development and testing, and the official mass production time is set for 2023.

Beyond the horizon, Huawei is another important player.

MDC 810, with a computing power of 400TOPS, has been put into mass production. MDC 810 is not equipped with a GPU that supports general computing, but uses the "domain-specific architecture" AI chip Ascend for computing.

The BAIC Arcfox αS Hi version, Changan Avita 11, and GAC's upcoming new car will all be equipped with Huawei MDC 810.

The other two domestic autonomous driving chips that have the potential for mass production are Huashan No. 2 and Jiazhixin V9.

They are from domestic startups Black Sesame Intelligence and CoreChi Technology respectively.

Heizhima's Huashan 2 A1000 is already on its way to mass production. With INT8 precision, the computing power of a single chip reaches 58TOPS. It will be mass-produced for the first time on JAC Group's new Sihao model, with the specific time yet to be determined.

Xinchi Technology has set its mass production target for autonomous driving chips directly on the future stage of high-level intelligent driving.

In the second half of this year, Xinchi will launch a dedicated chip for autonomous driving with a computing power of more than 200TOPS.

According to the general rules of the semiconductor industry, mass production plans will not be started before 2024.

Pros and Cons of Alternatives

According to CITIC Securities' latest "Autonomous Driving Chip Industry Research Report (2022)" analysis, these different players' solutions that can replace Nvidia Orin have obvious advantages and disadvantages in terms of performance and mass production rhythm.

Horizon Quest 5:

The biggest advantages are localized R&D and services, as well as the most advanced performance parameters in the country.

The disadvantages are the same: still not as good as Nvidia Orin.

In its research report, CITIC believes that the maximum computing power of Heizhima Intelligent A1000 is still behind that of Nvidia. In terms of computing platform solution, the computing power of about 250T similar to that of Nvidia single board can be achieved through four A1000s, which is also a generation gap.

But the advantage is also that mass production is imminent. If all goes well, JAC's new models will be launched and delivered at the end of this year.

As for Huawei, its real advantage lies not in its computing power or how fast its mass production is, but in its system engineering capabilities for smart cars.

Including chips, algorithms, cloud, V2X, operating systems and other levels.

This is why Huawei does not use general-purpose GPUs in the MDC solution, but instead chooses the Ascend chip, which is closely linked to Huawei's overall strategic layout.

CITIC Securities believes that Huawei has the ability to integrate the resources of major giants in the research and development of smart cars and autonomous driving, greatly accelerating the pace of real smart cars being put into production, and perhaps soon over the horizon.

But Huawei's cause for concern is equally clear: geopolitical friction.

As for CoreDrive's V9, we will have to wait and see its performance and commercialization progress after its official launch in the second half of this year before we can draw a conclusion.

So to sum up, it is not realistic to require domestic manufacturers to immediately come up with autonomous driving chips with performance similar to Nvidia Orin.