In a head-on competition with Qualcomm, what kind of cockpit products has Intel come up with?

"Intel is bringing AI PC to cars". During this year's US International Consumer Electronics Show, Intel announced its official entry into the automotive market and released its "first generation of software-defined automotive SoC" chip products, with Zeekr as its first customer. Now, it has released new results.

On August 8, Intel released the second product of its new automotive product line, the first generation of Intel ARC A760-A, an independent graphics card for automotive use . According to Jack Weast, vice president of Intel, the computing power of this graphics card platform is 229TOPS, and it is equipped with 16G of DDR 6 video memory. The floating point computing power per second is 4 times higher than that of integrated graphics cards . It supports 8 independent displays and 4K at the same time, and is equipped with Xe graphics engine and XMX computing engine.

"We are bringing discrete graphics cards into the automotive industry , so that large language models can be run in the car, and the same AAA PC gaming experience can be experienced in the car at home, while the chip can be used locally to get a computer-like experience." Jack Weast said, "Graphics cards will be mass-produced in 2025, which will usher in a new era of personalized customization and intelligent interaction in car cockpits."

As the dominant player in the PC stage, Intel does not have the upper hand in the field of smart cockpit chips. According to data from Gasgoo Automotive Research Institute, Intel will account for only 4.6% of the Chinese market in 2023, ranking fourth after Qualcomm , AMD , and Renesas . Obviously, Intel wants to change this situation, and it believes that it has a great opportunity.

01. Intel has a chance to make another effort in the automotive industry

This is not the first time Intel has made efforts in the automotive field. As early as 2016, Intel launched the Apollo Lake series of products, which quickly occupied the high-end car cockpit market. At that time, except for Mercedes-Benz and Audi, almost all high-end cockpit chips chose Apollo Lake. However, because the automotive market planning was too small at the time, Intel did not focus its main energy on this field and did not upgrade its products until 2023. This gave Qualcomm an opportunity.

Around 2020, Qualcomm's third-generation smart cockpit chip Snapdragon SA8155P was released, gradually replacing Apollo Lake and almost monopolizing China's mid-to-high-end market. The fourth-generation smart cockpit chip Snapdragon SA8295P, which was subsequently released in 2021, has also become the first choice for most domestic flagship models.

According to data from Gasgoo Automotive Research Institute, in the Chinese market in 2023, Qualcomm's cockpit SoC chip shipments reached 2.26 million, accounting for 59.2%; AMD ranked second, with a market share of 15.1%; Renesas ranked third, accounting for 8.6%; Intel ranked fourth, accounting for only 4.6%.

Intel obviously hopes to break the existing pattern. On January 9, 2024, Intel announced at CES 2024 that it plans to push the company's "AI everywhere" strategy into the automotive market, and also released a new series of artificial intelligence enhanced software-defined car system-on-chips . Zeekr will be the first "early adopter" and will be installed in 2024.

At that time, Jack Weast said, "Frankly, today's smart cars remind us of the first generation of laptops . In 2035, 80% of cars will be software-defined or electric . Now is the best time to enter the market." At the press conference, Jack Weast further updated the Chinese data. He said that by 2025, China will have an 80% penetration rate of smart cockpits, and many smart cockpit solutions will feature AI.

Intel's launch of ARC A760-A at this time is obviously just right for them. In fact, there is indeed an opportunity. At present, the standard configuration of the main control chip of the car's smart cockpit is SoC system-level chip + MCU chip. Among them, SoC chip is an important growth point in the market, Qualcomm occupies a dominant position, and the mainstream framework adopted is "ARM+Android".

With the rise of smart cars, users have a growing demand for differentiated cockpits such as 3D HMI, AI large models , and large-scale games. The "x86+ Linux " architecture has attracted the attention of car companies. Compared with the ARM architecture , the CPU under the x86 architecture has a large demand for computing, and its advantage lies in running large software. One of the leaders of the latter is Intel.

02. High computing power, Intel’s key weapon

As a "latecomer", Intel wants to take a piece of the pie from Qualcomm, so it naturally has a magic weapon: high computing power . According to Jack Weast, Intel ARC A760-A is an automotive-grade independent graphics card. In terms of AI capabilities, it can provide a platform computing power of 229 TOPS.

In comparison, the AI ​​computing power of Qualcomm's smart cockpit chip Snapdragon 8295 is 30 TOPS. In other words, the AI ​​computing power of Intel ARC A760-A is nearly 7 times higher than that of Qualcomm Snapdragon 8295. Another important indicator for measuring the computing power of deep learning , the FP32 single-precision computing power is 14TFLOPS, which is comparable to the mainstream GPU A100 used for large model training.

In addition, the Intel ARC A760-A has 28 Xe cores, 28 ray tracing units, and up to 448 XMX/vector engines for AI. At the same time, it inherits 16 GB of video memory and supports eight independent 4K displays running simultaneously. Jack Weast said that with its graphics processing capabilities, Intel's on-board discrete graphics card can support high-fidelity visual effects and complex 3D human-computer interfaces.

The reason why such a high-power AI product is launched in the cockpit field is based on Intel's judgment of the future trend of the smart cockpit market. Although the biggest brain of AI is in the cloud, Gao Yu, general manager of Intel China's technology department, said that rich and innovative AI application scenarios require a powerful local computing platform as support.

"We emphasize the need to embed AI computing power in the cockpit rather than just rely on the cloud. We are not denying the cloud, but we hope to see a reasonable distribution of computing power between the cloud and the end."

Jack Weast said that there are five considerations for placing computing power on the end.

First, local computing power is no longer dependent on the network;

Second, local computing power can ensure extremely low latency;

Third, for cost considerations, deploying computing power locally will reduce end-to-end costs;

Fourth, protect privacy and security. With the widespread deployment of cameras and sensors inside and outside the car , personal information security issues are becoming increasingly prominent. Local storage can effectively avoid such risks.

Finally, the memory capabilities of large models, including short-term memory, long-term memory, and dynamic fine-tuning capabilities, can be applied to achieve more humane deployment.

"For these five reasons, we strongly recommend that you consider sufficient local AI computing power when planning future models." Gao Yu called on car companies to pre-set AI computing power to meet the explosion of AI application scenarios in the cockpit within the next two years. Gao Yu said that to achieve a very good AI travel companion, a larger model is needed, and a model of 6B, 7B or even 10B or more is required to achieve better results.

"Thanks to the high computing power and large-capacity video memory, we can carry larger models. Judging from the current optimization results, it directly covers a wide range of models from 6B, 7B, 10B, 13B and even 18B." In Gao Yu's view, Intel ARC A760-A is currently the only choice that can run such large-scale models in the cockpit.

03. Intel combines software and hardware to make an all-out attack

"By 2030, the semiconductor market will undergo tremendous changes and the market size will reach $1 trillion." Jack Weast believes that Intel is the only company that can participate 100% in it because Intel has foundries that can manufacture artificial intelligence products and complete delivery.

At the same time, Intel is adopting open standards, namely open source software. In Jack Weast's view, the current industry is full of challenges, and Intel hopes to help the industry solve three challenges.

The first is the architectural challenge. With the help of software definition to solve the electronic/electrical architectural challenges, various industries, including the telecommunications industry, are transforming from the mindset of embedded microcontrollers to software definition.

The second is the energy challenge. Although the electric vehicle industry is developing very well, it is not easy to make money. Intel hopes to help the industry make profits by improving vehicle energy efficiency and reducing costs.

The third is the power and performance challenge. Many manufacturers are now pursuing scalability and hope that software can be reused. Intel hopes to use chips to help partners better realize expansion possibilities.

Based on this, Intel will include three solutions in the field of smart cockpit chips. For software-defined cockpits, Intel will bring PC-level processors , including PC-level CPUs and GPUs, to cars, and design cars as a system, which can improve efficiency and bring lower costs.