Intel is anxious to compete with Qualcomm for the automotive industry
Author: Liu Junhong
This article is reproduced from WeChat official account
| Lightcone Intelligence
The lesson of rejecting the iPhone and missing out on an entire era of mobile Internet is too painful. In the era of intelligent cars, Intel knows that it cannot be absent this time.
Recently, Intel released its second product in the automotive market - the ARC A760-A discrete graphics card. This product launch is Intel's second move to enter the automotive industry after it announced its entry into the automotive market and released the automotive cockpit SoC at this year's CES.
According to Jack Weast, vice president of Intel, the positioning of this in-vehicle graphics card is mainly used to meet the display and large model computing power requirements of the car cockpit. In terms of specific parameters, this graphics card is equipped with 229TOPS computing power and 16G DDR6 video memory, which can support the 14B parameter of the car side large model of the OEM.
In terms of configuration, the computing power provided by Intel is quite high, basically "excessively" covering the current needs of automotive smart cabins.
At the same time, the ARC series, as Intel's main PC graphics card brand in recent years, can also support 8 independent displays to display 4K content at the same time on the car side. Jack even claims that it can "play the top computer games in the car."
Perhaps it is a continuation of the idea of graphics cards moving from integrated motherboards (or SoCs) to independence in the PC era. Intel's automotive graphics cards this time, starting from the graphics and AI needs of the car cockpit, try to cut into the "gap" between Qualcomm's integrated cockpit SoC and Nvidia's autonomous driving chips. And then undertake the needs of the increasingly popular in-vehicle AI big models of current smart car manufacturers, and the possible future car-side intelligent driving big models.
Behind the redundant computing power and the flexing of muscles is Intel's eagerness to enter the smart car market, hoping to work miracles.
This graphics card, which is expected to be deployed in cars in 2025, has extraordinary significance for Intel. After losing miserably to Nvidia in the AI chip business and lagging behind Qualcomm in the mobile business, Intel hopes to fight again in smart cars.
"Intel had a five-year gap, but now we will take back everything we lost." After the press conference, Li Zhe, Intel's automotive business sales director, set this goal.
For Intel, which is currently facing declining profit margins, plummeting market value, and the "explosion" of its new-generation CPU process, seizing the automotive chip track has become the "last straw."
Defining smart cars from a PC perspective
Intel provides super-large computing power redundancy
In Intel's view, smart cars may be closer to PCs.
The most obvious feature is reflected in the specifications of this graphics card. According to the specifications, Intel's automotive graphics card is basically a "car-grade modification" of the existing desktop graphics card. In addition to meeting the necessary capabilities of wide temperature difference, reliability and stability for automotive use scenarios, it also has the PC application scenario capabilities of supporting multiple drivers, media decoding formats and virtualization (SR-IOV).
Image: ARC A760-A parameters (Source: techpowerup)
This model is mainly the result of Intel's operations based on existing experience. On the one hand, in terms of chip design, Intel referred to the idea of reusing the Snapdragon 888 design solution of Qualcomm's 8295 chip, and also "modified" the existing graphics card series to create this automotive graphics card. On the other hand, Intel, with the help of its past experience in PCs and data centers, further proposed the idea that automotive intelligence should be closer to hardware such as PCs.
"There are great similarities between PCs, data centers and today's cars. From a software architecture perspective, the underlying system level of a car is more like a data center. It is necessary to consider various aspects including virtualization, different domains and security. But for example, the upper-level experience of AI applications is very similar to that of PCs. Some of the traditional legacy applications in cars are based on mobile devices," explained Li Ying, vice president of Intel Corporation and general manager of Intel China Software and Advanced Technology Group.
Different from the industry consensus that "cars are more like mobile phones", Intel wants to continue the transformation of AI on PCs in cars. "Thanks to the accumulation of AI PCs, the in-vehicle platform has also achieved rapid adaptation to China's local large models," said Gao Yu, general manager of Intel China Technology Department.
In terms of specific AI capabilities, Intel's automotive graphics card has a computing power far exceeding Qualcomm's 8295 chip by 30TOPS, with a computing power of up to 229TOPS, supporting large models with more than 14B parameters on the end side. Based on the abundant computing power supply, Intel's automotive graphics card can also run multiple "smaller" large models at the same time. A series of application scenarios such as travel assistants, health butlers, and business secretaries are "transformed" with AI large models.
However, Intel's vision may still be a bit "ahead of its time".
Judging from the current smart cabin designs of leading smart car manufacturers, AI applications in cockpits are far from reaching the stage where high computing power is required. Although AI assistants are generally used in smart cars, referring to the recent updates of Weilai (Banyan), Xiaopeng (Dimensity), Great Wall (Coffee OS), Hongmeng Zhixing (Hongmeng OS), etc., the current focus of car smart cabins is still on interaction logic, functional design, and car machine jamming. There is a strong demand for display in the cockpit, and car companies are temporarily inclined to use dual Qualcomm 8295 chips similar to Zeekr 009, or Huawei's "external" tablets in the back row to "curve" solve the problem.
If we consider the camera input capability supported by Arc A760-A and the possibility of Intel's involvement in cockpit integration, then the parameters of this automotive graphics card are a bit embarrassing.
Although the graphics card memory bandwidth, which is the "bottleneck" of the intelligent driving large model, is not less than 204.8GB/s of Nvidia Orin X in the Arc A760-A (reference A770 is 560GB/s, actual not specified), but referring to the actual computing power, Intel's automotive graphics card is less than Nvidia Orin X computing power, and far less than Nvidia's next-generation intelligent driving chip's 2000TOPS computing power (Thor).
Smart cars in the future will definitely have greater demands for cockpit computing power, but with several current mainstream car cockpit chips, Intel is not sure whether it can control the cost with so much computing power redundancy. This may be the biggest concern of car OEMs who are very sensitive to price.
Facing defeat
Is the car the last straw?
Intel, which was once in its heyday in the PC era, is now in a slump. Developing smart car chips may be the most correct choice for Intel in the past decade.
Intel, which was once in its heyday in the PC era, is now in a slump. Developing smart car chips may be the most correct choice for Intel in the past decade.
Intel's current crisis has been reflected in the "sluggish" second quarter 2024 financial report released in August. The financial report shows that Intel's total quarterly revenue was US$12.83 billion, a year-on-year decline of 0.9%, and its net loss expanded from US$381 million in the previous quarter to US$1.61 billion.
The poor performance of financial data is because Intel has failed to reap the "dividend of the times" in the AI era.
Since Intel's revenue is mainly composed of client business, data center and AI business (accounting for more than 80% in total), by looking at the growth of its main business, we can find that Intel is "lagging behind in all aspects" in the AI era.
Among them, the client business (mainly PC) was the only growth item of the company in the second quarter of 2024, achieving revenue of US$7.41 billion, a year-on-year increase of 9.3%. However, considering the trend of AI-driven PC generational upgrades in 2024, Intel did not obtain excess returns.
On the surface, in the recovery trend of global PC shipments, Intel has "kept up" with the industry's 5.4% shipment growth with a revenue growth rate of 9.3%. Compared with AMD, another player in the PC CPU market, although its client business is still relatively small, according to its second quarter 2024 financial report, AMD is "eating away" Intel's client business growth space with revenue of US$1.492 billion, corresponding to a year-on-year growth rate of 49.5%.
On the other hand, in the data center and AI business (excluding the impact of Intel's spin-off of the Altera business from this business in the first quarter of 2024), Intel's actual revenue this quarter was about US$3 billion, a year-on-year decline of 3%. It can almost be said that in an era where AI computing power is in short supply, it has walked out of an "independent performance curve."
Although the new demands for data centers (or intelligent computing centers) in the AI era will also generate demand for CPUs, most of this demand has been packaged in the product portfolios of manufacturers such as Nvidia and AMD. In other words, in the AI computing era, Intel is basically still a "CPU specialty store."
NVIDIA computing power solution configuration list (Source: NVIDIA)
Intel's current decline is not just the result of the CPU being left far behind by the GPU in the AI era, but the result of the company's continuous wrong decisions.
In 2006, before the global smartphone boom, Intel rejected the opportunity to provide chips for the iPhone. Having missed out on the iPhone, Intel was "absent" from the entire mobile terminal era. It also laid the groundwork for Apple to completely "part ways" with Intel on the Mac side after developing a whole set of chips and tool chains on its own.
When AI was just beginning to demonstrate its practical value (2017-2018), Intel rejected the opportunity to acquire a stake in OpenAI. The reason was simply that generative AI could not enter the market in the short term and investment returns could not be realized quickly.
The opportunity Intel is eyeing is the IDM2.0 plan implemented on the eve of the AI boom (2021-present). In simple terms, the IDM2.0 plan means that Intel insists on designing and producing its own chips while also outsourcing chips to others and letting others outsource chips to itself. In response to this plan, Intel began the long and asset-heavy process of building wafer fabs and outsourcing some of its cutting-edge products to TSMC.
However, as Intel gradually fell behind in the AI era, its wafer fab construction and advanced process foundry eventually became a "knife stabbing itself". Intel gave guidance for the third quarter of 2024 that was far below market expectations, due to the "swallowed" net profit and process failures of the 13th and 14th generation chips.
On this basis, Intel's direct autonomous driving subsidiary Mobileye has been gradually "eliminated" by players such as Ideal Auto and Zeekr Auto in the field of intelligent driving in China, and the market share of intelligent driving solutions has been caught up by Horizon. Personally participating in the layout of smart cars has almost become an option that Intel "cannot refuse".
Market share of intelligent driving computing solutions for independent brand passenger cars in the Chinese market in H1 2024 (Source: Gaogong Intelligent Automobile)
From the internal competition of China's smart cars, Intel has seen the importance of maintaining technological leadership and forward-looking research and development. "We in the United States may not have seen this change yet, but it is happening in other countries around the world. New electric vehicle forces from China and Southeast Asia are building cars around batteries and computers," Jack said at this year's CES.
Perhaps, with the trend of automobile intelligence, Intel can also seize the "east wind" of the AI era. But at the same time, with the entry of Intel, the market structure of automobile chips has also entered the "Spring and Autumn and Warring States" era.
The Spring and Autumn Period and the Warring States Period of Automotive Chips
In fact, the market for automotive cockpit chips has become fiercely competitive.
Due to the demand for intelligent vehicle upgrades, the current shipments of traditional automotive cockpit chip manufacturers (such as NXP) are mainly based on existing projects, and their market share is declining year by year. The overall market is in the process of switching generations of smart cabin chips.
According to the list of "Top 10 Passenger Car Cabin Chip Delivery in China in the First Half of 2024" released by Gaogong Intelligent Automobile Research Institute, Qualcomm ranks second with a market share of 23.72%. At the same time, although MediaTek, NVIDIA, Intel, AMD, HiSilicon and other consumer chip manufacturers focusing on entertainment and high-computing domain control currently have a small market share, they are all participating in the process of dividing the entire market with their own competition models.
Ranking of passenger car cockpit chip delivery volume in the Chinese market in the first half of 2024
(Source: Gaogong Intelligent Automobile)
Among them, MediaTek, with its low-price strategy of $30 cheaper chips, has started a war between Dimensity and Snapdragon in the smart car market. After Nvidia won big in the smart driving chip market, it has deepened its influence in the field of car cockpits with its AI chip + CUDA smart car ecosystem architecture and the upcoming Thor chip that integrates cockpit and driving. As for HiSilicon, it has basically benefited from the Hongmeng Intelligent Driving, which was jointly launched by Huawei and car manufacturers.
However, in addition to the "civil war" among chip manufacturers, vehicle manufacturers and Tier 1 automakers are also trying to "get involved". The main reason is that the overall smart cabin solution provided by chip manufacturers is "slightly expensive".
Taking Qualcomm's cooperation model as an example, in terms of chip prices alone, the price of Qualcomm's 8155 chip is about US$180, while the 8295 is about US$300. At the same time, in order to make good use of this chip, the vehicle manufacturer must also bear the development licensing fee, project service fee, and team costs for developing chips and applications. After weighing it all, it is no wonder that some vehicle manufacturers have the idea of "getting" chips themselves. For example, Ecarx, a subsidiary of Geely Automobile, found Core Engine Technology to make a cabin-driving-parking integrated chip. Similarly, with the recent successful tape-out of NIO's smart driving chip, it is expected that the release of its smart cabin chip will not be too far away.
The "chaos" in the automotive cockpit market means that Intel is facing pressure from the entire industry chain. In addition to competition from chip peers and the "criticism" of service capabilities by OEMs, Intel also needs to "strengthen its momentum" with third-party partners. However, for Intel, whose main partners at this conference are only SenseTime (mainly DMS and OMS systems), Zhipu (AI large models), and Thundersoft (car display), this effort obviously still has a long way to go.
However, as AI changes all smart terminals, Intel has undoubtedly grasped the industry trend by using computing power and AI capabilities as the entry point for car cockpits. After all, no player can give a suitable answer to the question of how car cockpit AI can demonstrate differentiated capabilities. Cockpit AI, which is mostly assistant-based, is far less in-depth than the end-to-end large model of AI in intelligent driving.
Perhaps, the relatively blank car cockpit of AI application just gives Intel the opportunity to stand on the same starting line with other competitors. But in any case, entering the smart car chip market is one of the few opportunities for Intel at present.
京公网安备 11010802033920号