NVIDIA's DRIVE Thor chip wants to be a key part of the "grand unification" of automotive intelligence.

Electronic Engineering Album News 2024, new energy vehicles into a new round of price war, in this contest, the industry is accelerating the shuffle has been recognized as a fact. At the 2024 GTC conference, NVIDIA said that the launch of DRIVE Thor will completely change the landscape of the automotive industry. In this track, what will NVIDIA change?

The industry has a point of view is that the car "MCU + SoC" will be just a transition program, the past car does not have a few chips, can be counted out, such as the engine control unit ECU and transmission control unit TCU, etc., into the new energy era, intelligent driving and intelligent cockpit internal functions need to be increased, the car companies put forward the "domain control". The concept of "domain control" is proposed by automobile companies. With the improvement of arithmetic power, the previously adopted concept of "domain control" will go to the central processing route.

NVIDIA proposes the "Generative AI Defined Driving Experience Era", which introduces the transformative power of generative AI, Large Language Model (LLM) and Visual Language Model (VLM) into the field of mobile mobility, and through the powerful sensing, understanding and generating capabilities provided by deep learning and Transformer models, it can build in-vehicle assistants to enhance the driver's driving experience. The powerful perception, understanding and generation capabilities provided by deep learning and Transformer models are used to build in-vehicle assistants that enhance driver safety, security and comfort.

DRIVE Thor was born with NVIDIA's Blackwell architecture built specifically for Transformer, Large Language Model (LLM) and generative AI workloads.

The successor to DRIVE Orin, DRIVE Thor is a "centralized in-vehicle computing platform" that combines the Smart Cockpit chip and Autopilot chip, along with all of their functionality, into one centralized platform, which is expected to be in mass production as early as next year.

NVIDIA's newly released Blackwell architecture chip has a maximum of 208 billion transistors, and the number of transistors on a single die on the Blackwell chip reaches 104 billion, so two dies is 208 billion, 128 billion more transistors compared to the Hopper; Blackwell is manufactured using the customized Double Lithography Limit process. Size 4NP TSMC process manufacturing, through the 10 TB/s inter-chip interconnect, the GPU die connected to a unified GPU. mentioned in the official announcement, Blackwell's six key technologies also include the second-generation Transformer engine, the fifth-generation NVLink, the RAS engine, the security AI, the decompression engine.

DRIVE Thor was mentioned as early as 2022, and this year the first five car companies to work with DRIVE Thor were announced, with BYD, HaoPlatinum, Xiaopeng, Ideal, and Pole Krypton.

Hao Platinum's new model of next-generation electric vehicles using the DRIVE Thor platform will begin mass production in 2025, enabling L4-level autonomous driving, while Xiaopeng is utilizing the DRIVE Thor platform to power its self-developed XNGP Intelligent Assisted Driving System, which implements autonomous driving and parking, driver and passenger monitoring, and other functions.

BYD's long-term collaboration with NVIDIA extends from the vehicle end to the cloud. In addition to using DRIVE Thor to build its next-generation electric models, it plans to use NVIDIA's AI infrastructure for cloud-based AI development and training technologies, and to use the NVIDIA Isaac and NVIDIA Omniverse platforms to develop tools and applications for virtual factory planning and retail configurators. In addition, BYD did not specify which new BYD vehicles the DRIVE Thor will be equipped in, and it is speculated that BYD should only use this Thor chip in those models that require high computing power.

Automotive Intelligence "Grand Unification" is Coming?

DRIVE Thor is called an "in-vehicle computing platform", which is not just a chip, but can be understood as a set of small central computers consisting of a motherboard, CPU, GPU, RAM and memory, and its configuration can be specified by the customer, which is the meaning of Thor being defined as a computing platform. Manufacturers can set up DRIVE Thor in a variety of ways, and they can use all of the platform's computing all for autonomous driving processes, or use part of it for the smart cockpit and another part for driver assistance.

A few years ago, the new forces of new energy vehicles ushered in a wave of boom in the market, with a surge in sales and scenery. Until now, the mainstream vehicle chips mainly include control chip MCU, intelligent cockpit chip, and automatic driving chip. MCU is responsible for managing and controlling various electronic systems inside the car; intelligent cockpit chip is designed for the human-machine interaction system inside the vehicle, while supporting in-vehicle entertainment, navigation, voice recognition, intelligent assistants, etc.; the automatic driving chip is simply a high-computing chip that meets the needs of the development of intelligent vehicles. Automated driving chip is simply a kind of high computing power chip to meet the needs of the development of intelligent vehicles.

The automotive integrated chip integrates all functions into a centralized platform, which may improve the development efficiency of car companies and accelerate the speed of repeated software calculations.

Nowadays, "price war" has been regarded as a way and means of market competition, but the car itself has a high cost, this competition will continue, but not lasting.

Whether it is traditional or emerging forces car companies, are trying to AI deeper integration into the car application. With AI in-vehicle smart chips further complementing the intelligent functions of traditional car companies, this track will have more and more competitors, but ultimately will also be more intense.NVIDIA will undoubtedly become a key link in this competition.

Like the latest generation of NVIDIA DRIVE Orin, DRIVE Thor utilizes the productivity of NVIDIA's DRIVE software development toolset with ASIL-D functional safety level design and is based on an expandable architecture that allows developers to smoothly transfer past software development to the latest platform.

DRIVE Thor also dramatically improves the accuracy of deep neural networks and is the first in a family of self-driving car platforms to incorporate the Transformer engine.

"Accelerated computing is delivering transformative breakthroughs in areas such as generative AI, which is redefining autonomous driving and the global transportation industry," said Xinzhou Wu, global vice president and head of the Automotive Business Unit at NVIDIA."

What contributes to the current NVIDIA?

Looking back at NVIDIA's layout in the field of intelligent driving, NVIDIA launched 1TOPS PARKER in 2018, 30TOPS Xavier in 2022, and 254TOPS Orin in the same year, and Orin's chip has almost seized the entire high-end electric vehicle market.NVIDIA has only used less than five years to change from an "outsider" to an industry giant. NVIDIA in less than five years, from an "outsider" to an industry giant.

What are some of the reasons that led to NVIDIA?

In 2018, we are still looking at who has installed the Mobileye Q4 with 2.5TOPS arithmetic power, and by 2022, 200TOPS arithmetic power can no longer satisfy high-end cars.

With the rise of the new energy vehicle market, on the one hand, there is an absolute immediate need for high arithmetic power for the massive data of intelligent driving, and on the other hand, the head of the car enterprises need an open platform to realize software self-research, and there is an urgent need to realize the self-research of automatic driving algorithms.NVIDIA was the intelligent driving chip supplier at that time that could satisfy these two requirements at the same time, and the new intelligent driving chip DRIVE Thor released this time is able to DRIVE Thor, the newly released intelligent driving chip, can integrate intelligent driving with active safety, automatic parking, occupant monitoring, electronic mirrors, in-vehicle operating system, instrumentation, etc., and integrate intelligent driving and intelligent cockpit two domains of control, which can be seen to want to realize the central computing and electrical architecture of the car "grand unification".

The central computing architecture of the car is a big trend, just like a person only has a brain to manage all things, but before that, it needs to be able to undertake the task of the brain of the emergence of high computing power core chip.

Officially, "DRIVE Thor supports 8-bit Floating Point Format (FP8), which provides 1000 INT8 TOPS/1000 FP8 TFLOPS/500 FP16 TFLOPS performance while reducing overall system cost."

In addition to BYD, Hao Platinum, Xiaopeng, Ideal, and Extreme Krypton, Chery and Baidu's Geely joint venture, Extreme Yue, have all announced the adoption of the next-generation DRIVE Thor centralized in-vehicle computing platform.

Ahead of the launch of DRIVE Thor, which continues to be used by various car companies, DRIVE Orin is a SoC (system-on-chip) chip that provides 254 TOPS of computing power for intelligent driving.

At the Beijing International Automotive Exhibition (BIAE), the company will be joined by smart, a joint venture between Mercedes-Benz and Geely, the upgraded ET7 sedan launched by Azera Motors, Great Wall Motor's luxury premium SUV WEY Lanshan Intelligent Driving Edition, BYD, DENZA, and YANGWANG, as well as Guangzhou Automobile AION's luxury brand Hyper, Ideal Motors, Lotus, Momenta, Polestar, SAIC R Polestar, SAIC R Motors, Xiaomi SU7, and ZEEKR, among others, demonstrated adoption of the DRIVE Orin chip.

NVIDIA is known not only for its hardware, but also for its continuous updates on the software side, including Xiaopeng, BYD and others who will plan to use NVIDIA's software to help streamline the automotive design process as well as improve factory efficiency.

In NVIDIA's official announcement, it says, "NVIDIA Omniverse is an API, SDK and services platform that enables developers to easily integrate Universal Scene Description (OpenUSD) and RTX rendering technologies into existing software tools and simulation workflows to build AI systems."

"Developers and software vendors such as CARLA, MathWorks, MITRE, Foretellix and Voxel51 have emphasized the broad appeal of these Omniverse Cloud APIs to the entire self-driving car space."

Xiaopeng Automobile announced that it is using the NVIDIA Omniverse platform to streamline the design workflow for its flagship Xiaopeng X9.

BYD will utilize the NVIDIA Isaac and NVIDIA Omniverse platforms to develop tools and applications for virtual factory planning and retail configurators.

Automotive manufacturing is a complex process, and NVIDIA is exploring the "theory and practice" of in-vehicle, utilizing generative AI, visual language models (VLMs), the NVIDIA Avatar Cloud Engine (ACE), and multimodal language models through the combined use of these technologies and in concert with the NVIDIA DRIVE platform. Through the combined use of these technologies and working with the NVIDIA DRIVE platform, automakers can develop their own intelligent in-vehicle assistants.

Examples include Ceres' CaLLM, Wayve's next-generation AV2.0, Ideal Automotive's Multimodal Cognitive Model Mind GPT, Azure's NOMI GPT, Waabi is building AI for autonomous driving, Geely is leveraging generative AI and LLM technologies to deliver a smart cockpit experience and edge cloud acceleration collaboration, Lenovo is launching its new AI acceleration engine called UltraBoost, and SoundHub has launched a new AI acceleration engine called UltraBoost. AI acceleration engine called UltraBoost, SoundHound AI's in-vehicle voice interface, Tata Consultancy Services' Automotive GenAI suite, and MediaTek's Dimensity Auto Cockpit Platform product family of automotive-grade SoCs based on NVIDIA technology from the edge to the cloud.

As more car companies participate in the cooperation and NVIDIA accumulates data in the car end, NVIDIA will be further strengthened from the cloud to the car end, from computing power to algorithms.

Algorithms are not enough, arithmetic power comes to the rescue

According to Canalys research data in January this year, global electric vehicle (EV) sales will grow 29% to 13.7 million units in 2023, with a penetration rate of 17.1%. Greater China remains the largest EV market, with shipments of 7.6 million units and a market share of 55.5%. In 2023, Europe with 3.2 million units shipped and North America with 1.8 million units shipped will be the top three markets in 2023.

China's auto exports will reach 5.4 million units in 2023, with EVs accounting for 40% of that at 2.2 million units.

Canalys forecasts that the global electric vehicle market will grow 27.1% to 17.5 million units in 2024. Electric vehicles from Chinese automakers are expected to account for 78% of the market by 2024.