NVIDIA, the global leader in the field of artificial intelligence, has announced its new generation of AI chips and software for running AI models.
In the past, NVIDIA updated its GPU architecture every two years. Since the emergence of ChatGPT in 2023, large AI models have been emerging one after another globally, with most of them trained and generated on NVIDIA's previous-generation Hopper architecture.
Compared to Hopper, the new generation of NVIDIA's AI chip, Blackwell GPU, which was released last night, is composed of approximately 20.8 billion transistors, 2.5 times more than Hopper's transistor count. It is manufactured using TSMC's custom 4NP process and is based on the GB200 processor it has developed. The AI performance of Blackwell GPU reaches 20 PetaFLOPS, while H100 achieves 4 PetaFLOPS.
The training performance of Blackwell GPU is four times that of the previous-generation Hopper GPU, and the inference performance is 30 times higher, with an energy efficiency improvement of approximately 25 times.
[PetaFLOPS: a unit of computing speed equal to one thousand million million (10^15) floating-point operations per second]
Therefore, the first-generation AI supercomputer NVIDIA DGX SuperPOD, equipped with the NVIDIA GB200 Grace Blackwell super chip, was born. At the same time, NVIDIA also launched software services such as NIM, which can help companies that want to run their own AI models without buying access from companies like OpenAI, by using NVIDIA's accelerated computing library and generative AI model construction.
In the new era of the gradually heating AI computing power competition, NVIDIA, with absolute dominance, is no longer satisfied with just being a chip manufacturer. They want to become a provider of complete solutions that integrate software and hardware.
This is also evident in the intelligent automobile industry, which has an increasingly hungry demand for AI computing power. On the same night, NVIDIA announced that several mainstream Chinese car companies have joined the next-generation NVIDIA Drive Thor chip camp.
Leading the way in adopting the Drive Thor chip are companies like BYD and GAC Motor. Users who are interested in smart electric vehicles are likely familiar with the OrinX chip. As of now, the OrinX chip, with a single computing power of 256 Tops, has become the widely preferred autonomous driving platform for mid-to-high-end electric vehicles. For example, all models of NIO's second-generation vehicles are equipped with four OrinX chips as standard. Additionally, some models from Li Xiang, XPeng, Jidu, and other brands are built on the OrinX platform to achieve advanced autonomous driving capabilities.
In other words, NVIDIA's dominance in the field of intelligent automotive driving with the OrinX chip is comparable to Qualcomm's Snapdragon chip in the high-end smartphone industry, with no competitor able to replace it.
However, even so, the next-generation Drive Thor chip represents a leap in performance. With a single computing power of up to ...
It is worth mentioning that, with the support of NVIDIA NVLink-C2C chip interconnection technology, two Thor chips can be connected at high speed without loss, with a total computational power of up to 4000Tops, providing unlimited imagination space for the future evolution of smart cars.
And this AI car chip with such terrifying performance will soon meet you officially. It will be mass-produced for the first time in 2025 and has been announced to be adopted by Chinese car manufacturers including Li Auto, Xpeng, Xiaomi, and Great Wall.
Last night, NVIDIA announced again that BYD, a global leader in new energy vehicle sales, will also adopt the DRIVE Thor chip and plan to use NVIDIA's AI infrastructure for cloud AI development and training technology, as well as develop tools and applications for virtual factory planning and retail configurators using NVIDIA Isaac and NVIDIA Omniverse platform.
Meanwhile, GAC Aion has also announced that its high-end luxury brand, Haobo, will adopt the DRIVE Thor platform for its next-generation electric vehicles. The new models will begin mass production in 2025 and will feature Level 4 autonomous driving capabilities. Xiaopeng Motors has also announced that it will use the NVIDIA DRIVE Thor platform as the "AI brain" for its next-generation electric vehicles, helping its self-developed XNGP intelligent driving assistance system continue to evolve towards the future.
There is no doubt that the DRIVE Thor chip will continue to maintain the brilliant and dominant position of the DRIVE Orin chip in the field of in-vehicle artificial intelligence chips.
NVIDIA's ambition goes beyond chips. As mentioned earlier, simply acting as a chip manufacturer is no longer sufficient for NVIDIA. Providing a complete and mature autonomous driving solution is also an important task that NVIDIA is trying to succeed in. To this end, NVIDIA has previously recruited Wu Xinzhou, the former head of Xiaopeng XNGP autonomous driving, as its global vice president in charge of autonomous driving business.
Xiaopeng XNGP is currently the leading mass-produced advanced driving assistance system in China. From the perspective of end-user experience, Xiaopeng recently launched the "Nationwide Open" Infinite XNGP, which does not rely on high-precision maps to achieve urban navigation, leading the industry forward together with Huawei's NCA.
Wu Xinzhou is the key contributor to Xiaopeng XNGP's development from 0 to 1. He has played a crucial role in the successful implementation of high-speed NGP and urban NGP, as well as the establishment and improvement of the entire team and technical research and development direction.
In August last year, Wu Xinzhou left Xiaopeng and joined NVIDIA amicably, serving as a global vice president and fully in cha...
If Nvidia's integrated hardware and software autonomous driving solution can be successfully implemented in the future, it will undoubtedly be a good choice for car companies with weak self-developed capabilities.
In short, whether it is intelligent driving or intelligent cabin, the AI trend in the automotive industry has already begun, and it is likely to be a key factor in the battle for core discourse power in the next decade.
What's even more frightening is that cars are just the tip of the iceberg when it comes to artificial intelligence. The magic of AI reaches every corner of society, and under the sweeping influence of AI, everything is possible to undergo earth-shattering changes. And the computational power behind it is the source that allows AI to flow.
