Nvidia released the most powerful consumer graphics card RTX 40 Super, and the AI performance increased by more than 170%|Hit CES 2024

英伟达CEO黄仁勋(Jensen Huang)

After Nvidia launched a new product, its market value soared by nearly $80 billion (about 567.88 billion yuan) overnight.

Titanium Media App reported on January 9 that the 2024 International Consumer Electronics Show (CES) will kick off in Las Vegas, USA on the 9th local time.

In the early morning of January 9, Beijing time, American chip giant NVIDIA (NVIDIA) released three new graphics cards for consumers online before the opening of CES 2024, including RTX 4080 SUPER, RTX 4070 Ti SUPER and 4070 SUPER GPU (graphics processing unit) chip products.

Nvidia said that this is the company's new GPU chip specifically for AI PC devices, which is the best engine to power new laptops and desktop computers, with AI performance (image generation speed) increasing by more than 170%, which will allow gamers, designers and other users to better use AI PC (artificial intelligence personal computer) technology. The new consumer graphics cards from the RTX 40 Super series will be used primarily for gaming and AI applications, and are coming at a "great new price." Among them, the flagship RTX 4080 SUPER will be available on January 31 at a price of US$999 (8,099 yuan).

However, it is worth noting that due to the implementation of new semiconductor export control regulations by the US Department of Commerce, Nvidia RTX 4090 is prohibited from being sold in Chinese mainland and "specializing" 4090D, so some flagship chips such as RTX 4080 SUPER may not be fully available to Chinese players. It is reported that Nvidia will start mass production of AI chip products designed specifically for China in the second quarter of 2024 that comply with U.S. export control policies.

Affected by new products and market news, as of the close of the U.S. stock market on January 8, Nvidia (NASDAQ: NVDA) shares rose 6.43%, and once again refreshed a record high in intraday trading, and the market value increased by about 560 billion yuan overnight. Over the past year, Nvidia's stock price has risen by more than 230%.

The latest consumer-grade GPU for dedicated AI PCs has been released, and AIxing is more than 1.7 times faster

At the Nvidia CES 2024 special event, Jeff Fisher, senior vice president of NVIDIA GeForce, officially displayed the long-awaited RTX 40 Super product, saying that compared with Intel and AMD, the new chip products will show an "order of magnitude" improvement.

"Today, generative AI is becoming an inflection point in the latest technological developments, and NVIDIA is at the heart of this major technological transformation. Jeff Fisher said that Nvidia RTX GPUs have shipped more than 100 million units and play a key role in the PC and AI markets.

Specifically, NVIDIA's new GeForce RTX 40 SUPER-series GPUs deliver powerful gaming and generative AI performance.

Among them, the GeForce RTX 4080 SUPER is equipped with NVIDIA's full AD103 chip, more CUDA cores and faster memory, with a bandwidth of 736GB/s, power consumption of up to 320 watts, and support for 4K panoramic ray tracing games. In the most graphics-intensive games, the RTX 4080 SUPER performs 1.4x faster than the GeForce RTX 3080 Ti that doesn't support DLSS frame generation. With 836 AI TOPS, DLSS frame generation delivers an additional performance boost, with the RTX 4080 SUPER delivering 2x faster performance than the RTX 3080 Ti, AI-generated video more than 1.5x faster than on the RTX 3080 Ti, and AI-generated images more than 1.7x faster than the RTX 3080 Ti. The RTX 4080 SUPER will go on sale on January 31.

The other two products, RTX 4070 Ti SUPER and 4070 SUPER, the former has more cores, video memory capacity, and performance 1.6 times faster than RTX 3070 Ti, which will be on sale on January 24, with prices starting at 6499 yuan, and GeForce RTX 4070 SUPER mainly increases the number of cores, 20% more than RTX 4070, and the performance and power consumption are comprehensively improved, which is better than RTX The performance of the 3090 is increased by 1.5 times, and it will go on sale on January 17, with a price starting at 4899 yuan.

Justin Walker, vice president of Nvidia, said that the new graphics cards meet export control requirements and can be offered to Chinese consumers, which also provides another option for researchers who do not have access to Nvidia's most powerful chips. But he did not say which product was supplied to the Chinese market.

According to The Verge, Nvidia is aiming for the RTX 4080 SUPER to target existing RTX 3080 Ti or RTX 2080 Super users, rather than RTX 40 series adopters. The performance boost compared to the RTX 4080 should be between 1%-5%, depending on the game, so essentially, it performs at the RTX 4080 level (or better) but at a price of $200 less than the 4080.

Nvidia has said that ASUS, MSI, LENOVO, HP, Samsung and other OEM terminal manufacturers will provide the supply of AI PC devices based on the RTX 4080 SUPER.

In addition, at this CES, NVIDIA announced 14 new RTX cloud games for customers. And, with the launch of the new ACE Production Microservice for NVIDIA Audio2Face and NVIDIA Riva automatic speech recognition, leading interactive avatars and game developers are exploring how ACE and generative AI technologies can transform how players and users interact with non-player characters in games and apps. At present, the developers who are actively welcoming ACE include miHoYo, NetEase Games, PalmFun Technology, Tencent Games, Ubisoft, etc.

"GeForce RTX SUPER graphics cards support more than 500 RTX games and apps and will prepare users for the wave of generative AI applications on PCs," said Matt Wuebbling, vice president of global GeForce marketing at NVIDIA. ”

Jeff Fisher emphasized, "Thanks to the powerful RT Cores and Tensor Cores, every RTX laptop is AI-ready for the best gaming and AI experience." Nvidia says GeForce RTX has become the world's largest generative AI platform.

Xiaomi Auto and Ideal will use NVIDIA's autonomous driving chip platform

In addition to announcing new consumer graphics cards, NVIDIA also announced the latest developments in the NVIDIA DRIVE family of automotive businesses.

Nvidia says Li Auto has selected NVIDIA DRIVE Thor centralized in-vehicle computers to power its next-generation vehicles. In addition, Great Wall Motor, Zeekr, and Xiaomi have adopted the NVIDIA DRIVE Orin platform to power their intelligent autonomous driving systems.

Among them, Nvidia mentioned that the Xiaomi electric car SU7 is based on the dual NVIDIA DRIVE Orin configuration, which can realize highway driving functions. It will be available in two versions: one with a range of up to 415 miles on a single charge and the other with a range of up to 497 miles. The SU7 will be officially available in the first half of 2024.

Wu Xinzhou, vice president of autonomous vehicles at NVIDIA, said: "The transportation industry is adopting centralized computing to achieve a high degree of automation and autonomous driving. "The AI car computer of choice for today's OEMs is NVIDIA DRIVE Orin, and automakers are increasingly looking to their successor, NVIDIA DRIVE Thor, for the advanced features and AI performance to shape future vehicle roadmaps. ”

At the same time, during this CES, Mercedes-Benz, Polestar, Kodiak, Pebble, Lenovo and others showcased the latest AI automotive products, all of which use the NVIDIA DRIVE series platform. For example, Lenovo will showcase its vehicle computing roadmap at CES and showcase new products based on NVIDIA DRIVE Thor, including the Lenovo XH1, a central computing unit for advanced driver assistance systems and intelligent cockpits, the Lenovo AH1 ADAS domain controller unit, and the Lenovo AD1, a Level 4 autonomous driving domain controller unit.

According to Nvidia, since 2022, NVIDIA DRIVE Orin has been developed with automakers, trucks, AI taxis and shuttles, delivering up to 254 TOPS performance and scalable to support L2+ to L5 autonomous driving capabilities.

In addition, NVIDIA announced a partnership with Getty Images, a Getty Images company, iStock, has released a generative AI service based on NVIDIA Picasso, an AI foundry for visual design, that provides a platform for AI models to create 4K image models from text, trained on a massive volume of authorized, commercially secure creative content from Getty Images. A new editing API is also coming soon, which gives customers more control over the resulting images.

However, the performance of Nvidia's AI chips is so strong, especially chip manufacturers that occupy up to 90% of the AI training chip market, but they have no way to win the favor of Chinese companies under the US chip ban. According to the news quoted by the global network, Nvidia wants to sell its lower-performance AI chips to Chinese buyers in response to the Biden administration's export restrictions, but Chinese customers are not interested in the latest status quo.

According to the report, Chinese companies such as Alibaba, Tencent, and Baidu have revealed their intention to reduce their orders this year because Nvidia does not offer its "most powerful chip". And in terms of performance, there is not much difference between China's domestic chips and Nvidia's "compliant" chips, which means that these Chinese companies may start sourcing local chips. In addition, Chinese companies are concerned that the United States may further increase export restrictions.

Nvidia CEO Jensen Huang has said that U.S. export controls are like "tying our hands behind our backs." China has been an important market for Nvidia, which accounts for 20 percent of the company's revenue, he said.

Earlier, Nvidia launched NVIDIA RTX 5880 Ada and RTX 4090D consumer graphics cards on its official website, mainly for the Chinese market, in which the number of RTX 5880CUDA cores was reduced by 22%, the single-precision floating-point performance was reduced by about 24%, and the overall performance was reduced by nearly 1/4, and the actual performance was close to the sub-flagship performance of the RTX 5000.

U.S. Commerce Secretary Raimondo has said that the U.S. allows Nvidia to sell AI chips to China on a limited basis, but does not allow Nvidia to export the most complex and powerful AI chips.

(This article was first published on the Titanium Media App, author: Lin Zhijia)