On January 7, on the eve of the 2025 CES meeting, NVIDIA officially launched its much-awaited new generation PC graphics card-the GeForce RTX50 series.Also appearing along with the graphics card were the supercomputer Project DIGITS and the World Model Cosmos platform.

For more news, Hawk Insight brings you one article.

What is CES 2025?

CES 2025 will be held in Las Vegas from January 8 to January 11, 2025. As one of the largest and most influential consumer electronics exhibitions in the world, CES has always been an important event for investors in the technology industry to judge trends in the new year.This exhibition attracted many technology giants and start-ups from around the world to jointly showcase the latest technological innovations and products.

RTX 50

The release of this series of graphics cards marks another major breakthrough for Nvidia in the consumer GPU market.The RTX50 series adopts the new Blackwell architecture, which has significantly improved performance compared to the previous generation RTX40 series.Among them, the flagship model RTX5090 graphics card has 21760 CUDA cores and 32GB GDDR7 memory, with a memory bandwidth of 512 bits and a memory bandwidth of 1792 GB/sec.In terms of ray tracing and AI computing, the performance of the RTX5090 reaches 318 TFLOPS and 3352 AI TOPS, making it excellent in rendering realistic light and shadow effects and deep learning applications.

In terms of price, the RTX 5090 sells for as high as US$1999, while the RTX 5080, 5070Ti and 5070 sell for US$999, US$749 and US$549 respectively.Despite the high price, Huang Renxun emphasized at the press conference that the RTX5070 can provide comparable performance to the RTX4090 for US$549, mainly due to the significant improvement of DLSS4 technology.DLSS4 technology uses multi-frame generation technology to increase the frame rate of games by 8 times. In games such as "Cyberpunk 2077", after DLSS4 is turned on, the number of frames is generally increased by at least 30% compared with RTX4090.

Llama Nemotron

In addition to graphics cards, NVIDIA also launched an AI basic model adapted to RTX AI PCs at CES, which is provided in the form of NVIDIA NIM microservices.These models, combined with RTX50 series GPUs, can provide up to 2375 AI TOPS computing power and 32GB of video memory.Nvidia launched the Llama Nemotron series of basic models, including Nano, Super and Ultra versions, which are targeted at different application scenarios and performance requirements.The Nano version is suitable for deployment on PCs and edge devices, the Super version strikes a balance between computing efficiency and accuracy, and the Ultra version is designed for high-performance data center applications.

Project DIGITS

The supercomputer Project DIGITS released by Nvidia is equipped with a new GB10 super chip, making it the world's smallest AI supercomputer capable of running a 200B parametric model.Project DIGITS allows users to develop and run model reasoning on desktop systems and seamlessly deploy models on accelerated cloud or data center infrastructure.In order to further strengthen its strength in the SoC field, Nvidia chose to cooperate with MediaTek, which has rich experience in this field.The GB10 super chip uses a Blackwell GPU and is connected to a high-performance Grace CPU through NVLink-C2C inter-chip interconnection. The CPU uses ARM architecture and has 20 energy-efficient cores.

Autonomous driving and robotics

Nvidia's business expansion is not limited to the PC field, but its ambitions in the fields of autonomous driving and robotics cannot be underestimated.Nvidia demonstrated its new Cosmos platform for developing Physical Artificial Intelligence (Physical AI), which relies on the "Basic Model of the World" to allow companies to simulate multiple scenarios in a virtual environment and develop software for robots and autonomous vehicles.In addition, Nvidia has also launched the next-generation automotive processor "Thor" and announced cooperation with many car manufacturers such as BYD, Ideal, Xiaomi, Krypton, Tesla, and Mercedes-Benz.Nvidia expects its automotive business to grow by approximately $5 billion in fiscal year 2026.In the robotics field, Nvidia has launched NVIDIA Isaac GR00T Blueprint, which helps developers generate massive amounts of synthetic motion data to train humanoid robots through imitation learning.

market did not buy

Unfortunately, the market does not pay for the seemingly explosive product.What Wall Street is more concerned about is not long-term planning, but short-term benefits. For example, Blackwell's production, production capacity, and product yield were not mentioned in these press conferences.

After the product was announced, Nvidia plunged.22% intraday, the largest one-day decline since September 3, 2024.The market value evaporated by US$227.5 billion (approximately RMB 1,666.8 billion) overnight, closing at US$140.14/share.U.S. stocks and chip stocks generally fell, with the Philadelphia Semiconductor Index falling 1.84%.TSMC fell more than 3%, Broadcom fell more than 3%, ON Semiconductor fell more than 2%, Maiweir Technology fell more than 2%, and Micron Technology rose more than 2%.