Sophia Windsor
[Gearbest Technology]At 10:30 am on January 7, NVIDIA founder and CEO Jensen Huang gave a keynote speech at the CES 2025 conference. The speech focused on AI and unveiled a series of new products, such as the RTX 50 series graphics cards. At the same time, he also shared the company’s innovative results in the fields of automobiles and robotics. There are many highlights in this speech. Gearbestwill next take you to review NVIDIA’s actions at CES 2025.
RTX 50 series graphics card
The RTX 50 series graphics cards are undoubtedly the focus of this conference, and its price strategy also surprised many people. The RTX 5090 is priced at $1,999, while the RTX 5080, 5070 Ti, and 5070 are available at $999, $749, and $549 respectively. According to official claims, the performance of the flagship model RTX 5090 has doubled compared to the previous generation RTX 4090. What is even more remarkable is that from RTX 5090 to RTX 5070, the performance of the entire new generation GPU series has been significantly improved compared to the previous generation. Improvement, which also triggered heated discussions among netizens, jokingly saying that Nvidia “squeezed too much toothpaste” this time. Huang Renxun also demonstrated an ultra-thin notebook equipped with RTX 5090 graphics card at the scene, priced at US$2,899.
Usually, such a huge performance leap is not common, which makes people wonder whether Nvidia has incorporated some kind of “black technology” into the new graphics card. Through in-depth analysis of the benchmark results, the answer is gradually found, and that is DLSS 4.
It can be seen from the RTX 5090 performance chart released by NVIDIA that compared to the RTX 4090 using DLSS 3, the RTX 5090 has achieved a significant improvement in performance with the multi-frame generation technology of DLSS 4. RTX 5090 is based on the new GB202 core and has up to 21,760 CUDA cores, becoming the first GeForce GPU with a core count exceeding 20,000. In addition, it is equipped with a 512-bit memory bus and 32GB GDDR7 memory, with a power consumption of 575W, an increase of 125W compared to the RTX 4090.
However, not all games enjoy such significant performance improvements. For example, the performance improvement data of “Far Cry 6” and “A Plague Tale: Requiem” are relatively conservative. Among them, “A Plague Tale: Requiem” only supports DLSS 3, so it cannot take advantage of the multi-frame generation advantages of DLSS 4; and “Far Cry 6” does not support DLSS technology at all, making it a relatively weak performance among Nvidia’s display games.
The RTX 5070, which will be launched in February, has also attracted much attention. Huang Renxun boasted at the press conference that the performance of the RTX 5070, which sells for only $549, is comparable to the RTX 4090. This statement is also mainly due to the multi-frame generation technology of DLSS 4.
However, it is worth noting that multi-frame generation technology also comes with latency issues. Similar to DLSS 3, using frame generation technology essentially means gaming with a slight lag. While latency is minimal, DLSS 3 can be problematic at lower base frame rates. Still, it’s a blast for gamers at large. For those players who have recently purchased 40 series graphics cards, they may feel a little regretful.
In addition to its graphics card products, Nvidia also announced a series of other noteworthy products and large models.
The world’s smallest personal AI supercomputer—Project Digits
According to Huang Renxun, Project Digits is equipped with the latest GB10 Grace Blackwell super chip, which has a built-in NVIDIA Blackwell GPU. This GPU is equipped with the latest generation CUDA core and fifth-generation Tensor Cores, which can drive complex AI models and processing capabilities. Astonishingly, it is said to be able to easily handle AI models with up to 200 billion parameters.
Project Digits not only performs well in terms of hardware configuration, each device is equipped with 128GB of unified memory as standard, and also provides up to 4TB of NVMe high-speed storage space, ensuring efficient and smooth data processing and storage. In addition, users can simply connect two Project Digits to build a more powerful computing cluster, which can easily handle ultra-large models with up to 405 billion parameters. It is worth mentioning that Project Digits will be officially launched in May this year, with a starting price of US$3,000.
The core competitiveness of Project Digits lies in its new Grace Blackwell super chip (GB10). This system-on-a-chip (SoC) based on the Grace Blackwell architecture can provide up to 1 petaflop AI performance at FP4 precision. GB10 integrates NVIDIA Blackwell GPU (equipped with the latest generation CUDA cores and fifth-generation Tensor Cores), and is closely connected to the high-performance NVIDIA Grace CPU (20 energy-saving cores using Arm architecture) through NVLink-C2C chip-to-chip interconnect technology. It is worth mentioning that MediaTek also participated in the design of GB10, which played a certain role in improving energy efficiency, performance and connectivity.
Next-generation automotive smart driving chip——Thor
Thor is also a major attraction of this exhibition. This is a disruptive robotic computer designed to process massive amounts of sensor data. NVIDIA uses the Omniverse platform to conduct in-depth testing of driving scenarios and generate synthetic data to accelerate the development process of artificial intelligence models. He is proud to announce that NVIDIA Drive OS has successfully obtained certification, becoming the world’s first programmable artificial intelligence computer functional safety software that meets the highest automotive safety standard ASIL-D.
In terms of partners, Huang Renxun revealed Nvidia’s extensive layout in the automotive field. Domestic partners include well-known companies such as BYD, Ideal, Xiaomi, and Jikrypton, while overseas cooperative car companies include giants such as Tesla, Mercedes-Benz, and Toyota. In addition, Toyota, AURORA and Continental have also officially joined NVIDIA’s cooperation camp, working together to launch the next generation of highly automated and self-driving fleets.
Huang Renxun said that Nvidia’s automotive business will achieve a revenue scale of US$5 billion in fiscal year 2026.
large model
Huang Renxun pointed out at the scene that there are about 1 billion knowledge workers worldwide, and predicted that AI agents are expected to become an industry worth trillions of dollars.
To this end, NVIDIA has launched a basic model of the world called “NVIDIA Cosmos”. This is a basic platform designed to help AI understand the real environment and can receive various types of information such as text, images, and videos. The platform integrates functions such as world basic model generation, advanced word segmentation and video acceleration processing channels, and can generate virtual world images.
At the application level, Huang Renxun specifically mentioned the two popular areas of intelligent driving (end-to-end) and AI robots. Both application scenarios require understanding the physical world, converting data into machine-recognizable language, and then further converting it into images to provide services for humans.
In addition, NVIDIA also demonstrated the Llama Nemotron language basic model. The model is available in three sizes: the Nano version is cost-effective and optimized for low-latency real-time applications, ideal for deployment on PCs and edge devices; the Super version is a high-precision model capable of delivering on a single GPU Excellent throughput; the Ultra version is the highest accuracy model, designed for data center scale applications with high performance requirements.
According to Huang Renxun, the Llama Nemotron model performs well in areas such as chat, function calling, programming and mathematics. At the same time, the model is optimized in size to run on a wider range of NVIDIA devices, providing higher accuracy and greater model throughput.
humanoid robot
In his speech, Huang Jen-Hsun also shared his insights on humanoid robots. He predicted that the era of humanoid robots is gradually approaching, which indicates that general robotics technology is about to usher in a major breakthrough. At the same time, he revealed the list of manufacturers that Nvidia has cooperated with in the field of humanoid robots, including domestic Yushu Technology H1 and Xiaopeng Laotie.
Huang Renxun said that developers around the world are working on building the next generation of humanoid robots. This process requires collecting and processing massive amounts of real-world data, and the cost is considerable. To this end, NVIDIA launched Isaac Groot to solve these problems and provide developers with four key supports: basic robot models, data pipelines, simulation frameworks, and Thor robot computers.
write at the end
The press conference lasted for about an hour and a half. The insights about AI shared by Huang Renxun and the products launched all showed Nvidia’s leading position in the field of AI chips. With the help of these products, Nvidia’s stock price is expected to hit new highs this year.
Login with Google