NVIDIA has unveiled the new generation of GeForce RTX 50 series GPUs at CES 2025, including the GeForce RTX 5080, GeForce RTX 5090D, and other products for both mobile and desktop platforms. NVIDIA expects to officially release these graphics cards by the end of this month. Compared to past GPUs, the biggest feature of the GeForce RTX 50 series is the significant increase in AI computing power, making AI an important part of game rendering. NVIDIA recently held a media briefing to provide domestic media with a detailed introduction to the features of the GeForce RTX 50 series GPUs, as well as many new rendering technologies specifically developed for the RTX 50 series.
First, the RTX 50 series graphics cards use the Blackwell architecture. For gamers, the most significant change in the Blackwell architecture is the addition of a new FP4 unit, which, with the help of the fifth-generation Tensor Core, achieves up to 4000 TOPS of AI computing power, thus supporting more advanced AI applications. In the Ada architecture, each SM unit is divided into traditional FP32 computing units and FP32/INT32 hybrid computing units. In the Blackwell architecture, to make AI computing power more efficient, all rendering units are now FP32/INT32 hybrid computing units to meet the needs of AI computation.
The RTX 50 series graphics cards also feature GDDR7 memory, built on Pam3 signaling, offering higher frequencies and lower voltages. The memory frequency can reach 30Gbps, which is twice as efficient as GDDR6. Additionally, Blackwell’s ray-tracing units have been fully upgraded, allowing for more realistic lighting effects while reducing performance overhead.
Regarding the highly anticipated DLSS 4, NVIDIA provided an introduction. The company stated that DLSS research has been ongoing for six years, during which the DLSS algorithm has been continuously improved for greater efficiency. In previous versions of DLSS, NVIDIA used CNN convolutional neural networks for calculations, but in DLSS 4, NVIDIA switched to the more efficient Transformer model, improving the efficiency of the AI algorithm. In DLSS 3, the GPU renders one frame of the image while the AI simulates and calculates another frame, effectively inserting one frame between two original frames to improve game frame rates. DLSS 4 builds on DLSS 3, allowing up to three frames to be inserted, significantly reducing the computational burden on the GPU. This means the GPU is only responsible for 1/16 of the image rendering, with the AI handling the rest of the calculations and simulations.
This method has proven highly effective, with games using DLSS 4 achieving up to eight times the performance of native frame rates. Compared to DLSS 3, performance will increase by 70%. NVIDIA also announced that more than 75 games and applications will support DLSS 4 upon launch. At the event, we experienced Black Myth: Wukong, where with all effects enabled, DLSS 4 boosted the game’s frame rate from 23 frames to 193 frames, demonstrating impressive performance. Additionally, for esports games, NVIDIA introduced Reflex 2 technology, which further reduces game latency by up to 75% through AI calculations and other techniques. Currently, only The Finals and Valorant support this technology. Furthermore, NVIDIA also showcased several AI rendering technologies that game developers can use to create more realistic in-game visuals.
NVIDIA’s ACE technology is also becoming increasingly mature. We experienced the PC versions of Naraka: Bladepoint Mobile and Animal Punk, both of which have integrated this technology. ACE enables NPC responses to be more natural, allowing them to understand user language and perform corresponding actions to enhance immersion. Thanks to the improved AI computing power, this technology is no longer limited to cloud services. Game developers can store and run the relevant model algorithms locally for more efficient inference. We also tested the AI teammate feature in Naraka: Bladepoint Mobile, where you can give natural language commands to your teammate to perform various actions, making the game experience more realistic. However, this feature is currently only available in human vs. AI mode; PvP support is still in development.
At the event, we also experienced various new rendering technologies powered by AI, including the use of AI to generate a massive number of polygons to create more beautiful and immersive visuals. For example, NVIDIA’s dragon demo was built with millions of polygons, with incredibly detailed modeling of scales and claws, making the dragon appear lifelike. Such real-time rendering technology was previously difficult to achieve, but now, using AI, it can be effectively implemented while maintaining smooth performance. Through AI, NVIDIA also made object textures more realistic, ray tracing more natural, and significantly reduced the use of video memory, thus improving frame rates in games and applications. NVIDIA will release these technologies to developers through an SDK, allowing future games to use these techniques to create visuals that rival the real world.
Undoubtedly, in this era where AI takes center stage, NVIDIA’s next-generation graphics cards have made AI a major player in game rendering, delivering higher frame rates, as demonstrated by the RTX 50 series. Regarding traditional rasterization performance, NVIDIA acknowledges the challenges of improvement, as current core areas are already large, and adding more transistors is a huge challenge. Traditional rasterization may face bottlenecks, so leveraging AI computing power is an effective solution. For NVIDIA, the goal is to use technology to benefit gamers, ultimately allowing them to enjoy smoother and more realistic visuals.
The Resources comes from the Internet
Related:
Disclaimer:
- This channel does not make any representations or warranties regarding the availability, accuracy, timeliness, effectiveness, or completeness of any information posted. It hereby disclaims any liability or consequences arising from the use of the information.
- This channel is non-commercial and non-profit. The re-posted content does not signify endorsement of its views or responsibility for its authenticity. It does not intend to constitute any other guidance. This channel is not liable for any inaccuracies or errors in the re-posted or published information, directly or indirectly.
- Some data, materials, text, images, etc., used in this channel are sourced from the internet, and all reposts are duly credited to their sources. If you discover any work that infringes on your intellectual property rights or personal legal interests, please contact us, and we will promptly modify or remove it.
