Leifeng Network(Public number: Leifeng.com)It is reported that at the GTC event on September 20, NVIDIA CEO Huang Renxun released the much-anticipated RTX40 series of NVIDIA’s new generation gaming graphics card in his speech.
This time NVIDIA released a total of three graphics cards, the RTX 40 series flagship graphics card RTX 4090 and the high-end RTX 4080 16GB and RTX 4080 12GB.
RTX 40 series graphics cards adopt a new generation of RTX architecture, and the computing power has once again achieved a leap development. The new upgrade of light chasing technology and the addition of DLSS3 also brings a new upgrade experience to gamers.
The RTX 4090 graphics card will be available on October 12, while the RTX 4080 graphics card will be available in November.
The increase in quantity also increases the price, and the performance and price of the 40-series graphics card is a double “jump”
The RTX4090 uses the new third-generation RTX architecture and is manufactured using TSMC’s 4N process.
The new generation of RTX is named after the world’s first computer programmer Ada Lovelace. Huang Renxun said in his speech that thanks to the close cooperation between NVIDIA and TSMC,The 4N process optimized for GPU manufacturing allows the Ada Lovelace architecture to integrate 76 billion transistors and more than 18,000 CUDA cores, which is 70% more than the previous generation Ampere architecture and doubles the utilization of energy compared to Ampere.
Figure A new Generation of Ada architecture
The 40-series graphics cards using the Ada architecture perform much better than the 30-series graphics cards of the predecessors.
bannerThe onboard RTX 4090 integrates 76 billion transistors, 16,384 CUDA cores, and has 24GB of video memory.
NVIDIA said that with the blessing of powerful hardware indicators, the RTX 4090 graphics card can still achieve a smooth gaming experience of more than 100FPS even at 4K resolution.
In terms of energy efficiency, the power of the RTX 4090 is 450W, which is consistent with the RTX 3090Ti, but Nvidia claims that under the same power, the game performance of the 4090 is twice that of the RTX 3090Ti.
The RTX 4080, located at a lower level, has two memory versions, 16GB and 12GB, and the 16GB version integrates 9728 DUDA cores. Nvidia claims the RTX 4080 16GB has twice the gaming performance of the RTX 3080 and outperforms the previous generation’s flagship RTX 3090 Ti.
The 12GB version of the RTX 4080 integrates 7680 CUDA cores, and its theoretical performance exceeds the flagship RTX 3090 Ti of the previous generation.
The new architecture also includes the Micromap Opacity engine, which doubles the performance of the Alpha-Test ray-traced collection. The new Micro-Mesh engine can improve the richness of graphics without bringing more BVH build and storage resource usage.
Finally, the new Tensor Core adds the Hooper FP8 Transformer Engine to provide 1.4 petaFLOPS of tensor processing performance. This generation of Nvidia gaming graphics cards has also reached record prices while getting a huge upgrade in performance.
The suggested retail price for the RTX 4090 starts at 12,999 yuan, which is 1,000 yuan higher than the 11,999 yuan of the previous generation RTX 3090.
RTX 40 series graphics card price figure
The price of the RTX 4080 has reached 9,499 yuan, which is a “consumption upgrade” compared to the starting price of the previous generation RTX 3080 of 5,499 yuan. Even combined with the rumor that the current generation graphics card will not have NVIDIA RTX 4080Ti, the price of RTX 4080 16GB is slightly higher than the 9499 yuan price of the previous generation of the same position RTX 3080 Ti.
The price of the RTX 4080 12GB which has a lower form is also as high as 7,199 yuan, and the price of 4,499 yuan for the RTX 3070Ti with the same location as the previous generation has also risen sharply.
Chase light and upgrade again, AI in the game “steal the sky and change the sun”
One of the most important upgrades of this RTX 40 series graphics card is the upgrade of ray tracing technology.
RTX series graphics cards include ray tracing when they first came out, and the 40 series update is even more “toothpaste squeezed”, as NVIDIA CEO Huang Renxun said: “NVIDIA has redefined graphics .”
The new Ada Lovelace architecture has up to 90TFLOPs of shader capacity, twice the throughput of the previous generation. separately,The Ada architecture’s SM multi-unit stream processor also uses Shader Reordering Technology (SER), which enables real-time task reordering. Jen-Hsun Huang emphasizes that the technology is revolutionary for GPUswhich is similar in principle to the CPU’s out-of-order execution technology.
For this technology, Huang Renxun explained in his speech that a ray tracing workload requires different thread processing, and it is difficult to combine work between different shaders. SER technology can schedule shader loads in real time, thereby improving efficiency implementation and better Make better use of GPU resources to create better ray tracing effects.
Figure SER technology works
This technology can improve the ray tracing performance of NVIDIA graphics cards by 2-3 times and improve the overall performance of the game by 25%.
In addition to improving computing power, NVIDIA also improves the final game experience in a variety of ways. In his speech, Huang Renxun said that like accelerated computing, computer graphics is also a full-stack challenge. Achieving breakthroughs also requires innovation in architecture, design and algorithms.
DLSS technology, which has always played an important role on RTX graphics cards, has also presented a major update this time.
On RTX graphics cards, DLSS technology has always been an important “partner” of ray tracing technology: ray tracing simulates game scenes that are more realistic and closer to reality, while DLSS technology uses AI algorithms to reduce the calculation required for real simulation. physical environment to ensure a smooth gaming experience for users.
This time, the third-generation DLSS technology emerged with the RTX 40 series graphics card and brought an innovative optical multi-frame generation function.
The biggest difference from the previous one is that the third generation DLSS technology no longer produces pixels, but a comprehensive picture frame.
Based on the new optical flow accelerator based on Ada Lovelace architecture, the third-generation DLSS technology can analyze two consecutive frames of game images, and input pixel-level movement direction and speed information from frame to frame to the neural network, and then the neural network model will be calculated accordingly out of the middle frame.
Such intermediate frames are generated by neural network calculations, no longer involve graphics rendering, and are completely independent of the game. This can greatly reduce the computational load of real-time calculations in the game.
In layman’s terms, the pictures in the game are created by the “painter” of the GPU through the “brush” of computing power. DLSS3 technology can insert an AI-generated “picture” between the two paintings, thereby completely freeing up the GPU. “The artist’s hands.
Nvidia said that the third generation DLSS technology can improve game performance by 4 times compared to pure rendering.
Huang Renxun said that DLSS 3 can bring a much higher frame rate to Ada GPU than the CPU can calculate, so that some games with high CPU requirements can also benefit from it.
At the scene, NVIDIA showed two games with high CPU requirements. After turning on DLSS3, the frame rates of “Cyberpunk 2077” and “Microsoft Flight Simulator” both more than doubled to more than 90 frames .
Figure Before and after DLSS3 is turned on, the frame rate of “Microsoft Flight Simulator” changes
“Porth” connects the past and the future,Nvidia is rushing into the metaverse
Gaming was an important topic when the RTX 40 series was released.
At the same time as the release of the 40 series graphics card, this time NVIDIA also brought a remake of the classic game “Portal” in a retro spirit.
The game, called Portal RTX, is actually a mod created by Nvidia based on its metaverse creation platform, onmiverse.
“Portal” Remake figure using ray tracing technology
Nvidia cleverly used “Portal” to connect the past and the future.
Just like the gameplay of this classic game: open the portal and go to the other side.
In Nvidia’s opinion, the other end of the portal is undoubtedly the world belonging to the metaverse.
Huang Renxun said that mod is a game culture with a huge audience, and the number of downloads of various game mods has exceeded billions of times every year. In 9 of the 10 most popular competitive games, mods can be seen.
Nvidia has created an Ominverse app called RTX Remix for game mod designers.
Based on this tool, game developers can make mods for various classic games to add RTX ray tracing effects.
Based on this tool, mod makers need to import the game to USD, then use the modding tool to create a ray tracing mod and export it to the RTX renderer. The tool also uses AI technology to improve the texture of mod materials.
Nvidia has created a metaverse creative platform called Onmiverse a few years ago, allowing metaverse developers to connect and collaborate with each other through this 3D platform.
Whether it is the stronger GPU computing power of the RTX 40 series graphics cards or the more realistic light chasing effect, it will undoubtedly contribute more power to the ecology of the Metaverse.
At the beginning of Lao Huang’s speech, there is a demo of RacerX. RacerX is a fully interactive simulation demo built on the Omniverse, where everything is based on real world physics, and all lights, reflections and refractions use ray tracing technology.
This may indicate the future in NVIDIA’s eyes: a more realistic gaming experience and metaverse scenes complemented by more powerful graphics performance and AI capabilities.