Author | ZeR0
Edit | Moying
This Wednesday, the "new ceiling" of the game graphics card - the GeForce RTX 40 series GPU - is about to be officially launched. Since the official release of
in mid-September, the performance improvement brought by Nvidia RTX 40 series GPUs has always attracted much attention. With the recent corresponding evaluation results released one after another, I believe that this new graphics card series, known as the "ultimate platform for gamers and creators", has left a deep impression on many players in terms of performance improvement.
snippet the controversy over price and power consumption, now RTX 4090 has become the well-deserved new "Ki" and has achieved comprehensive suppression of competitors in terms of unprecedented computing power and cutting-edge technologies: 's modern game performance has been improved by up to 2 times compared to the previous generation RTX 3090 Ti, and ray tracing game performance has been improved by up to 4 times.
Even the second flagship RTX 4080 has achieved higher performance than the previous generation flagship graphics card RTX 3090 Ti. In the words of Nvidia founder and CEO O Huang Renxun , the performance that used to cost $2,000, but now it can be bought for only $900.
Interestingly, the core contributor to the soaring performance and cost-effectiveness of the RTX 40 series GPU this time is the artificial intelligence (AI) technology that Nvidia has been proud of in recent years.
In the past, AI was more of the protagonist of Nvidia's data center product line. Why did it also begin to highlight the role of AI in the consumer-level field that focuses on games? Behind this, Nvidia's ambition has spread to a broader future of potential AI applications through games.
1. frame rate is higher, AI frame supplementation becomes a "killing move"
Speaking of the most eye-catching technology of the RTX 40 series, AI tool DLSS 3 is the only one to take charge.
AI is not uncommon in the gaming field. Since AlphaGo developed by DeepMind defeated the world Go champion, AI has successively dominated chess, Japanese chess, StarCraft, Dota 2, Honor of Kings and other chess and card games and real-time strategy games. In the field of game development, AI has gradually taken on more creative tasks such as writing line scripts, dubbing, and translating to generating real-time 3D facial animations.
For top game masterpieces, plot, system, level design and visual effects are indispensable. People are always pursuing higher-definition picture quality, smoother fluency, cooler special effects, and more simulated scenes... These are the directions that Nvidia has been investing heavily in research and development, and optimizes various experiences of PC games by introducing AI.
Among them, the pioneering DLSS 3 (deep learning super sampling) technology is an important "killer" to widen the performance gap between the RTX 40 series and the previous generation.
The previous generation of DLSS technology can "brain" the clear image from the blurred image, that is, let AI predict what the current frame with a higher resolution is roughly like by analyzing the current frame with a low resolution and the previous frame with a high resolution, and then generate high resolution images to achieve image quality improvement.
image quality has been improved, but the frame rate is still a difficult problem.
If the frame rate cannot be raised, it is equivalent to switching between PPT of the 3D picture with naked eyes. No matter how shocking the visual effect is, it will be so stuttering that it makes people lose patience. Nvidia's newly launched DLSS 3 technology was born to "fill the frame".
DLSS 3 directly achieves "making something out of nothing". By adding pioneering optical multi-frame generation technology, allows AI to "brain" a brand new complete high-quality picture in real time, achieving doubled frame rate.
NVIDIA test shows that in the overdrive mode and 4K resolution of the classic modern ray tracing game "Cyberpunk: 2077", when ray tracing is turned on, if DLSS 3 is turned on at the same time, the frame rate can reach 100FPS.
Since DLSS generates frames on the GPU as post-processing, even if the game is limited by CPU performance, such as physical computing-intensive or large-scale scene games such as Microsoft Flight Simulation, DLSS 3 can also allow the RTX 40 series GPU to achieve up to 2 times the performance computable by the CPU to render games.
With this technology, Nvidia RTX 40 series graphics card can support the rendering needs of large-scale games, allowing players to enjoy the shocking visual effects brought by image quality improvement, while playing smoothly due to the surge in frame count, making it more enjoyable. The "fill-frame" principle of
DLSS 3 is not complicated, but if you specifically disassemble the series of technologies involved, it is not difficult to understand why it is the first to make such pioneering R&D results.
2. Moore's Law "battery endurance" is not enough, AI must take over
to deal with the slowdown in Moore's Law. Nvidia's core idea in recent years can be summarized into one word: full stack .
According to Huang Renxun, the era of Moore's Law providing the same performance at half the cost every year and a half has ended. The cost-effectiveness increase driven by the increase in the density of transistor is no longer as fast as that of the past. The process is continuing to improve, but the cost has also increased much.
▲NVIDIA founder and CEO Huang Renxun
Taking RTX GPU as an example, from the RTX 30 series to the RTX 40 series, the chip process process has been upgraded from the Samsung 8N process to the more advanced TSMC 4N process. However, according to Huang Renxun, this upgrade only brings about 15% performance improvement to the RTX 40 series, and the remaining performance improvement comes from R&D innovations such as DLSS 3, Tensor Core, and Shader Execution Reorder (SER).
The key leverage to realize these technology combinations is AI.
DLSS 3 includes 3 components: deep learning super resolution (that is, the previous DLSS 2), deep learning frame generation (that is, the newly added interpolated frames in DLSS 3), and Reflex that reduces PC latency.
Among them, the optical flow accelerator built into Nvidia's new generation Ada architecture is one of the hardware cores that supports the implementation of DLSS 3. It can analyze two consecutive game images and provide neural network with pixel-level frames to the motion direction and speed information of objects and elements in the frame, knowing how to correctly render lighting and shadows when objects move. Another key architectural innovation in
is Tensor Core, which is good at matrix computing in Ada architecture, which provides a hardware foundation for accelerating AI computing. Nvidia says that the fourth-generation Tensor Core, which is upgraded in the RTX 40 series graphics card, has added the Hopper FP8 Transformer Engine, and the FP8 processing performance is as high as 1.4 Petaflops, which is 5 times that of the previous generation graphics card.
DLSS super resolution technology uses AI to build image frames with higher resolution. After integrating a pair of super-segment image frames, motion vectors, and optical flow field information in the game, all information is input to the convolutional neural network AI frame generator that NVIDIA has trained. After calculation and analysis, the AI frame generator can accurately reconstruct geometric graphics and effects, and efficiently render and generate a new high-quality game frame.
Combining the new frame generated by DLSS with the DLSS super resolution frame, can use AI calculation to reconstruct pixel up to 7/8 of two consecutive frames, generate a new high-resolution frame, and then insert it between normal game screens. Compared with no DLSS, DLSS 3 can improve gaming performance by 4 times.
Considering that interpolation will bring additional delay, Nvidia has also introduced Reflex low-latency technology , which ensures the best response speed by synchronizing GPU and CPU and eliminating useless frames in the GPU render queue.
Although multi-frame generation technology relies on "brain repair", it is not a series of violent calculations from modeling maps to lighting information like traditional rendering, but because of AI, the big killer, the effect of frame insertion is enough to be scattered.
Finally, the three elements that affect the game experience: fluency, response speed, and image quality are all guaranteed.More than 35 games including
" Anti-Shui Cold ", "Black Myth: Wukong", "Cyberpunk 2077" have been announced to support DLSS 3 soon, which can be powered by specific hardware acceleration on RTX 40 series graphics cards to achieve a smoother frame-complement effect.
Recall that when DLSS was first launched, the integration of games and app developers was not fast. In order to promote the rapid popularity of DLSS 3, Nvidia also launched the Streamline open source development plug-in in April, allowing developers to easily add various new technologies to the game in a modular way, significantly reducing the difficulty of integrating new technologies. As the threshold for application is lowered, more and more game developers will choose to quickly embrace new technologies.
It can be seen that AI technology has complemented Nvidia's other "golden signboard" - hardware-accelerated real-time ray tracing , further expanding the gap between Nvidia's flagship graphics cards and other graphics cards in terms of leading performance.
3. AI super sampling + real-time ray tracing, knocking on the door to rendering innovation
In 2018, the launch of the Nvidia RTX series officially ushered in the era of real-time ray tracing.
Ray tracing is a graphic rendering technology that simulates the light emitted by various light sources in the computing environment being reflected, refracted, shadowed and indirectly illuminated, and finally enters the human eye. It has higher and more accurate visual fidelity than traditional rasterization, but also has higher computing overhead and is very hardware configuration.
Four years later, modern games like Cyberpunk perform more than 600 ray tracing calculations per pixel to determine light, requiring up to 16 times more computational increase, but the increase in the number of transistors responsible for such calculations in the GPU can no longer keep up with this speed.
It is precisely through the previous article that uses AI to calculate some pixels and predict most pixels, and Nvidia has been able to continuously improve its performance within 4 years to meet this computing demand growth.
As of now, there have been more than 280 ray tracing games released in the past three years. Although the newly launched mainstream graphics cards have announced that they support ray tracing, most games are still in a hybrid rendering mode. Ray tracing is only used to express the light effects of certain scenes, and the overall effect still has a lot of room for improvement.
Since the game is not completely ray tracing, why does Nvidia still have to make such an advanced technical layout?
As Huang Renxun said, the cutting-edge technological innovations such as real-time ray tracing and AI super sampling introduced by the Ada architecture are all paving the way for future games based entirely based on simulation.
Looking back on history, this is actually Nvidia's usual style of behavior - if the demand does not exist yet, then create demand.
When Nvidia invented the colorable programming GPU, the GPU application market was still quite quiet; when Nvidia created the world's first 3D electronic game platform, the electronic game market was very small; when Nvidia proposed a new computing method "accelerated computing" and spent a lot of money to develop general computing GPUs and unified programming software CUDA, few people believed in the blueprint it drew...
Results, history has repeatedly verified Huang Renxun's accurate judgment and insight on future market trends. From gaming graphics cards to AI computing platforms and meta-universe platforms, Nvidia has repeatedly "breaked the circle" and continuously explored new business growth points through full-stack innovation combining software and hardware to open up new business growth points for its longer-term market space. Similar to
, is also planning to cultivate ecological advantages in the long run in in optimizing the real-time ray tracing experience.
According to data from market research institutions, Nvidia's shipment volume in the independent GPU market accounted for as high as 80%, which makes its gaming GPU advantage snowball-like forming a positive cycle - the more users of its ray tracing games, the more feedback it gives, and the more friendly optimization of such games will be. The better the game optimization experience, the wider the applicability, the stronger the compatibility and stability, the higher the user stickiness.Game players such as
are accustomed to the more layered light and shadow changes brought by "AI super sampling + real-time ray tracing" and the silky high frame rate experience. It is difficult to accept the stiff visual effects made by traditional rendering methods after looking back. By then, the position of real-time ray tracing technology in the gaming field will shift from "better than nothing" to "slaying all sides", and Nvidia's voice in the high-end gaming graphics card market will also be more stable.
Because of this, Huang Renxun has a very high confidence in the RTX 40 series, saying that this means "a new era of real-time ray tracing and neural network rendering using AI to generate pixels has arrived. ".
Nvidia is also focusing on lowering the threshold for developing ray tracing games, and at the same time, it provides the possibility of RTX to those popular old games. For example, it has created an AI tool for game MOD RTX Remix to quickly trace old games.
developers can use their AI aid tools to improve the resolution of textures and materials, or convert materials into materials with precise physical attributes, while presenting realistic light and shadow effects, giving the game texture a new look.
As more games can be quickly "RTXed" through RTX Remix, Nvidia is optimistic about the ray tracing game market, which is expected to be accelerated.
4. Conclusion: Don’t underestimate Nvidia
"The past ten years have been the revolution in algorithm invention of AI, and the next ten years have been the decade of AI application." In the GTC keynote speech, Huang Renxun said that the engine that promotes the development of computing technology is accelerated computing, while fuel is AI.
Ten years ago, Professor Geoffrey Hinton's team at the University of Toronto, Canada won the ImageNet large-scale visual recognition competition using deep learning algorithms, which began to set off a tide in the popularization of AI. The hardware foundation that supports this breakthrough progress is the Nvidia GPU, which made its fortune in graphics processing and gaming.
Today, Nvidia has grown into the recognized "number one player" in the field of AI computing, but the potential market foreseeable AI applications is far broader than the current existing market. From 3D games to the metacosmic world, AI has too many places to show off. AI tasks become more and more complex, and more powerful computing power is required. This is exactly what Nvidia's technology and products are good at.
RTX 40 series GPU has given the example of AI-driven graphics processing, and through the support of a series of AI capabilities, it opens up a new path for optimizing computing graphics and changing the gaming experience.
In the long run, Nvidia's full-stack advantages in technological innovations such as architecture, design, and algorithms will continue to promote it to provide higher gaming performance at lower prices, and provide long-term capital for Nvidia to stand at the top of the GPU pyramid.
