NVIDIA shows off its muscles, gives AI computing power, and also uses computing power to do AI

On May 30, NVIDIA's stock price hit a new high, reaching $404. At this time, NVIDIA's market value was more than $1 trillion, becoming the first chip company on the planet with a market value of more than one trillion dollars.

Now that nearly two months have passed, NVIDIA's market value has gone further, and it is currently stable at around 1.1 trillion US dollars. This round of NVIDIA's stock price surge, the outside world believes that there is only one direct propellant, and that is AI.

Whether it is OpenAI's ChatGPT or some large domestic models, the source of its computing power is inseparable from the Tensor Core GPU provided by NVIDIA (different from the CUDA Core GPU, this GPU is suitable for the matrix budget required by artificial intelligence deep learning). Now on NVIDIA's official website, they define themselves as "the leader in the field of artificial intelligence computing."

In fact, in addition to Tensor Core GPUs like A100 or H100, NVIDIA's self-driving chips are naturally closely connected to AI, because it needs to process data and information from sensors such as lidar, millimeter-wave radar, and cameras in real time to make real-time driving decisions, which involves a lot of artificial intelligence calculations. As for Nvidia's old business, the consumer-grade graphics card is actually inseparable from AI.

DLSS 3, Reflex, and the games that support it

On July 20, 2023, NVIDIA will hold a GeForce RTX 40 series media tasting in Shanghai, focusing on the newly launched GeForce RTX 4060 series graphics cards, DLSS 3 technology, and various applications accelerated by RTX graphics cards.

DLSS (Deep Learning Super Sampling) technology uses AI to output higher-resolution frames from low-resolution input. For example, DLSS can use AI to increase 1080P image quality to 4K while using 1080P computing power, which means a significant increase in performance. DLSS would not be possible without AI and dedicated Tensor Core acceleration.

The latest generation of GeForce RTX 40 series GPUs supports DLSS 3. DLSS 3 adds AI-driven frame generation technology to the DLSS 2 super-resolution technology to further improve performance. It can be said that without the advancement of AI, there will be no playable frame rate with panoramic ray tracing turned on.

On the tasting conference line, most of the area is reserved for various games, such as the open world action RPG game "Naruto", "Naruto" is a two-dimensional action game with next-generation image quality. In the battle, we can manipulate the character to hit beautiful high-speed combos; we can also perform extreme dodge in the face of the enemy's fierce offensive; and under some specific moves, drifters can also use tricks to fight back against the attack of the BOSS; there are also QTE synergy gameplay and so on. Mechanisms all greatly increase the mobility of combat.

As the first game that Kulo Games and NVIDIA have reached a cooperation, the demo provided this time supports DLSS3 technology. In addition, it also supports NVIDIA's Reflex, which is a technology for measuring and reducing latency in competitive games. In FPS shooting games such as "Overwatch: Return", we click the mouse to shoot, and there is a lot of "delay" in the millisecond time when the bullet hits the enemy on the screen:

  • Peripheral Latency: The time it takes for an input device to process mechanical input and send those input events to the PC
  • Game Latency: The time it takes for the CPU to process input or changes to the game world and submit new frames for the GPU to render
  • Render Latency: The time between when a frame is queued for rendering and when the frame is fully rendered by the GPU
  • PC Latency: The time it takes for a frame to travel on the PC. This includes game lag and rendering lag
  • Display Latency: The time it takes for the display to render a new image after the GPU has finished rendering the frame
  • System Latency: Time covering the entire end-to-end measurement – ​​from the start of peripheral latency to the end of display latency

NVIDIA Reflex includes two main new technologies: NVIDIA Reflex SDK and NVIDIA Reflex Latency Analyzer. The former allows the game engine to complete rendering tasks in real time, thereby eliminating GPU rendering queuing and reducing CPU backpressure; Reflex Latency Analyzer can monitor mouse clicks and measure the time it takes for the corresponding pixels on the screen (such as gun flames) to change, thereby noticing the delay time between player operations and screen display, taking advantage of a few milliseconds to tens of milliseconds.

In addition to "Naruto", NVIDIA also exhibited a series of games that support DLSS 3 and Reflex, including "Edge of Rebirth", "The End of Time", "Diablo IV", "Cyberpunk 2077" and more.

In addition to these newer games, there is also an RTX version of "Portal: Overture". "Portal" is a puzzle-solving work that was popular in the game circle 15 years ago. For current players, this old phone looks quite simple, but NVIDIA RTX Remix technology can provide game MOD lovers with simple and stunning ray tracing production effects.

The RTX version of "Portal: Overture" has been re-made on the basis of the original version with materials and light, enriching more picture details, including smoke, water surface effects, etc., making the game more authentic. The image effect can be further denoised to make the picture quality more delicate. Players can re-experience a classic work 15 years ago from a modern perspective. Its release and popularity also means that the technical integration threshold of RTX is very low, and even old games more than ten years ago can be easily RTXized. This has encouraged many game developers, who can implement RTX's advanced special effects support by simply modifying games that are under development or have been released and have received critical acclaim, including realistic ray tracing special effects and ultra-high frame rates under the blessing of DLSS 3.

The game that best embodies Reflex technology is "Fearless Contract". Combined with Reflex mouse and monitor, this technology can provide e-sports players with a new ultra-low-latency e-sports experience. The GeForce RTX 4060 device equipped in the e-sports area can achieve 1080p, fps 360+ with a delay of less than 10ms; 1440p, fps 240+ with a delay of less than 20ms. The top e-sports experience.

NVIDIA Powers Local AI Creation

Although the computing power of the large language model service is basically on the cloud side, the computing power of the local device basically does not affect the experience, but some applications on the device side actually test the AI ​​capabilities of the device side, so in a sense, the performance of the graphics card is also approximately equal to the AI ​​​​performance of the device side.

In addition to displaying AI-related game technologies such as DLSS 3 and Reflex, NVIDIA's support for local AI creation is also an important part of the event.

The most typical example is the CG work "Flower" created by NVIDIA Studio acceleration and AI-assisted creation first released by the well-known UP master special effects brother at station B.

This work was accelerated with RTX GPU, rendered with NVIDIA Omniverse and Unreal Engine, and created with Stable Diffusion. For Stable Diffusion running locally, GeForce RTX 40 series graphics cards can provide users with an image generation speed up to 1.5 times that of the previous generation, improving the efficiency of creators using AI for creation.

In addition to the fit with open source AI tools such as Stable Diffusion, NVIDIA also demonstrated that the Clipping Professional Edition uses the eighth-generation NVENC encoder and AV1 encoding technology to increase the video export speed by 2 times with the support of the 40-series graphics card.

On NVIDIA Canvas, powered by AI and NVIDIA RTX GPUs, users can quickly convert simple strokes into realistic landscapes. Now that Canvas supports 360° panoramas, users can not only create beautiful wrap-around environment maps and use them as the background of 3D scenes, but also test various concepts and atmospheres.

As the biggest beneficiary of this round of AI wave, NVDIA’s graphics card, AI technology, and AI applications displayed at this event are actually the closest to users, and they also have the lowest threshold for users to experience. As for cloud-side AI computing such as farther and stronger autonomous driving and large language models, there is still a distance from users, but just as cellular networks have made the era of mobile Internet, the computing power distributed on the cloud side and the device side is the power source of the AI ​​​​era, and it is the air and water of AI.

In the plastic greenhouse of fate, every cabbage that has been sprayed with too many pesticides once had a dream of becoming a pollution-free organic vegetable.

#Welcome to follow Aifaner's official WeChat public account: Aifaner (WeChat ID: ifanr), more exciting content will be presented to you as soon as possible.

Ai Faner | Original Link · View Comments · Sina Weibo