Everything you need to know about DLSS 3.0

DLSS 1.0

was trained to recognize a game scene. From then on, this scene is initially rendered with fewer pixels, that is, with a lower resolution. This saves processing power.

In the next step, DLSS 1.0 enters. The technology reconstructs the game scene using what it learned from the neural network, improving image quality. Thus, we have a frame that was rendered in a lower resolution but that, after undergoing an upscaling process, was presented in a higher resolution.

was trained to recognize a game scene. From then on, this scene is initially rendered with fewer pixels, that is, with a lower resolution. This saves processing power.

Furthermore, the first version of DLSS was presented along with Ray Tracing. This technology calculates how light rays affect each game element in real time. The result is much more realistic light, shadow and reflection effects. If you want to know more about Ray Tracing, we also have an article dedicated to this topic:

What is Ray Tracing?

The big problem with DLSS1.0 was that few games supported the technology. It launched with a library of just 35 games. The reason for this is that the neural network needed to be trained for each game specifically. And that takes time and resources.

DLSS 2.0

The second version of DLSS was developed to run from the Tensor Cores of the RTX family of graphics cards. These processing cores are dedicated to artificial intelligence and deep learning tasks.

Furthermore, the artificial intelligence model applied in DLSS 2.0 made Tensor Cores work at least twice as fast. The result of this is an increase in the frame rate of games and a significant improvement in graphics. The technology manages to make the details sharper and use only half of the pixels to form the frame.

The user can still choose between three image quality modes of DLSS2.0 technology. There's the quality mode, which prioritizes beautiful graphics, the balanced mode, and the performance mode.

However, the great advantage of "DLSS version 2.0" is that it uses a more generalist neural network. This means that it is no longer necessary to train the neural network for each game individually. So much so that several games currently support DLSS 2.0 technology.

And what's new in DLSS 3.0?

The third version of DLSS was released by NVIDIA promising up to four times more performance when it comes to native rendering of a game. DLSS 3.0 is designed to work exclusively with the new 4th generation Tensor Cores, which utilize the Ada Lovelace microarchitecture .

That is, only powerful and expensive GeForce RTX 4090 and RTX 4080 graphics cards are capable of working with DLSS 3.0. This will certainly leave a huge portion of users unable to experience the new version of technology. Unfortunately, GPUs from the RTX 20 and RTX 30 series were left out.

The big difference between DLSS3.0 and the two previous versions is that now it creates new frames between the frames natively generated by the video card. DLSS 1.0 and 2.0 only added pixels to the image via artificial intelligence to increase the resolution and quality of the graphics. "DLSS 3.0" will create entirely new frames.

In addition, DLSS 3.0 works in conjunction with NVIDIA Reflex and Optical Stream Accelerator. This technology, present only in the RTX 40 series cards, feeds the DLSS neural network with information about the movements of pixels in subsequent frames. This ensures that performance is accelerated even in games that rely more on the processor than the graphics card.

Due to the high computational power of the Ada Lovelace cores, the Optical Flow Accelerator is able to analyze two sequential game frames, calculate their movement and predict what the pixel change will be in the next generated frame. The analysis is so thorough that it works at the pixel level, analyzing particles, shadows, reflections and all the lighting in the scene.

Finally, NVIDIA Reflex is a technology built into DLSS 3.0. Together, the two technologies deliver optimal response time with low latency. Just to give you an idea, latency is reduced by at least half.
No comments
Post a Comment