NVIDIA DLSS 3.5 technology, also known as flash reconstruction, has become one of the most important advances that gaming AI has seen, and has shown to be key to further progress towards achieving higher quality ray tracing without incurring too much cost in terms of performance.

What can NVIDIA DLSS 3.5 do?

It allows create high quality images with much more intensive ray tracing in games and professional applications. So its value is not limited to the world of video games, and it makes a big difference in rendering tasks and other professional scenarios where ray tracing is used.

The end result that this technology is capable of achieving has been spectacular ever since amplifies all effects that occur due to light interactionsincluding direct and indirect lighting, what we know as global lighting, shadows, ambient occlusions, reflections and caustic effects.

How flash reconstruction works

As already explained by NVIDIA, its basic task is generally the same as that of a noise limiter, but unlike traditional manually tuned models, NVIDIA DLSS 3.5 works through a neural network and uses AI to achieve excellent results. This allows you to improve image quality as well as performance.

If we want to understand it better, it is necessary to review how ray tracing works. Yippee rendering technique that can realistically mimic lighting the scene and the elements that make it up, including the smallest objects. This includes both reflections, shadows, caustics and global lighting of the scene.

To do this, it follows the path that the light follows by emitting rays from an image camera located in each scene that determines the point of view within it. These rays interact with the 3D scene and create reflections that eventually return to the light source. Two things can happen with these rays: failure or success.. If there is an impact with a reflective surface, we will have a reflection and its intensity will depend on the surface of the object on which it occurred.

Generally speaking, it is the same thing that happens with light rays in real life. A light bulb that illuminates a room will also cause more or less lit areas, will generate shadows and also reflections. Using ray tracing, this behavior of light is recreated in a realistic way, although of course it has its limitations.

It is a very demanding and intensive technique. Rendering multiple rays per pixel can take hours to render an image This is not viable in video games as we need to generate at least 60 frames per second to achieve a good level of fluency. In this case, a smaller number of rays are used to obtain a representative sample of the overall illumination of the scene.

There is a positive side to this, and it is allows you to apply ray tracing to complex scenes in a few milliseconds, but it also has a negative side, and that is the resulting image it has a lot of noise and it lacks pixels and information. This problem is solved by noise reduction, which fills in the missing pixels using the information present in the scene. They resort to two techniques:

• Temporary accumulation of pixels in multiple frames.
• Spatial pixel interpolation to connect adjacent pixels.

It works, but they increase the complexity and cost of the development process, and ultimately negatively impact frame rates in ray-tracing intensive games. NVIDIA DLSS 3.5 solves this problem by using AI to create smart noise limiterwhich is able to:

• Use additional information from the graphics engine.
• Recognize the different ray tracing effects used as well as the patterns you have been trained for.
• Distinguish between good and bad pixels, both spatially and temporally.
• Maintain a higher level of data for the scaling process.

The difference this technology makes is huge in terms of image quality, and the best part is that it also improves performance in games and applications that make extensive use of ray tracing. NVIDIA DLSS 3.5 is Compatible with GeForce RTX 20 and higher graphics cardsand is hardware accelerated using tensor cores.