DLSS 3 is an evolution of the well-known rescaling and reconstruction technology. supported by NVIDIA artificial intelligence, which maintains that base but introduces, at the same time, a very important novelty that has allowed it to make a significant leap in performance without relying on other components, the generation of frames.
DLSS 3 technology has two parts, one which is the process of re-scaling and intelligent reconstruction of the image, which was already present in DLSS 2, and another which is the frame generation using artificial intelligence. The latter is the one that represents that very important innovation we have referred to, and we are going to go into it in depth in this article.
However, we are not going to limit ourselves to talking about frame generation. To give you a better understanding of what DLSS 3 is, how it works, and what value it brings. we are going to go through its two parts in depth. However, if you have any questions after reading this article, you can leave them in the comments and we will help you to solve them. Without further ado, let’s get down to business.
NVIDIA DLSS 3: reconstruction and rescaling of the image.
DLSS 3 technology maintains, as we have said, the image reconstruction and rescaling process. Thanks to artificial intelligence, the algorithms used by this technology can create frames with high final resolution and high image quality, starting from images rendered with a lower number of pixels. than the target resolution.
The process is very easy to understand with an example. When we use DLSS 3 in performance mode and we have as target resolution 3,840 x 2160 pixels, the rendering resolution becomes 1,920 x 1,080 pixels. The algorithms combine previous frames with the current frame.and use spatial and temporal elements, and also to motion vectors, to create a high quality frame that will maintain an excellent level of detail in both near and far objects, and even those with a low geometric loading.
Traditional rescaling systems do not use artificial intelligencenor do they resort to temporal elements, they simply are limited to filling pixels starting from a spatial basis which, in the simplest cases, is equivalent to “stretching pixels”. Other more advanced ones resort to fixed algorithms that slightly improve the result, but do not reach the level that we would obtain with DLSS 3, since they tend to show much lower definition and a lower level of detail.which ultimately results in a blurred and washed-out appearance. They are also prone to problems with flickering and fading of distant objects, especially when these have a low geometric loading.
Thanks to the image reconstruction and rescaling process offered by DLSS 3, and also present in DLSS 2, it is possible to create a frame with a quality very close to native 4K starting from a 1080p resolution.. This implies a significant performance improvement, and a minimal sacrifice in terms of image quality. However, reducing the resolution increases the dependence of the graphics card on the CPU, which makes it infeasible to reduce the resolution to levels at which the processor may end up producing a bottleneck.
Frame generation: the solution to the per-CPU bottleneck.
When DLSS 3 was unveiled from NVIDIA, they confirmed that the most important new feature introduced by this technology was the frame generationand indicated that it would work in conjunction with image reconstruction and rescaling to further improve performance.
The process of rescaling and image reconstruction using artificial intelligence. is performed on the tensor kernelswhile the frame generation is executed on the “Optical Flow Accelerator”a component that is integrated into the GeForce RTX 20, GeForce RTX 30 and GeForce RTX 40 graphics cards.
This graphics engine accelerates the entire workload which involves the generation of frames, and works in conjunction with the DLSS Frame Generation AI.which decides how all the information obtained during the frame generation process is used and defines the final result.
Frame generation aims to. improve performance, but it does not resort to a reduction in resolution, nor is it an optimization of the reconstruction and rescaling process. This technology analyzes two sequential frames that have been rendered in the traditional way, i.e. with the intervention of the CPU, and uses the information from those frames and previous frames to generate an additional frame totally independent, i.e. without the intervention of the CPU.
We might think that DLSS 3 “invents” a frame, but this would be a mistake, because what it actually does is to generate an additional frame with the information it obtains from other frames, and it uses artificial intelligence, motion vectors and optical flow (“Optical Flow”). to get an accurate prediction of how all objects and elements within that frame should look, including such complicated things as shading and lighting.
With this technology we have one extra frame for every two frames generated in the traditional way.. This frame is not affected by the CPU, since it is totally independent as mentioned above, and thanks to the use of motion vectors, optical flow and intelligent frame analysis, DLSS 3 manages to maintain a very good image quality.
The performance improvement we can achieve with frame generation is so great that, depending on the game, this. can be increased by as much as 548%. performance over the native mode. For comparison purposes, the improvement we would achieve with DLSS 2 alone in the same scenario would be 329%.
However, frame generation comes at a cost at the latency level. This could be a problem in specific cases, but NVIDIA has solved it with the Reflex technologytechnology, which is automatically activated when it detects that frame generation is running, and performs a synchronization job between the CPU and GPU. which manages to reduce latency considerably, and makes the response rate in games good enough for an optimal experience.
DLSS 3 compatible graphics cards and supported games.
The center frame has been generated using DLSS 3.
We have already said that DLSS 3 resorts to tensor cores and “Optical Flow Accelerator”, and we know that the frame generation runs on the latter. The GeForce RTX 20, GeForce RTX 30 and GeForce RTX 40 have tensor cores and also mount an “Optical Flow Accelerator”, but only the latter support frame generation.
This has an explanation, and that is that the “Optical Flow Accelerator” present in the GeForce RTX 20 and GeForce RTX 30. does not have enough computing power, nor enough precision.to achieve a good result with frame generation. Therefore, NVIDIA has limited DLSS 3 support to the GeForce RTX 40. GeForce RTX 20 and GeForce RTX 30 support DLSS 2, and NVIDIA Reflex.
All DLSS 3 compatible games are compatible with DLSS 2, but not vice versa, it is necessary to implement the frame generation technology. At the moment the list of titles that already include support for such technology is very small, but we know that it will be gradually expanded, as it happened at the time with the second generation DLSS.
It should also be noted that in order to enable DLSS 3 and frame generation in games, games must use DirectX 12because this technology does not work under DirectX 11. This is not a problem, since DirectX 11 is an obsolete and disused API, but it is important to be clear about this because, in games that allow us to choose between one and the other, we have to choose the first one to be able to activate the frame generation.
