They are here. Finally. PC graphics hardware enthusiasts were eagerly awaiting the arrival of the first cards of the GeForce RTX 40 family, and NVIDIA has not disappointed us. A few minutes ago Jensen Huang, the CEO of this American company, unveiled the new GeForce RTX 4090 and 4080, two high-end graphics cards that promise us amazing performance when using ray tracing.
However, they have not arrived alone. Beyond the many new features that the Ada Lovelace microarchitecture has in store for us, and which we will delve into in a special article we are already preparing, the GeForce RTX 40 arrives backed by DLSS 3 image reconstruction technology. And yes, what NVIDIA promised us during its presentation looks very good. Astonishingly good.
Here’s the promise of DLSS 3: 4x performance with ray tracing
NVIDIA’s DLSS (Deep Learning Super Sampling) technology arrived alongside the first generation of NVIDIA GeForce RTX graphics cards with a promise under its belt: to allow us to enjoy our video games with a higher frame rate per second even if our graphics demands were very demanding. Even when ray tracing is enabled.
The purpose of this innovation is to relieve the GPU of some of the effort involved in rendering images to increase the frame rate without affecting the graphics quality.
The idea is far-reaching and, as users can guess, the technology that makes it possible is complex. The image reconstruction technique employed by NVIDIA relies on real-time analysis of our game frames using deep learning algorithms.
The strategy used by NVIDIA to alleviate the effort required by the GPU is similar to that used by other graphics hardware manufacturers: the rendering resolution is lower than the output resolution that the graphics card finally delivers to our monitor.
In this way, the stress on the graphics processor is lower, but in exchange, it is necessary to resort to a procedure that is responsible for scaling each of the frames from the rendering resolution to the final resolution. Moreover, it must do it efficiently because, otherwise, the effort that we have avoided in the previous stage could appear in this phase of the generation of the images.
This is the stage where the artificial intelligence that NVIDIA has fine-tuned comes into play. And the GPU’s Tensor cores. The graphics engine renders the images at a lower resolution than we expect to get, and then DLSS technology scales each frame to the final resolution by applying a deep learning sampling technique to try to recover as much detail as possible.
In the images, we have used to illustrate this article we can see that the procedure implemented in DLSS 3 is more complex than the one used by DLSS 2. NVIDIA’s new image reconstruction technique takes advantage of the presence of the fourth-generation Tensor cores of the GeForce RTX 40 GPUs to enable the execution of a new reconstruction algorithm called Optical Multi Frame Generation.
Instead of approaching the reconstruction of each frame by working with isolated pixels, which is what DLSS 2 does, this strategy generates complete frames. To do so, it analyzes two sequential images of the game in real-time and calculates the vector information describing the movement of all the objects that appear in those frames, but are not processed by the game engine itself.
NVIDIA claims that this image reconstruction technique achieves a fourfold increase in the frame rate delivered by DLSS 2. And, importantly, it minimizes the aberrations and visual anomalies that appear in some games when using the previous revision of this image reconstruction strategy. It sounds great, so we’re looking forward to testing it to see if its performance is as compelling as NVIDIA is promising us.
One more interesting note: the processing of high-resolution frames and motion vectors are fed, NVIDIA tells us, by a convolutional neural network that analyzes all this information and generates in real time an additional frame for each frame processed by the game engine. To conclude, here is another promise from this company: DLSS 3 can work in tandem with Unity and Unreal Engine, and over the next few months it will reach more than 35 games. It is possible to enable this technique in a short time in those titles that already implement DLSS 2 or Streamline.