With the arrival of the NVIDIA RTX 3080, the company has presented a series of quite interesting novelties, of which it already dropped many pearls a short time before. The launch to the market brings us a fight between the two versions of Deep Learning Super Sampling, more specifically between DLSS vs DLSS 2.0 . What news and improvements are there between these two versions? Is it worth the change?
That DLSS has revolutionized games at the level of Ray Tracing, it is undoubted, but it is also true that the first version of this technology of Temporal SuperSampling did not finish really well and the criticism arose strongly, although some time later between the improvements in the games and NVIDIA support all went much better.
Although it was still not what was promised, NVIDIA presented a second version called DLSS 2.0 that, this time, promises to redefine the representation in real time using the AI and the Tensor Cores present in the RTX.
NVIDIA DLSS vs DLSS 2.0, an update with important news
DLSS is based on NVIDIA’s NGX technology as the main function to use AI, for this, the RTX function has to be integrated into the game in question, where certain files will be interconnected with some NVIDIA driver DLLs so that neural networks ( DNN ) can give access to RTX technologies through the SDK.
The previous training of these DNNs in the Saturn V supercomputer allows the images to be rendered at a very low FPS speed and with 64 samples per pixel, which offers us a lower resolution input and an output to our monitor with high images. resolution.
For this, DLSS is fed by low resolution images that are rendered by the game engine, where the movement vectors of those images will indicate to the SDK the direction of the scene to jump from one frame to another, allowing the next one to already is offered with high resolution.
Key points where improvements are concentrated
As expected, everything discussed has a series of key advantages where this new version surpasses the previous one in the DLSS vs DLSS 2.0 dispute. Specifically, we talk about four points that NVIDIA emphasizes to show the virtues of its new Super Sampling technique:
- A neural network that does not depend on the specific game. One of the most objectionable points of DLSS was precisely that each game needed to be previously trained and independently, making the whole process very complicated. With DLSS 2.0, the AI network is generalized and for all games, which should result in a much faster adoption of this technology and in a greater number of games.
- Personalization: now we can choose three levels of image quality with three different levels of performance -> Quality, Balanced and Performance.
- Better GPU scaling in RTX: Tensor Cores are used by the driver and SDK much more efficiently, specifically and according to NVIDIA, up to 2 times faster, thereby eliminating any bottleneck in the GPU and allowing to win more FPS.
- Image Enhancement: DLSS 2.0 offers improved image quality, with sharper details and better FPS stability.
In general, everything that the first version did has been improved, so that the dispute between DLSS vs DLSS 2.0 does not currently exist, since its successor has achieved a qualitative leap forward. In terms of pure performance, NVIDIA figures an improvement of up to 76% in depending on which scenarios, so we are facing a step and a half ahead to not only improve the image in our games, but to be able to gain performance with it or in its defect, alleviate the performance drop of Ray Tracing.