The race to photorealism has led the videogame industry to
invest not only in increasingly powerful machines, but also to develop new
technologies capable of collaborating with the hardware and carrying out a
synergistic work to achieve an optimal and uncompromising result.
NVIDIA most of all has pushed on the accelerator to speed up
this process, thanks to a series of insights that have allowed the entire
industry to benefit from technologies otherwise too early for the computing
power available. DLSS, now a cornerstone of PC Gaming , is undoubtedly the best
of these insights.
A revolution dictated by needs
The biggest step forward in recent years in gaming graphics
is undoubtedly Ray Tracing , a tool that has definitively changed the concept
of lighting and its interaction with the surrounding environment. The most
recent implementations of this technology have shown once and for all how a
correct calculation of reflections, light sources and their consequences on
surfaces can really overturn the concept of graphics . The price to pay for
such a satisfying visual rendering is obviously in terms of performance. While
the use of RayTracing can literally take your breath away, the collapse of the
framerate could compromise the overall experience to the point of wanting to
give it up. For this reason,NVIDIA has developed a system that would be an
understatement to define revolutionary . We are talking about DLSS, a
technology that can drastically improve the performance of the games that
implement it, as long as you use video cards that support it at the hardware
level.
The video cards of the RTX range have been designed by
NVIDIA specifically to be able to get the most out of the titles that support
the new technologies, thanks to the implementation of three types of
processors. Inside the NVIDIA GeForce RTX GPUs there are Streaming
Multiprocessors flanked by Tensor Cores , responsible for the execution of
Artificial Intelligence models and Deep Learning, and finally the Ray Tracing
Cores.
NVIDIA's secret weapon allows you to get an incredible
Framerate boost using Artificial Intelligence. Not surprisingly, DLSS stands
for Deep Learning Super Sampling . Today we will try to understand a little
more deeply how this technology has evolved over time and we will discover how
it is possible to obtain an increase in FPS without particularly affecting the
image, or sometimes even improving it.
A brilliant idea
DLSS is one of the upscaling methodologies . That is to say,
that particular type of technologies that, starting from an image at a given
resolution, are able to obtain a final result with a higher resolution . In
DLSS this process is proposed and improved thanks to AI and Deep Learning. The result
of many years of research work, NVIDIA releases the DLSS in its first iteration
in 2019 . What at the time appeared as a mysterious object, obscured by the
much more promising Ray Tracing, immediately showed the muscles on the albeit
fewsupported games. In fact, research on AI-based rendering in NVIDIA labs had
already been around for several years. The first hypotheses of use in the
purely photographic field immediately made us understand how the application in
the Gaming PC could shake the industry from its foundations.
Arrived on tiptoe, it immediately showed its full potential
in improving the performance of RTX GPUs, also thanks to the inherent beauty of
titles such as Battlefield V , Metro Exodus and Shadow of the Tomb Raider..
With the obvious technical limitations, also due to the young age of
technology, DLSS already allowed to obtain excellent results in terms of
performance, managing to provide up to 40% more frames per second, but required
a cost in terms of visual quality.
The operating mechanism was as simple as it was brilliant:
set the game resolution, for example 1440p, it allows the GPU to carry out the
normal activities of rasterizing and rendering the image at a lower resolution
and then perform the so-called Super Sampling . This recreates an image as
close to native resolution as possible, despite the lack of additional pixel
information.
The process of generating the final image is by no means
random. Harnessing the potential of machine learning , NVIDIA trained its
supercomputers to generate artificial intelligence models . These libraries,
then distributed through the Game Ready drivers , were the result of sampling
countless frames from each single supported game and then creating, through 64x
supersampling methods, a "perfect frame". The resulting image pairs
were eventually used to train the AI to fill lower resolution frames with
missing details.
Precisely this passage helps us to understand the real value
of this technique andits substantial difference from the more traditional SSAA
, which allows to perform the supersampling rendering locally the image at
higher resolution. In DLSS, this step is carried out remotely and then only the
final model is distributed to the end user to be fed to the Tensor Cores.
Good the second
The life cycle of the first DLSS, due to its cumbersome
limit of allowing training in watertight compartments for each single supported
title , lasted relatively short, also due to a not optimal final result in
terms of sharpness, especially in some scenarios. After less than a year, he is
literally blown away by Control's impetuosity . The masterful work of Remedy
Entertainment first exploits the DLSS 1.9 and then veers towards the definitive
version 2.0.
DLSS 2.0 marks the definitive consecration of NVIDIA's miracle
and manages to make us understand its philosophy once and for all. From this
moment on it becomes in fact an essential component for new generation games,
with and without Ray Tracing.
The keystone is precisely the removal of the limits imposed
in the first generation. NVIDIA's new neural network allows you to overcome
study barriers for each individual game by allowing the processing and learning
unit to learn generic rules that apply to all supported titles.
The novelties of DLSS 2.0 do not stop there, as the numerous
optimizations implemented by NVIDIA allow the Tensor Cores to maximize their
efficiency and perform calculations at almost twice the speed of the initial
version, without considering the possibility of choosing between three different
presets., allowing the end user to decide whether to set it to maximum quality,
maximum performance or an intermediate solution.
An impressive generational leap but above all without
contraindications, given that support for the second version was also guaranteed
for owners of first generation RTX video cards , namely the NVIDIA Series 20.
In our technical
analysis of Control , the gem of Remedy, we were fascinated by the progress
made by NVIDIA. Thanks to the comparison between the pre version and the post
update version to DLSS 2.0, we were amazed to see how the 1080p resolution DLSS
2.0 managed to churn out better frames than the previous version at 1440p and
even better even than the native version. No good magician reveals his tricks
but here we are faced with a masterful development work, which has allowed the
implementation in the DLSS of a filter so high that it is able to reduce if not
completely eliminate the temporal ghosting .
Among the best titles to highlight the prowess of your
Tensor Cores, we undoubtedly recommend returning to the wastelands of Metro
Exodus with the beautiful Enhanced Edition , free for owners of the original
version. A brilliant title from a technical point of view, which allows you to
fully enjoy Ray Tracing in one of its best implementations ever.
Never as in this title, the use of DLSS 2.1 is essential .
Its activation will accompany us in a fluid game world and simply perfect in
terms of sharpness, without practically ever coming to terms with the devil. A
look towards the new generation, through the lenses of a world devastated by
atomic war.
The DLSS today
Our journey ends with DLSS 2.1, the state of the art of this
technology. NVIDIA's latest update raises the bar even further, putting native
resolution rendering in check and also targeting 8K. Thanks to the Ultra
Performance Mode it is in fact possible to draw on an additional tank of Frames
on titles such as Wolfenstein: Youngblood and Cyberpunk 2077 . However, this
mode is strongly recommended only at very high target resolutions, under
penalty of a heavy compromise of the aesthetic component.
With the advent of 2.1, the DLSS finally lands also in VR ,
however it will take a few months to enjoy the first compatible titles. Today,
getting lost in the deep space of No Man's Sky in search of glyphs and portals
has never been so satisfying.
Wearing the headset, the procedural universe of Hello Games
is ready to take the breath away of old and new navigators with doubled
performance compared to the past and now really within everyone's reach,
allowing to obtain a stable framerate of 90FPS at Ultra details on an RTX 3080.
The future of the industry
In NVIDIA labs, the impossible has become possible. But what
can we expect from tomorrow in PC Gaming and, by extension, in the gaming
landscape?
DLSS and other optimization techniques such as Resizable BAR
undoubtedly represent an essential element for the next generations but the
real revolution is already underway, as NVIDIA has managed to offer for the
first time the possibility to access Ray Tracing without compromise even
through mainstream and entry level configurations . This will therefore be the
biggest news of the generation. Benefits so far reserved only for high-end
solutions that are finally accessible and within everyone's reach.
