Launched in the final version last December, MechWarrior 5: Mercenaries the last chapter of a glorious fighting series that has its roots even in the 80s. Inspired by the board game BattleTech RPG and wargame publisher FASA, MechWarrior a mech piloting simulation (in part RPG and management), with first and third person view. The series consists of numerous titles, from the first chapter released in 1989 to MechWarrior 2: 31st Century Combat (1995), MechWarrior 3, in 1999 and MechWarrior 4: Vengeance (2000), almost all of which have become cult figures with the public of fans.
The game passed through the hands of many development teams, from Dynamix to Zipper Interactive, from FASA Interactive to Piranha Games, the Canadian software house not to be confused with the German Piranha Bytes (Gothic, Risen). Thanks to Unreal Engine 4Piranha Games with MechWarrior 5 manages to offer surprisingly detailed graphics.
MechWarrior 5: Mercenaries allows players to build, manage and pilot their unit through a campaign that takes place over several years during the Third and Fourth War of Succession of the imaginary of MechWarrior. Whether you go on missions alone, together with AI-controlled pilots, or with the help of up to three friends in cooperative play, players are faced with complicated situations that don't provide a perfect solution. They need to decide which targets to prioritize and how to balance repairs and upgrades without wasting valuable resources on their unit.
Since a few weeks NVIDIA introduced its algorithm to Deep Learning Super Sampling (DLSS) in version 2.0, with which it has revised and corrected some principles and assumptions of operation of this anti-aliasing technique implemented in hardware in the video cards of the family GeForce RTX. We have already seen how in the case of Minecraft RTX DLSS goes to solve a series of performance problems related to the use of Ray Tracing. With MechWarrior 5 we once again record benefits in terms of very important frame rates, without loss of quality or, even, with quality improvement.
DLSS is a futuristic technique that could solve many anti-aliasing implementation problems. Apply anti-aliasing without compromising on quality using the artificial intelligence and hardware part of the GeForce RTX GPU used for it, i.e. Tensor Core. DLSS 2.0 uses a deep learning neural network, called NGX, Neural Graphics Framework, which is trained offline with a supercomputer to which tens of thousands of high resolution images are administered. The supercomputer renders them at very low frames using 64 samples per pixel, so as to promote quality without worrying about performance. After the network is trained, NGX provides the results of its calculations to PCs via NVIDIA driver update or other types of OTA updates. The Tensor Cores of the Turing architecture, therefore, thanks to the computing capacity up to 110 teraflop for the artificial intelligence with which they are equipped, they identify if in the game there are images similar to those already processed offline and propose the available results in real time. We talked about it extensively here.
With DLSS 2.0, NVIDIA promises superior image quality through the use of a better AI bot for remote image processing. The initial algorithm, in fact, allowed an improvement in the frame rate, but at a high cost in terms of image sharpness. The result was that DLSS 1.0 was put aside, while being fundamental in terms of performance to make up for the reduction caused by Ray Tracing.
What really changes the logic behind the bot's operation. In fact, he is no longer trained on the basis of individual games but with generic images. The system still needs to be trained, and NVIDIA must provide specific integration for each new game that comes out, updating the drivers accordingly. But it does not receive "game" images related to the game: it can therefore be used in a transversal way, and not expressly in the gaming field. The ability of AI to recreate a game image that we could compare to a puzzle that consists of a million pieces with only a handful of elements at its disposal, and with almost perfect precision, to say the least surprising. Of course, a GeForce RTX video card is still required, because the DLSS function linked to theTuring architecture and in the presence of Tensor Core in it.
Just like in Minecraft RTX, there are three DLSS options available: Performance, Balanced and Quality, which identify an ideal compromise between quality and performance according to the processing resources required. With DLSS enabled in combination with a pre-existing anti-aliasing technique, not only do you not lose performance, but the graphics are improved overall. Based on ours benchmark, with DLSS 2.0 you gain approximately 23% of the performance with GeForce RTX 2080 Ti, at 4K resolution and DLSS 2.0 in the Quality variant enabled, together with TxAA. A very important improvement, which highlights NVIDIA's new anti-aliasing technique.
Off
Performance
Balanced
Quality
Off
Performance
Balanced
Quality
But the developer Piranha Games has stuffed his simulation game with many graphic options, offering very detailed scenarios that will make fans of the genre happy. Visual distance, textures, effects, vegetation, shadows, post-processing and anti-aliasing can be configured. It is precisely this last item that interests us most in this article, together with FidelityFX CAS. In addition to DLSS, therefore also for Radeon cards, we have two other anti-aliasing filters available, FXAA and TXAA. Here they are.
FidelityFX Content Adaptive Sharpening the hardware-agnostic dynamic sharpening filter to improve visual details in all circumstances, useful in this case to allow players to grasp any detail even in dark areas. The contribution of FidelityFX CAS here is interesting, because it not only increases the level of detail by adjusting the contrast ratio, but improves the overall sharpness of the image. A decidedly superior result to what we had found in Resident Evil 3.
MechWarrior 5: Mercenaries does not have an integrated benchmark tool, but the hangar area, very detailed polygonal, has proved particularly suitable for testing GPUs. Graphically MechWarrior 5 very interesting, and also in terms of environmental destruction proves pleasant. The mechs are visually very faithful and the playing areas are pleasant. It suffers, however, from obvious pop-in effects, that is, some polygons appear suddenly when you quickly travel the game map (it occurs mainly with the third person view). The lighting is also not so advanced, especially in those areas where direct sunlight does not reach.
It is a game, as well as all those based on Unreal Engine 4, particularly optimized to take advantage of GPUs, while it does not require a particularly performing CPU. A good test for our experiments on DLSS and FidelityFX. We have it tried with seven video cards, with these performance results.
Ultimately, using an artificial intelligence technique to solve the age-old problem of performance loss with enabling anti-aliasing could be a point of no return for video game graphics. Of course there are still limits: that is, NVIDIA has to do a lot of work on a per game basis to implement it, and you need a GeForce RTX video card, usually expensive. For its part, AMD must react to this technological effort. Its open-source platform GPUOpen has already proven that it is not firm, and FidelityFX Content Adaptive Sharpening (CAS) is an example.
As we have seen towards us, this technique actually manages to improve the level of detail, especially of objects distant from the observation point, without loss of performance with Radeon video cards. to be refined, and it depends a lot on the game and its type of graphics, but it can really add details where the programmers have not been able to deepen the level of realism that certain visual elements deserve.
