It seems incredible, but it has been more than two years since NVIDIA introduced its Deep Learning Super Sampling technology on its RTX graphics cards. Although at first it was not good at all, now with the release of DLSS 2.0 things have improved a lot, but the reality is that now you are going to face something logical: competition in the sector. Although they do not work in the same way and perhaps they do not share characteristics, DLSS will face on the one hand FidelityFX SR, on the other Radeon Boost and finally Microsoft DirectML. Are they compatible with each other? And above all, which one is better?
Rumors suggest that soon, AMD will launch a new driver suite that promises two new features and updates: FidelityFX SR and Radeon Boost, which although the latter is already enabled in its drivers, would come with improvements to boost performance. NVIDIA came first, hit before and now it has to compete, but can it do it with its eternal rival and against Microsoft?
One standard or four different? Where is the video game industry going?
This is really the question we must ask ourselves, where are the players going and where is the industry looking? And it is that with the panorama that we are presented with, plus some options that are loose in the industry, we are going to have to face DLSS vs Fidelity FX SR vs Radeon Boost vs DirectML.
As we have mentioned, they all have their peculiarities, but at the same time they all do the same in essence: scale the resolution to improve image quality. So why is each company developing its proprietary technology and not being offered as an open standard or joining forces under one?
It is a great question to which we have no answer, but surely since it is not an API as such, but is a mode of rescaling with AI, each company intends to implement it in the engines of the main developers with the excuse of exclusivity for their hardware.
NVIDIA DLSS 2.0
We have talked at length about DLSS, so we will summarize quite a bit for those reasons. It’s a technology that uses NVIDIA’s Tensor Cores AI processors to boost both performance and graphics quality.
All through a deep learning neural network, which focuses on four main improvements over its predecessor: superior image quality, great scaling across all RTX GPUs, a neural network for all games, and the ability to get options customizable.
In summary, NVIDIA is the one that develops and maintains it, the implementation is carried out by the game engine, it has a selected game support that can enable it, it is Full Frame and is only available for the Volta, Turing, Ampere architecture and successive.
AMD Radeon Boost
Although it is not the direct competitor as such, it is true that Radeon Boost overshadows the NVIDIA pool. The problem that AMD has is that the form of implementation is through the company’s Adrenalin driver , where for more problems the games enabled for this technology are not so abundant.
In addition, and due to its nature, if we analyze it carefully we will see that although it is Full Frame, its way of working is totally dynamic, although it is capable of supporting architectures from GCN to RDNA and higher.
The problems to compare it with NVIDIA is that it really works worse because it is supported by the driver and it is the one that does the work and the dynamic range that in many cases makes it lose quality instead of gaining it.
It was the last to arrive as such and it is no less important for that. Unlike the two rivals, it has more “open” characteristics, and its implementation requires only Windows 10, the game or the driver having nothing to do with the performance or optimization of said technology.
In addition, and although it is a Full Frame technology, it has the drawback that there is no game that supports it, which is curious since the developers would have to make minimal changes to support it.
In addition, you are curious that it will be compatible with all GPUs that support DX12 , without discriminating models or architectures, which is a great advantage and could be postulated as the reference in the sector.
AMD FidelityFX Super Resolution
It is the last to arrive, to the point where it remains a mystery to this day, but we do know several things. In the first place, it will be a technology that will be implemented in each game that supports it, which leaves this type of implementation in the hands of the developers.
It will be the same as DLSS, a Full Frame technology and although we do not know what architectures it will support, it is more than likely that it will be RDNA forward. But what sets it apart from Radeon Boost? Well, FidelityFX SR does use Deep Learning algorithms and it will require graphic hardware to move it, but this point is not clear.
There is a lot of controversy here, since as we do not know how AMD is going to move the hardware resources, it is speculating that FidelityFX SR could be based in some way on Microsoft DirectML, but always within the software package that receives its name: FidelityFX.
This results in that today we do not know if AMD is following in NVIDIA’s footsteps, if it shares a path with Microsoft or if it undertakes its own. Everything indicates that it will be the last and that the company will have to assign software engineers so that the developers know how to implement it correctly in their games, since if it does not happen to NVIDIA when DLSS comes out, which was a rather notorious failure.
So which one is better?
Well, it is an unanswered question if you have been attentive in reading. The only reality are DLSS 2.0 and Radeon Boost, but the latter does not use Deep Learning and only changes the resolution by driver.
DirectML is very green and Microsoft support projects seem to be going slowly but surely, although at the moment there is nothing on the market that is of the stature. For its part, NVIDIA came first, struck first with DLSS, and now has a mature technology that works and to work on. Developers are increasingly implementing their technology in engines, so for now it is the company and technology that is working the most.
We’ll see if FidelityFX SR can overshadow it …