GPUs in recent years are having a great impact in many more areas than in games as such. For years, AI and its learning have been helping a huge number of sectors, expanding its potential throughout the world. But as we well know, AI has to be trained and to speed up tasks the GPUs have special units that speed up the work, which together with software like Microsoft DirectML can enhance any environment, but what exactly is this software?
Before getting into the whys, you always have to know exactly what you are talking about. For this reason, when it comes to talking about Windows 10 and AI, GPUs and super-resolutions, we must talk about software-based technologies like the one we are dealing with today, what is it, what is it used for and what is its direct competition?
Microsoft DirectML, the standard for accelerating artificial intelligence
Microsoft DirectML is the acronym for Direct Machine Learning and is part of a suite of software developed by the company as a low-level API for machine learning. It is a very simple idea that is quite complex when developed, since everything works based on what is called “primitive calls”, which according to Microsoft are grouped into blocks.
The idea of the blocks was developed by the fact of being able to computerize large amounts of more or less simple data that allow working with large volumes of data. Unsurprisingly, DirectML is part of DirectX 12 and has a core API oriented workgroup and programming interface.
The good thing about this software is that it allows from training machine learning models for any GPU that is compatible with DirectX 12 as such, giving results both in photography applications to Deep Learning, so in this sense there are no limits. DirectML was introduced in Windows 10 since version 1903 and has been updated several times, but at the moment it is not the standard that Microsoft is looking for to master all the inference fields.
What is Microsoft’s competition?
It is the big question, because unlike its rivals, DirectML is much more complete than these and above all, it is a standard, which implies that it can be used for a multitude of scenarios with the same language.
But in gaming the options are much more concrete and concise. NVIDIA has DLSS in its version 2.0, which works really well and is highly appreciated by programmers and game engines. AMD has not released any version, but it does have it almost ready at the moment thanks to its FidelityFX technology, from which it receives the name as FidelityFX SR , where obviously the SR tag comes from SUPER RESOLUTION.
As the options are not yet ready on the market, there is no possible comparison and, having seen what has been seen with NVIDIA and DLSS, it will take a few years to be able to compare the three technologies that are going to define the supersampling jumps in the same game.
