An overview of Nvidia’s DLSS 4.5 technology and what it means for PC gamers.
The PC graphics market might be in peril in hardware terms, with prices spiraling and availability inconsistent, but on the software side things are at least reliably moving forward – with Nvidia today announcing the latest upgrade to its DLSS technology.
DLSS 4.5 is shadow dropping today, and is already available across a suite of hundreds of compatible PC games. It’ll offer performance enhancements to a variety of games across a wide gamut of Nvidia gaming hardware.
If you’re among the uninitiated, DLSS stands for Deep Learning Super Sampling, and is an AI-powered piece of tech aimed at increasing game image quality and performance. This is, it has to be said, one of the more readily accepted pieces of AI used in and around video games – used to stretch your hardware further rather than replace the work of any humans.
The way it works is complex, obviously, but it can be summed up fairly simply: DLSS initially renders a game at a lower resolution – for the sake of this example let’s say 720p rather than 1440p. Rendering at a lower res in turn allows the game to display at higher frame rates thanks to the lowered visual fidelity. But then that super sampling kicks in – using a custom-trained neutral network to upscale the 720p output image to your target resolution (in this case, 1440p). The result is you get the 720p frame rate, but you get a visual fidelity and image quality that is very close indeed to a native 1440p output.
DLSS is Nvidia’s proprietary version of this tech, but other versions exist. PlayStation 5 has PSSR, its equivalent. AMD PC GPUs have FidelityFX Super Resolution. Nvidia has always sat at the cutting edge of this technology, however – and with an Nvidia chip inside, the Switch 2 actually uses a version of Nvidia’s DLSS too – so upgrades to DLSS have potential implications there.
This tech is useful at a base line, but it can also be especially handy in games that make use of graphical features that absolutely tank performance, like real-time ray tracing. In the end, there is always an argument to be made about untouched rendering – raw rasterisation – versus these ‚processed‘ images, and if a ‚pure‘ and pristine frame is better.
