Posts: 2,046 +234
TL;DR: The ultimate question that most gamers have been asking is: is DLSS or FSR "better" That's a near impossible question because there are lots of factors to consider: visual quality, performance, artifacting, specific game support, and so on. Neither technology will be the best in all those categories. But in the area that perhaps matters the most – the balance of visual quality and performance – generally speaking DLSS 2.0 is better, though to the degree it’s better varies significantly depending on the game and playing conditions.
We've covered DLSS in-depth a few times as it's progressed and improved, and we recently took an early look at FSR, so without rehashing much about how they work, you should know these are two fundamentally different technologies attempting to achieve the same outcome: improve performance without a significant loss to visual quality.
AMD’s FSR is the more simple technique. It’s a spatial upscaling technology that combines edge detection upscaling with a sharpening pass. Nvidia’s DLSS on the other hand is more complex, incorporating temporal data including prior frames and motion vectors, and uses an AI network to assist with the image reconstruction process. Both require games-specific integration, ideally before final post-processing and UI effects are added, and both feature several modes in their standard configuration.
There are several games now supporting both techniques, and the two we're focusing on in this video are Marvel’s Avengers and Necromunda Hired Gun. Avengers is developed by Crystal Dynamics and uses their in house Foundation Engine, similar to the prior Tomb Raider games they have worked on. Meanwhile, Necromunda is a game from Streum On Studio set in the Warhammer 40K universe that uses Unreal Engine 4. The integration of FSR in Avengers came as a bit of a surprise as it wasn’t a title AMD listed in their launch presentation.
Both titles use the first iteration of FSR which AMD are calling FSR 1.0, and they are making it pretty clear in their documentation that they’d like people to use “FSR 1.0” branding in their games. Meanwhile, both of these titles use a variant of DLSS 2.0.
This is the kind of test that has both a performance and image quality components, so check out the video above for all those. We’re mostly interested in which modes of DLSS and FSR perform the same, rather than comparing both at the same base render resolution, which is not that relevant for real world usage outside of “for science” investigations.