top of page

AMD FSR: An enticing competitor to DLSS

  • Writer: Sriyashas Kalluri
    Sriyashas Kalluri
  • Jun 27, 2021
  • 2 min read

During AMD’s phenomenal keynote at computex 2021, one of the things they announced is here. AMD FSR released on June 22nd as an open source upscaling technology.


A Brief Overview of FSR

Image From AMD's Website


FSR or FidelityFX Super Resolution is AMD’s response to NVIDIA’s DLSS(Deep Learning. What these both do is render the image at a lower quality (eg: 1080p) and upscale it to whatever desired resolution, hence granting higher framerates in supported games. While DLSS is only supported on NVIDIA’s RTX series GPUs, FSR follows AMDs trend of making their software open source and more widely supported and runs on older models. Even older NVIDIA GPUs which don’t support DLSS can run FSR.


FSR works based on a spatial upscaling algorithm in contrast to DLSS’s reliance on machine learning and complex temporal algorithms.


Most people expected FSR to be somewhere between the horrible failure called DLSS 1.0 and the successful DLSS 2.0. One major reason people felt that FSR would fall short of DLSS 2.0 was that while DLSS is supported only by NVIDIA’s new RTX graphics cards (released in 2019) using their proprietary and expensive ‘tensor cores’, FSR has a wider support base. It can be used by almost all GPUs(even integrated ones). One user reported it working on a NVIDIA GTX 650 (from 2012), a GPU not even supported by NVIDIA DLSS.


This is a breath of fresh air, especially considering the current market situation and how hard it is to get any video card.


How FSR Works


Image from AMD Computex 2021 Keynote


FSR is based on a spatial upscaling algorithm as mentioned before. What that means is that it renders the image at a lower resolution than what is to be displayed and then attempts to fill in the details and sharpen the edges.

Image From AMD keynote(Computex 2021)


It’s integrated into the graphics pipeline as a shader, so the hardware it runs on doesn’t matter. Only the image is processed and the HUD and other elements are added at the end. This approach allows FSR to be much simpler than DLSS and does not require any of the extensive game specific support needed for DLSS.


BenchMarks


FSR is supported by 7 games right now, none of which support DLSS, so no direct comparisons as of now.


FSR starts out with stellar results, with an average of around 30% - 40% increase with ultra quality which scales up to around 75% with balanced. That’s a lot of performance gains, especially for open source software.


Even the visual quality doesn’t suffer a big hit, with some softer edges and shimmering artifacts visible only at the performance and balanced modes.


I won’t go extensively into the benchmarks as I don’t have a sufficient list of first party comparisons.


Conclusion

AMD has really done well (I’m saying that a lot, I know). The numbers are on par or sometimes even surpassing DLSS. I really love what AMD is doing here by making everything open source and helping all gamers out there. So keep up the good work AMD.








Σχόλια


bottom of page