Nvidia just launched its latest ICAT utility to showcase the image quality of its DLSS algorithm, specifically looking at how it compares to AMD's FSR. We've discussed FSR vs. DLSS in more detail elsewhere, but the new tool (ICAT stands for Image Comparison and Analysis Tool) provides some new options for comparing image quality. Nvidia's hope is to showcase how much better DLSS looks… but it's not quite an open-and-shut case.
The real problem seems to be rather wildly varying implementations of FSR. We noticed this in our own discussion of the two techniques, specifically calling out Deathloop as having rather poor results for FSR. But it's one thing to take still screenshots and compare the two resulting images. What happens when you look at things in motion? That's something ICAT helps to provide.
It's of course possible to clip together video segments using Adobe Premiere or some other video editing utility, but ICAT helps to streamline the process. You can shift each video sequence one frame at a time to get things to line up (as closely as possible), and when you zoom in, it applies the same zoom parameters to all videos. You can pause and pan around, and you have the option of a split-screen mode where you compare just two videos, or a multi-video side-by-side mode showing all the videos at the same time.
The result is pretty slick, though trying to convey all that via internet video can be problematic. That's because YouTube recompresses any uploads, making slight differences in image quality much harder to see. Even with the original video captures, it's still not going to provide a perfect "what you see on your monitor" experience, unless you have a hardware capture card recording uncompressed video files. Disclaimer: We do not have that (at least not yet), which means we had to use Nvidia's ShadowPlay (part of GeForce Experience) or AMD's equivalent to capture the videos. And since we wanted to compare DLSS with FSR, that meant using ShadowPlay exclusively for this initial look.
We captured the same sequence in a couple of games at 4K and 50Mbps bitrates using ShadowPlay, then imported the results into ICAT and set about lining things up. Nvidia also provided some video footage, which we elected not to use (though the above Chernobylite screenshot capture used Nvidia's images). The important bit is to start with a game that supports both FSR and DLSS. Here's the short list of current games with both technologies available: Back 4 Blood, Baldur's Gate III, Call of Duty Vanguard, Chernobylite, Deathloop, Edge of Eternity, Enlisted, F1 2021, Marvel's Avengers, The Medium, Myst, and Necromunda: Hired Gun.
We only checked out Back 4 Blood and Necromunda for this first look, since capturing and lining up all the videos still takes quite a bit of time and effort. Here's a look at ICAT, running at 25% speed playback — because our PC couldn't handle real-time decoding of five separate 4K streams — along with some pausing and zooming in on some of the details. We'll show the split-screen and side-by-side modes in Necromunda, and just the side-by-side mode in Back 4 Blood. Note the filenames shown below the clips, where it indicates whether the video is native, DLSSQ, DLSSP, FSRUQ, or FSRP. We've also added text to make it clear which video uses what mode.
We've uploaded the videos in the original 4K resolution we captured from our desktop — again, using ShadowPlay — but YouTube can take a while to process the content. Ideally, you should watch the content in fullscreen mode on a 4K display, though differences should still be visible on a 1440p or 1080p monitor. You can pause the video as needed, though again, YouTube can mangle the quality with its higher compression ratios.
When you watch the videos, you'll see that while the quality does vary, it can be a bit difficult to say which is universally "better" — well, except FSR Performance, which clearly looks worse (and the above captures still don't convey the whole story). Part of the problem is that I opted to capture the videos at 4K. That means even in the performance modes of both DLSS and FSR, the algorithms have 1920x1080 data to work with. That's enough to get a reasonable upscaling result. If you start at 1080p native, upscaling 960x540 starts to show more artifacting. Anyway, the above are at least another interesting comparison point.
The image quality results of DLSS and FSR can vary quite a lot, depending on the game and what area of the game you're looking at. Nvidia's DLSS tends to do a better job with anti-aliasing, at the cost of sometimes looking quite blurry. FSR in contrast uses a sharpening filter and can at times get a halo sharpening effect. Nvidia also provides a sharpening filter via its drivers, but you can't (currently) capture the resulting content via ShadowPlay – it requires a separate capture card – so it's possible to make DLSS look a bit less blurry if you desire.
Nvidia's ICAT software is available as a free download if you want to try it out. I believe it requires an Nvidia graphics card to run, though I'll check that in a moment to confirm. Obviously, you need an Nvidia GPU if you want to capture DLSS videos for comparison.
You may be interested in:
>> Is a Chromebook worth replacing a Windows laptop?
>> Find out in detail the outstanding features of Google Pixel 4a
>> Top 7 best earbuds you should not miss
Related Posts:
>> Recognizing 12 Basic Body Shapes To Choose Better Clothes
>>Ranking the 10 most used smart technology devices
>> Top 5+ Best E-readers: Compact & Convenient Pen
0 Comments:
Post a Comment