DSR/Downsample Guide: Is 1620p a viable alternative to 2160p/4k?

The Witcher 3 Blood and Wine Toussaint Ultra Widescreen Panorama
Click this image to experience its full size glory.

You may ask yourself this question because a 2160p/4k resolution is not an option for you. Let it be your personal preference of playing at 60fps and the lack of hardware power to maintain this level of performance or let it be your hardware’s lack of performance to even maintain stable 30fps at 4k resolutions. So, what alternatives do you have in order to increase your image quality beyond 1080p/FullHD?

The answer seems to be simple. Just increase your resolution (via Nvidia’s Dynamic Super Resolution or AMD’s Virtual Super Resolution) until you hit the sweet spot between image quality and performance and be done with it.

DSR Nvidia Control Panel

This is true in principle, but because of how upscaling works you could end with even worse image quality* than with 1080p. Why is that?

Every other resolution than your monitor’s native res—if it’s a modern LCD/LED monitor—has to be interpolated. Think of playing a game in a sub 1080p resolution being scaled over your whole screen. It’s a blurry mess and gets worse the lower you go in res. Now, on PC, you can change the scaling behavior to ‘do not scale’, but you get a smaller ‘window’ that is (almost) pixel perfect and crisp. It’s small, however.

DSR/VSR give you the option to ‘downsample’ a picture to your native resolution and that comes basically with great benefits as downsampled pictures are built with more (source) information, simply spoken. It is scaled nonetheless and if you don’t use a resolution (from that you downsample to your native resolution) that scales to a shared divisor (or an even multiple, respectively), your image quality will suffer.

Why is that? Imagine you have an object on the screen (1920×1080) that takes up the size of four pixels. Two in horizontal direction and two in vertical (2×2). What happens when you double your horizontal and vertical resolution to 3840×2160? Your object on the screen gets scaled accordingly and is now rendered with four pixels horizontally and four pixels vertically—thus with more details.

What happens when you increase your resolution to 2560x1440p? 2560 horizontal pixels are one and a third of 1920, so if you multiply that with the two pixels of your object you get 2,67 pixels… and this is the point where you should see the problem. It’s either 2 or 3 pixels and the ‘missing information’ gets blurred; or jaggy like an un-antialiased image.

This is where Nvidia’s ‘DRS smoothness‘ comes into play. It smooths-out the uneven edges from uneven (multiplied) pixels with a Gaussian filter; its default setting at 33%.

DSR Smoothness Nvidia Control Panel

What does that mean for you, the user, in a practical sense? There is no such value for ‘perfect’ DSR smoothness, scientifically. It depends on the resolution you use in your game, on your personal preference and should be set accordingly. What works with 20% DSR smoothness for you and at 1620p may not work for another person and/or at 1440p. Therefore, the rule of thumb should be to choose a resolution your PC can handle in most games and can maintain your desired frames per second and set the smoothness respectively.

If you can maintain a 4k resolution (or, to be more precise, 4x of your total native resolution’s pixel amount), the best option is to set the DSR smoothness factor to zero altogether, deactivating it completely.

Should you even use super 1080p resolution, given the tradeoffs and imperfect image? In short: yes, you should! It’s not only about image quality, antialiasing and pixel density. It is too about added details that come with greater resolutions. As I mentioned earlier, simply spoken, higher res carries more information and some of that remains even if the image gets downsampled afterwards. Especially tiny details, in distances, that can barely be distinguished because the image doesn’t count enough pixels, could be improved with a higher-than-native res. See this example of Dishonored 2 below:

Dishonored 4k comparison with added details over 1080p

Dishonored 4k comparison with added details over 1080p
You can open these screenshots in a new tab to zoom in.

If you look at small details like fences at the top right of the images, the shadows at the roof tops, the anchor at the left of the images, you see details that are simply absent or very hazy in the second screenshot.

In conclusion, if your hardware allows it and if you find the sweet spot of DSR smoothness—in case you won’t use an even multiple of your monitor’s native resolution—downsampling is the way to go for increasing image quality and adding details to game’s scene.

Nvidia’s DSR offers option like ‘1.78x’ or ‘2x’ and what does that mean? It counts the total amount of pixels respective to your native resolution. 1920×1080 are 2 073 600 pixels. If you double the vertical pixels and double the horizontal pixels—to 3840×2160—you have four times the amount of pixels: 8 294 400. You could, then, think ‘2x’ is an even multiple and, therefore, should be fine! Thing is, if you multiply 2 073 600 with two, you have 4 147 200 pixels. Since there is no 16:9 resolution for this amount, you get a resolution like 2715×1527 which is not exactly 4 147 200. And it’s not exactly 16:9, either. This will mess up your image. If you choose ‘2.25x’ however, which are 4 665 600 pixels, you get a smooth number of 2880×1620.

This should be fine, shouldn’t it? Based on the last paragraph’s numbers, yeah. But think of my example with the 2×2 object. If you now use the 2880×1620 resolution, you multiplied each your horizontal and vertical res with 1.5. What happens with our 2×2 object? Right, it becomes 3×3. What if it is 5×5 object? 7,5×7,5 is, again, problematic. 1.5 is still an uneven number, after all. And that’s why you basically need a post-processing filter like Nvidia’s Gaussian blur that is questionable of course in terms of its quality and whether Nvidia should give their users more option to choose a resampling filter since there are more methods than Gaussian.

I hope this guide gave you an understandable heads-up of DSR/VSR/downsampling but, more importantly, what your best options are and how to get the most out of your image regarding its quality, cleanness and general appeal to your personal preferences.

* We use the term ‘image quality’ at many occasions. In case you’re not familiar with the difference between image quality and graphical fidelity: The former describes the impression of the rendered image according to common quality properties, as in artefacts, color bending, aliasing and flickering (cf. image quality of a 100% JPEG or highly compressed one with e.g. 50% JPEG quality). Graphical fidelity describes the visual effects and graphical features of a game, as in texture resolution and quality, shadows (e.g. shadow map size, drawing distance), lighting etc. pp. It is usually untouched by a game’s resolution and therefore independent of image quality. There are occasions where graphic features—like shadows or geometry details—scale with the resolution to a certain degree (cf. Dishonored 2 screenshots above), hence increasing graphical fidelity. It’s not the main purpose, however.

Kommentar verfassen