Answer:
The best way to figure out which type of AA (Anti-Aliasing) works the best for you is to simply try every AA available until you find the sweet spot between the image enhancement and tolerable performance hit.
We’ll go into detail about all popular anti-aliasing types to give you a better idea of what each of them does.
Are the graphics in your video games jaggy or suffer from the staircase effect? Can you see the rough edges instead of a round shape? Don’t worry, there are ways to improve this.
The prominent jaggies appear due to the low resolution, so the most effective way to get rid of it is by getting a higher resolution monitor.
In fact, on a 27″ 4K monitor, for instance, the image is so crisp and sharp that you most likely won’t even need any sort of AA.
However, if you can’t afford a new monitor or the PC gear required to run higher resolutions, you can use one of many anti-aliasing techniques available to improve the image quality.
Some are more efficient than others but usually come at a higher performance cost resulting in FPS (Frames Per Second) drop.
Different Anti-Aliasing Types
There are two main groups of anti-aliasing.
The first group we’ll get into increases the sample rate by rendering more pixels than the screen actually needs and then down-samples it to your resolution.
Alternatively, the second group of AA types blurs the rough edges on the screen after the rendering process. Since this version of AA is post-processing, there’s only a tiny impact on your performance, but the image quality improvement is also less noticeable and can appear too blurry, especially with fast motion.
So, if you can afford to sacrifice performance for the picture quality, you will likely opt for some sort of sampling anti-aliasing.
If every frame per second is precious to you, then you will have to settle for a post-processing AA, which may slightly blur the image, but at least it will eliminate the unpleasant staircase effect.
Super-Sampling Anti-Aliasing
First off, we have the SSAA (Super-Sampling Anti-Aliasing) or FSAA (Full-Scene Anti-Aliasing), which were the first forms of AA to be available and still deliver the arguably best image quality enhancement but with a substantial performance cost.
A more popular AA is the MSAA (Multi-Sample Anti-Aliasing), which only applies the sampling to the edges, where it is most needed, and will thus save you the performance cost while still noticeably improving the image quality.
Moreover, NVIDIA and AMD have their own propriety versions of MSAA. Both Nvidia’s CSAA (Coverage-Sample AA) and AMD’s EQAA (Enhanced Quality Anti-Aliasing) are derivates of MSAA and function pretty much the same way.
Post-Processing Anti-Aliasing
The most common post-processing AA is FXAA (Fast-Approximate Anti-Aliasing). As previously described, enabling this will blur out the annoying jaggies with a minimal performance cost.
Overall, this peculiar anti-aliasing algorithm may repulse many gamers due to the blurry image. Still, if you have limited system horsepower, it is likely your only choice as it’s the best anti-aliasing method for performance.
AMD’s MLAA (Morphological Anti-Aliasing) is similar to FXAA, you get a blurry picture at a tiny performance cost, but it smoothes out the rough edges.
Combined Anti-Aliasing
NVIDIA’s TXAA or TAA (Temporal Anti-Aliasing) combines MSAA (down-sampling) and post-processing (blurring) with temporal filters for an overall better outcome. However, you will need a graphics card based on NVIDIA’s Kepler GPU, a GTX 600-series or higher, for this technology.
Naturally, TXAA also requires more power than the standard FXAA as it handles fast in-game motion much more efficiently.
Another in-between solution is SMAA (Sub-Pixel Morphological Anti-Aliasing), which is the middle ground between FXAA and MSAA regarding both image quality and performance cost. Basically, it works like FXAA but also detects the edges to smooth them out specifically.
Then we have CMAA (Conservative Morphological Anti-Aliasing), which is a combination of FXAA again and SMAA. So, you get less blurring than with FXAA, but a softer image than SMAA while the performance cost is in-between the two.
Super Resolution
Lastly, depending on your graphics card, you may also want to try out NVIDIA’s DSR (Dynamic Super Resolution) or AMD’s VSR (Virtual Super Resolution).
This technology allows your GPU to render the screen up to 4K (depending on the max resolution of your monitor) and then down-sample it to your monitor’s native resolution.
NVIDIA’s newer DLDSR (Deep Learning Dynamic Super Resolution) does the same thing but uses AI for better results.
Additionally, this isn’t only great for gaming as you can increase the resolution of your desktop as well, which will give you more screen real estate.
AMD FidelityFX Super Resolution (FSR)
Inversely to VSR, AMD’s FSR technology makes the GPU render at a lower resolution and then uses its upscaling algorithm in order to provide you with a higher frame rate.
The trade-off between how much extra FPS you get and how much the image quality is affected is up to you as you can choose between different presets (Ultra Quality, Quality, Balance and Performance). The ‘Performance’ mode offers the highest FPS gain at the biggest toll on image quality, while ‘Ultra Quality’ has the lowest FPS gain but also has the smallest impact on picture quality.
AMD’s FSR is open-source, but in order to use it, the game must support it and you will need a compatible GPU (GTX 10-series or AMD RX 460, or newer).
You can find a list of all games (including upcoming titles) that support FSR here.
Some modders have also succeeded in adding FSR support to games that don’t support it natively, for instance, GTA V.
FSR 2
AMD’s FSR 2.0 uses temporal upscaling instead of spatial (like FSR 1.0).
FSR 1.0 is most beneficial at high resolutions, such as 4K UHD, and when using the ‘Ultra Quality’ or ‘Quality’ presets, whereas FPS 2.0 offers better image quality and performance at 4K, but also great results at lower resolutions and with modes that favor performance.
For optimal performance, AMD recommends at least RX 5700 or RTX 2070 for 4K UHD, RX 5600 or GTX 1080 for 1440p, and RX 590 or GTX 1070 for 1080p. However, FSR 2.0 is still supported on all RX 500-series and GTX 10-series or newer GPUs.
The newer versions, AMD FSR 2.1, 2.2, etc. further improve the upscaling performance, mainly by reducing the amount of ghosting artifacts.
FSR 3 & AFMF
AMD’s FSR 3 is a frame-generation technology, which creates interpolated frames between the normally rendered frames. As a result, you get a higher frame rate for smoother gameplay, but input latency is not improved, which makes this technology impractical for competitive gaming.
What’s more, FSR 3 can introduce various visual artifacts and for the best results, you’ll need a base frame rate of at least 60FPS (ideally 100FPS+). At the moment, it’s only available in a few titles.
Some games with FSR 3 support also have a new ‘Native AA‘ mode, which doesn’t use any upscaling, just anti-aliasing to improve the image quality (better than native) at a small performance cost (~10%).
AFMF (AMD Fluid Motion Frames) is also a frame-generation technology, but it’s driver-based, allowing you to use it in games that otherwise don’t support FSR 3 (or DLSS 3.0).
AMD recommends a base frame rate between 55 and 70FPS for this technology. However, at the moment, the technology isn’t very usable as the generated frames are blurry in addition to higher input latency.
Just like with FSR 3, we expect AFMF to improve with future driver updates.
AMD Radeon Super Resolution (RSR)
AMD’s RSR technology uses the same algorithm as FSR 1.0, but it’s driver-based, allowing you to use it in any game in full-screen mode. However, it requires an AMD RDNA-based GPU (RX 5000 series or newer).
NVIDIA DLSS 2.0
Just like AMD’s FSR, NVIDIA DLSS 2.0 (Deep Learning Super Sampling) makes your GPU render at a lower resolution, but it uses artificial intelligence for better upscaling. So, you get a higher frame rate with a lower impact on image quality in comparison to FSR.
However, DLSS 2.0 is not open-source. In order to use it, a game must support it and you need to have a compatible NVIDIA GPU with Tensor Cores (RTX 20-series, or newer).
Newer versions of DLSS (2.1, 2.2, etc.) are also available. They work on the same principle but offer improved performance and reduced visual artifacts.
You can still find the older DLSS 1.0 version in some games and while it can sometimes improve performance, the impact on the image quality is usually not worth it.
Just like DLSS 2, DLAA (Deep Learning Anti-Aliasing) uses machine learning to improve image quality but doesn’t use upscaling. So, you get a better image quality than native with a performance cost.
DLSS 3.0
With DLSS 3, 3/4 of the first frame is upscaled via AI, while the second frame is reconstructed via DLSS Frame Generation for an even higher gain in frame rates in comparison to DLSS 2.
Although you get higher FPS, frame generation actually increases the input lag as DLSS 3 requires data from the ‘next’ frame. So, as a result, you get smoother motion, but not as responsive gaming experience. The generated frames can also cause some visual artifacts.
If you have a high base frame rate (~120FPS), DLSS 3 can boost it up to ~200FPS. In this case, the visual artifacts and the added input latency aren’t that noticeable, but if your base FPS is below 100FPS, DLSS 3 isn’t very useful – you get a higher frame rate, but the responsiveness is noticeably worse and you get visual artifacts.
Furthermore, DLSS 3 cannot be used with V-Sync or capped frame rates, so it’s mostly suited for high refresh rate (240Hz+) displays with VRR. And due to the increased input lag, it’s not useful for competitive gaming.
In some games, you can use DLSS 3 with capped frame rates, allowing you to prevent your FPS from exceeding your monitor’s maximum refresh rate.
DLSS 3.5
While the original DLSS and DLSS 2.0 are super-resolution technologies, and DLSS 3.0 is a frame generation technology, DLSS 3.5 is a Ray Reconstruction technology. Further, DLSS 3.0 only works on 40-series or newer RTX GPUs, while DLSS 2.0 and 3.5 work on all RTX graphics cards.
DLSS 3.5 improves ray-traced images by replacing the regular game denoisers with denoising processed by GPU’s AI tensor cores. At the moment, DLSS 3.5 is only available in a few games, such as Cyberpunk 2077, using the game’s demanding path tracing mode, so it’s viable if you have a high-end GPU, such as the RTX 4080 or 4090.
The newest DLSS 3.7 version features a new Preset E DLSS scaling that eliminates smearing artifacts visible in some games.
Intel XeSS (Xe Super Sampling)
Intel’s super-sampling technology is called XeSS and it has two modes: DP4a, which is supported on most GPUs and XMX, which only works on Intel dedicated graphics cards and offers better results.
The latest version (XeSS 1.3) improves performance and image quality, and also adds more presets.
While it’s good that there are more presets to choose from to find the ideal ratio between image quality toll and FPS gain for you, the preset names bring confusion when comparing them to AMD and NVIDIA’s presets with the same name.
For instance, FSR Balance uses 1.7x resolution scaling, DLSS Balance also uses 1.7x, but XeSS Balance now uses 2.0x scaling. So, when comparing all three technologies, you’d need to use XeSS Quality in this case for 1.7x scaling, which is unnecessarily confusing.
DLSS 3.7 vs XeSS 1.3 vs FSR 2
While all three technologies provide you with roughly the same FPS gain at the same resolution scaling, there are big differences in image quality. In most cases, DLSS offers the best image quality, followed by XeSS XMX, then XeSS DP4a and lastly, FSR, which often has the worst image quality.
Also, note that in order to use the latest version of super sampling (if it’s not natively supported by the game), you’ll need to manually download the latest version and replace the .dll file in your games. File location and file name will vary depending on the game and the super-sampling technology.
Conclusion
So, what is the best option for you?
There are many more types of AA, such as NVIDIA’s SGSSA and OGSSAA, which need more complex driver editing and adjusting.
For games that don’t have any AA or not the ones you need, you can download software such as Reshade, which can inject certain anti-aliasing methods, including SMAA, among others. Or simply use NVIDIA’s or AMD’s super-resolution feature if your GPU supports it.
Hopefully, after reading this, you’ll have a better idea of what to do in the anti-aliasing section of the video settings in your games.