Why Do Game Edges Still Look Jagged? SSAA, MSAA, FXAA, SMAA, TAA & DLAA Explained — Mechanisms & Trade-Offs Guide
Key Takeaways
Game edges look jagged because a display is a grid of square pixels while game geometry is continuous, so a diagonal edge forced into whole pixels becomes a staircase. Different anti-aliasing methods attack different kinds of jaggedness, and no setting fixes them all. This guide compares six methods by mechanism and shows how to diagnose which kind you are seeing.
A diagonal line in a game rarely looks like a line. It looks like a staircase, each step exactly one pixel wide, climbing across the screen in small square jumps. Turn the camera, and the staircase shimmers. Open the graphics menu, turn the anti-aliasing setting up as high as it goes, and the staircase may barely change, or the whole picture may go soft, or a moving object may leave a faint trail. The natural conclusion is that something is broken: the monitor, the cable, the driver. In many cases, nothing is broken. The staircase is the visible result of sampling a continuous world onto a square grid, and different anti-aliasing methods repair different parts of that problem. Asking which setting is strongest is therefore the wrong first question. The right first question is: which kind of jaggedness am I actually looking at?
The Root Cause: A Continuous World on a Square Grid
Picture graph paper. A diagonal line drawn across it must pass through square cells, and each cell can only be coloured one way: inside the line, or outside it. There is no half cell. A game faces the same constraint at far higher resolution. The three-dimensional scene is continuous geometry, but the display is a grid of discrete square pixels. When a renderer decides what each pixel shows, it typically samples one point inside the pixel and asks which object that point belongs to. If a diagonal edge passes through a pixel, the whole pixel takes one side or the other, and the edge emerges as a staircase. That staircase is geometric aliasing, and it is the first of the two kinds of jaggedness in this guide.
The staircase also explains why the problem is worst on long diagonals and curves, and why straight horizontal and vertical edges never show it. A horizontal edge runs along the grid, so it aligns with whole rows of pixels. A diagonal edge crosses the grid at an angle, so it must be approximated, and the approximation is visible wherever the contrast across the edge is strong: a dark roofline against a bright sky, a pale railing against a dark wall. Anti-aliasing, in all its forms, is an attempt to soften that forced choice, either by sampling more carefully or by blending the result more cleverly.
The Second Kind of Jaggedness: Shimmer
There is a second kind of jaggedness, and mistaking it for the first is the most common reason anti-aliasing settings seem not to work. Even when geometric edges are handled well, fine detail smaller than a pixel can still flicker. Dense textures, transparent objects such as wire fences, foliage, and railings, and sharp specular highlights all contain detail at frequencies higher than the pixel grid can represent. From frame to frame, as the camera moves, a given pixel sometimes catches that detail and sometimes misses it, so the detail flashes on and off. In motion, that reads as shimmering or crawling.
This shimmer is not a geometric edge problem, and geometric sampling methods cannot cure it. That sentence is the dividing line of the whole guide, and it will be repeated in the method descriptions below, because every disappointment with anti-aliasing traces back to applying a geometric remedy to a shading problem, or the reverse. A still screenshot can also hide the distinction: a method that produces a beautiful still image may shimmer badly in motion, and a method that looks slightly soft in a screenshot may be rock steady in motion. Still and moving judgements must be made separately, a point the diagnostic sequence returns to.
Six Methods Compared
| Method | Stage | Sampling or post-processing | Which jaggedness it treats | Main weakness | Direction it suits |
|---|---|---|---|---|---|
| SSAA | During rendering, whole scene | Sampling: renders above output resolution, then averages down | Geometric edges, textures, and highlights together | Highest cost of the group, in relative terms | Older games, or when performance headroom is ample |
| MSAA | During rendering, edges only | Sampling: multiple coverage samples on geometric edges, shading usually computed once | Geometric edges only | Leaves texture, highlight, and transparent-object shimmer untouched | Sharp geometric edges where performance still matters |
| FXAA | After rendering, on the finished picture | Post-processing: finds high-contrast edges and blends them | Geometric edges, approximately | Can soften fine text and texture detail along with the edges | Lowest cost, when performance comes first |
| SMAA | After rendering, on the finished picture | Post-processing: recognises edge shapes and blends selectively | Geometric edges, with better detail preservation as its design goal | Still a single-frame guess, without true scene knowledge | Post-processing quality with less whole-picture softening |
| TAA | Across frames | Temporal sampling: jitters sample positions each frame and blends aligned history | Geometric edges and much shimmer, in motion | History errors appear as ghosting, trailing, or softness | Stable motion, where shimmer is the main complaint |
| DLAA | During rendering, at native resolution | Model-based, in the DLSS family: anti-aliasing applied directly at native resolution | Geometric edges and shimmer, where implemented | Requires a supporting game and supporting hardware | Native-resolution quality where the game and hardware support it |
SSAA: The Thorough Method, at the Highest Cost
Super sampling anti-aliasing, SSAA, is the most straightforward idea in the field, and the most expensive. The entire scene is rendered at a higher resolution than the output, and then shrunk down, averaging several high-resolution pixels into each output pixel. Because every output pixel is an average of several samples, geometric edges, fine textures, and highlights are all smoothed together, and the result is generally the cleanest picture any method in this guide produces. The cost follows the same logic: the whole scene, not just its edges, is computed at the higher resolution, so SSAA sits at the highest end of the relative cost ranking. That cost is stated only as a ranking. No frame rate loss figure, no time figure, and no percentage is given for any method in this guide, because such figures depend on the game, the scene, and the hardware, and a universal number would be fabrication.
SSAA appears rarely in the menus of modern games. That absence is not a fault, and it is not a sign that a machine fails to support it. It reflects the cost: for a modern, demanding scene, rendering everything above output resolution is usually more than the performance budget allows. Where it does appear, it suits older or lighter games, or machines with ample headroom, where the cost can be paid without consequence.
MSAA: Sharp Edges, and Nothing Else
Multi-sample anti-aliasing, MSAA, is the surgical version of the same sampling idea. Instead of rendering the whole scene at higher resolution, it takes multiple coverage samples only where a geometric edge passes through a pixel, and works out what fraction of the pixel the edge covers. The edge is therefore drawn with a smooth gradient of coverage rather than a hard staircase step, and the result is a sharp, clean geometric edge at a cost well below SSAA.
The limitation is structural, and it must be stated without softening: shading is usually computed only once per pixel, even where several coverage samples were taken. MSAA knows where the edge is, but it does not sample the texture or the highlight several times. Fine textures, transparent objects, and specular highlights therefore keep their original shimmer, untouched by the method. A fence or a tree canopy that shimmers before MSAA will shimmer after it, and that is not a settings error. It is the boundary of what the method does.
One further warning belongs here, and it is a hard rule for the rest of the guide. The numbers on an MSAA menu, 2x, 4x, and 8x, are edge sample counts only. They count how many coverage samples are taken on geometric edges. They are not a strength scale that can be compared with the numbers or sliders of other methods. An MSAA 8x setting is not stronger than temporal anti-aliasing in any general sense, because the two numbers measure different things, and they treat different kinds of jaggedness. Never compare anti-aliasing numbers across methods. Compare what each method does.
FXAA: Cheap, Blind, and Often Good Enough
Fast approximate anti-aliasing, FXAA, works after the picture is finished. It scans the completed frame for high-contrast edges, the places where a staircase is likely, and blends those edges to soften the steps. Because it operates on the finished picture, it costs very little relative to the sampling methods, and it can be applied to almost any game. That combination of low cost and universality is its whole appeal.
The price is knowledge. FXAA does not know the geometry behind the picture. It only sees the finished pixels, so it must guess which high-contrast transitions are true edges and which are fine detail. Fine text, delicate texture, and thin lines can be caught in the same blending and softened along with the staircase. The whole picture can therefore look slightly hazy, as though a thin softening filter had been laid over it. For readers who value performance above all, that softness is an acceptable trade. For readers chasing crisp detail, it is the reason FXAA is often the first method to be replaced when a better option exists in the same game.
SMAA: The Same Stage, With Better Manners
Subpixel morphological anti-aliasing, SMAA, works at the same stage as FXAA: after rendering, on the finished picture. Its difference is in how it decides what to blend. Rather than blending every high-contrast transition it finds, it recognises edge shapes, straight lines, corners, and diagonals, and blends selectively according to those patterns, with additional handling at the subpixel level. Its design goal is to reduce the whole-picture softening associated with FXAA while keeping the low cost of a post-processing method.
That design goal must not be hardened into a universal claim. This guide does not state that SMAA is always better than FXAA, because the actual result depends on the implementation in each game. In some games the selective blending preserves noticeably more detail. In others the difference is small. The honest statement is directional: SMAA aims to be a more careful post-processing method than FXAA, and readers should judge it game by game, on the same scene, rather than trusting a ranking.
TAA: Borrowing the Past to Steady the Present
Temporal anti-aliasing, TAA, changes the dimension of the problem. Instead of sampling more within one frame, it borrows information across frames. Each frame, the sample positions are jittered slightly, so successive frames sample the same edge from slightly different positions. Motion vectors are then used to align the previous frames to the current one, and the history is blended with the new frame. In motion, the result is steadier than any single-frame method in this guide, and shimmer is greatly reduced, because detail that flickers in one frame is averaged against its neighbours.
The weakness is also temporal. The blend is only as good as the history alignment. When motion is fast, when the camera cuts, when occlusion changes, or when motion vectors are inaccurate, old colours leak into the new frame in the wrong place. The visible result is ghosting, a faint trail behind a moving object, or an overall softness as history smears detail. That ghosting is a rendering history problem. The two share only the loose word ghosting, and the diagnostic difference is exactly that test: change the scene, and temporal ghosting goes with it.
TAA therefore demands a different kind of judgement from the other methods. A still screenshot of a TAA picture can look excellent, while the same setting in motion shows trailing. Conversely, a method that looks slightly rougher in a screenshot may be perfectly clean in motion. Readers who judge anti-aliasing from screenshots alone will misjudge TAA in both directions, and the diagnostic sequence below insists on separate still and moving tests for that reason.
DLAA: Native Resolution, Model-Based Smoothing
Deep learning anti-aliasing, DLAA, is the DLSS family approach applied at native resolution. The distinction from DLSS must be stated in one clean sentence, because the two names are constantly confused: DLSS renders below output resolution and reconstructs upward, while DLAA anti-aliases directly at native resolution, without lowering the resolution first. The two are not interchangeable, and they must not be compared as though one were simply a stronger version of the other.
DLAA requires a game that implements it and hardware that supports it, and this guide states those two preconditions in that general form only. It names no hardware model list and no driver version, because support changes and any list printed here would age into error. Where both preconditions are met, DLAA offers native-resolution smoothing in the temporal family, with the same broad trade profile as TAA: strong stability in motion, with a dependence on the quality of the implementation in that game.
For completeness, a brief note on a further post-processing relative: some games offer CMAA or similar single-frame post-processing methods. Those belong in the same category as FXAA and SMAA, namely post-processing on the finished picture, and this guide mentions them only to place them. They are not given a separate treatment, because their stage and their limitations are the same ones already described.
Why Modern Menus Often Have No MSAA
Readers who remember MSAA as the standard option often ask why it has vanished from many modern menus. The answer is structural, not a fault and not a cost-cutting trick. Modern engines commonly use deferred rendering, in which geometric edges and shading are processed in separate stages. MSAA, which needs to combine edge coverage with shading in a particular way, fits awkwardly into that separated pipeline. The menu therefore offers temporal or post-processing methods instead. The absence of MSAA is an engine choice, not evidence that a machine or a driver is at fault, and no engine version or interface version list is offered here, because the structural reason is sufficient and version lists would add nothing to the diagnosis.
Resolution Helps, but Does Not Abolish the Staircase
Higher resolution reduces the visible staircase, because the same diagonal edge occupies more pixels, and each step becomes relatively smaller. The direction is reliable, and it is one reason higher-resolution displays look cleaner even before any anti-aliasing is applied. It is not a cure. Fine geometry such as wires and railings, and bright highlight shimmer, can still alias at high resolution, because the detail is still finer than the pixel grid in places. This guide therefore does not state that any particular resolution removes the need for anti-aliasing. The need shrinks, but the method choice remains, especially for the shimmering kind of jaggedness.
Sharpening Is Not Anti-Aliasing
Sharpening deserves a short, firm paragraph, because it is often mistaken for a seventh anti-aliasing method. Sharpening increases contrast at edges, which makes a soft picture look more crisp and energetic. It does not add sampling information, and it does not repair the staircase or the shimmer at their source. Where a temporal method has left the picture soft, adding sharpening is a reasonable trade: the picture looks livelier, while the underlying ghosting, if present, remains untouched. Sharpening treats the symptom of softness. It is not a substitute for anti-aliasing, and it must not be presented as one.
The Diagnostic Sequence, in Order
Run these steps in order. Each step changes exactly one thing, and the pattern of which changes matter is what identifies the cause. Changing several settings at once destroys the evidence.
Step 1: Name which kind of jaggedness this is. A staircase on a still diagonal edge is geometric aliasing. Flickering fine texture or highlights in motion is shading shimmer. A generally soft whole picture is more likely a scaling or resolution problem, covered in article 174, article 235, or article 242, than a pure aliasing problem, and it should not be forced into this guide.
Step 2: Compare on the same scene. Choose one scene and one camera angle, look at it with anti-aliasing off, then try each anti-aliasing option the game offers on that same scene. Comparisons across different scenes are meaningless, because scene content changes the jaggedness more than the setting does.
Step 3: Judge still and moving separately. Examine the scene frozen, then in motion. Temporal methods can look excellent frozen and trail in motion, while sampling methods can look sharp frozen and still shimmer in motion. A screenshot verdict is not a motion verdict, and both are needed.
Step 4: Exclude resolution and scaling interference. If the game uses dynamic resolution scaling, covered in article 235, its real rendering resolution may be below the output resolution, which worsens aliasing. Fix the resolution first, then judge the anti-aliasing. If the picture is softened by display or graphics scaling, article 174, or by operating system scaling, article 242, that softness is not something anti-aliasing treats.
Step 5: Test transparent objects deliberately. Look at wire fences, foliage, and railings under MSAA. If they still alias, that is the method limit, not a settings error. Treating that kind of detail points toward a temporal or upscaling direction, touched on in article 162, rather than toward a higher MSAA number.
Step 6: Choose by trade-off direction, not by number. Ample performance headroom and a love of sharp edges point toward the geometric sampling direction. Shimmer as the main complaint points toward the temporal direction. Performance first, with some softness accepted, points toward the light post-processing direction. No game-specific recommendation list is given, and no performance figure is quoted, because both would age and mislead.
Boundary Cases Worth Knowing
First, the cross-method number ban deserves restating as a formal rule: never compare anti-aliasing numbers across methods. MSAA sample counts, super sampling factors, and temporal strength sliders belong to different measurement systems. A larger number in one system is not better than a smaller number in another.
Second, upscaling and anti-aliasing overlap but are not the same subject. Upscaling reconstruction, covered in article 162, has an anti-aliasing effect of its own, because reconstruction averages and stabilises detail. This guide touches that overlap only at the DLAA boundary and in diagnostic step 5. The full upscaling subject belongs to article 162 and is not repeated here.
Third, frame generation, covered in article 175, is a different layer again. Frame generation invents intermediate frames, while temporal anti-aliasing blends real previous frames. Their ghosting artefacts may look similar at a glance, but their causes differ, and the remedies differ with them.
Fourth, panel-side smearing, covered in article 214 on overdrive and article 220 on flicker, is a display time characteristic, not a rendering aliasing problem. Where the reader cannot tell them apart, the first diagnostic step, still versus moving, combined with changing the game or the picture, separates the layers: rendering aliasing follows the game scene, while panel smearing follows the panel across content.
Eight Myths, Dismantled
Turning anti-aliasing to maximum always gives the best picture. Wrong. Maximum is not the same as most suitable. The most expensive sampling costs the most performance, and the strongest temporal blend can show the most ghosting. Suitability depends on the scene and the performance headroom, not on the slider position.
MSAA 8x must be better than temporal anti-aliasing. Wrong twice over. The 8x figure counts geometric edge samples only, and it treats a different kind of jaggedness from the one temporal methods address. A number from one system cannot outrank a method from another system.
Anti-aliasing fixes all flicker. Wrong. Geometric anti-aliasing does not treat shading and highlight shimmer. That kind of flicker needs a temporal direction, and even then it is improved rather than guaranteed away.
At high resolution, anti-aliasing is unnecessary. Wrong. Higher resolution shrinks the staircase, but fine geometry and highlight shimmer can persist. The need lessens. It does not vanish by rule.
A menu without MSAA means the machine or the driver is faulty. Wrong. Modern deferred rendering pipelines make MSAA awkward to apply, so many engines offer temporal or post-processing methods instead. The missing menu entry is an engine choice.
Temporal ghosting means the monitor is faulty, or has burn-in. Wrong: temporal ghosting vanishes when the scene changes, while panel burn-in stays whatever content is shown. Change the scene, and the layer reveals itself.
DLAA is just another name for DLSS. Wrong. DLSS renders below output resolution and reconstructs upward. DLAA anti-aliases directly at native resolution. The directions are opposite, and the two must not be swapped.
Sharpening is a form of anti-aliasing. Wrong. Sharpening raises edge contrast and makes a soft picture look livelier, but it adds no sampling information and repairs neither the staircase nor the shimmer at their source.
When It Is Worth Having the Machine Checked
Most cases end at step 6 of the diagnostic sequence: a named kind of jaggedness and a trade-off direction that follows from it. A check is worth considering when the tests point in conflicting directions, for example when the picture is abnormal regardless of game and setting, when jaggedness appears together with other display misbehaviour, or when the display behaves oddly under other content. Those patterns point beyond settings to the graphics card, its configuration, or the display, and that is diagnosis work rather than settings work. Published inspection and repair pricing for this kind of work falls within the shop three published tiers, from HK$400 to HK$2,800, depending on what the diagnosis finds; no figure outside that range should be quoted for this category. Our shop is at Room 1209, Dali Plaza, Shek Mun. The telephone number is 2186 8010 and WhatsApp is 6558 6806. There is no Mong Kok shop, and there is no on-site service; diagnosis is done at the shop. Before bringing anything in, write down the results of steps 1, 2, and 3, the kind of jaggedness, the same-scene comparison, and the still-versus-moving judgement, because those three observations narrow the investigation more than any description of the staircase alone.
Frequently Asked Questions
Why does my game menu offer no MSAA option?
Because many modern engines use deferred rendering, in which geometry and shading are processed in separate stages, and MSAA fits that pipeline awkwardly. The menu offers temporal or post-processing methods instead. The missing option is an engine choice, not a fault in the machine or the driver.
I switched anti-aliasing on, so why do trees and fences still flicker?
Because foliage and fences are transparent, fine detail, and their flicker is shading shimmer, not geometric aliasing. Geometric sampling methods do not treat that kind. A temporal direction is the one that can improve it, and the improvement should be judged in motion, not in a screenshot.
Is temporal ghosting a sign that my monitor is faulty?
Usually not. Temporal ghosting comes from history blending in the renderer, and it vanishes when the scene changes. Change the scene: if the trail goes with it, the layer is rendering, not the panel.
At high resolution, do I still need anti-aliasing?
Usually yes, though less urgently. Higher resolution makes each staircase step relatively smaller, but fine geometry and highlight shimmer can still exceed what the pixel grid can represent. Judge on the actual scenes played, rather than assuming resolution alone settles it.
How is DLAA different from DLSS?
DLSS renders below output resolution and reconstructs upward to the output size. DLAA anti-aliases directly at native resolution, without lowering the resolution first. They belong to the same technology family but run in opposite directions, and they must not be used interchangeably.
Can sharpening replace anti-aliasing?
No. Sharpening raises edge contrast, which makes a soft picture look crisper, but it adds no sampling information. It cannot repair the staircase or the shimmer at their source, though it can be a reasonable companion to a temporal method that has left the picture soft.