By Walter Jobs, Technology writer and editor · Published 9 October 2026
Most guides to graphics settings rank them by how much frame rate they cost. That is half an answer, because the setting you want to turn down first is not the most expensive one, it is the one with the worst ratio of cost to what you can actually see. And nearly every guide makes the same structural mistake: it treats all frame rate cost as one thing, when some settings bill the graphics card and others bill the processor.
Get that split wrong and you will spend an evening lowering shadow quality on a machine where shadows were never the problem.
The split that organises everything else
A frame is built in two stages by two different chips. Your processor decides what is in the scene, how many objects there are, where they are and what each one is doing, then hands a list of draw calls to the graphics card. The card does the pixel work: lighting, shadows, reflections, post-processing.
Settings therefore fall into two groups, and the group matters more than the magnitude:
| Bills the graphics card | Bills the processor |
|---|---|
| Ray tracing and path tracing | View distance and object draw distance |
| Shadow resolution and filtering quality | Level of detail transitions and population density |
| Ambient occlusion | Crowd and traffic density |
| Screen-space and ray-traced reflections | Shadow draw distance, as distinct from shadow quality |
| Volumetric lighting, fog and clouds | Physics detail and destruction |
| Resolution and resolution scale | Grass and foliage density, in many engines |
| Anti-aliasing |
Before changing anything, find out which chip you are waiting on. Open your GPU software’s overlay and look at graphics card utilisation while you play. If it sits at 95 to 99 percent, you are graphics-bound and the left column is where your frames are. If it sits below about 80 percent while one processor thread is pegged, you are processor-bound and the left column will do almost nothing. You can turn ray tracing off entirely and watch the frame rate refuse to move. Our bottleneck calculator will tell you which of your two parts is the likely limit before you even start, and the signs of a processor limit covers how to recognise it in play.
Ranked by what you give up, not by what it costs
These are approximate costs on a modern graphics-bound setup at 1440p. Treat them as the order of operations rather than as measurements; the figures vary enormously between engines, which is itself worth knowing.
Turn these down first: large cost, small visual loss
Ray-traced reflections and global illumination. Comfortably the most expensive thing in any modern options menu, frequently 25 to 50 percent of your frame rate on its own, and the setting with the widest gap between what it costs and what you notice in motion. Ray-traced shadows are a smaller hit than reflections. If you want one ray tracing feature, pick global illumination, which changes the look of a scene more than reflections do.
Volumetric lighting and cloud quality. Expensive, typically high single digits to low teens in percentage terms, and a setting whose highest tier is often indistinguishable from the one below in anything other than a screenshot of a god ray.
Screen-space reflections at their maximum. The step from high to ultra on reflections usually buys a more accurate puddle and costs several percent. The step from off to medium is the one that matters visually.
Ambient occlusion above medium. Contact shadowing is worth having. The difference between medium and the most expensive tier is a slightly softer gradient in corners for a few percent.
Leave these alone: small cost, large visual gain
Texture quality. Costs essentially no frame rate as long as you have the video memory for it, and does more for how a game looks than any other single control. This is the setting people lower first and should lower last. The caveat is the video memory one, below.
Anisotropic filtering. Keeps ground textures sharp into the distance, costs about one percent at 16x, and has done for a decade. There is no reason to run it below maximum.
Anti-aliasing, on modern implementations. This one has aged out of the expensive list and most guides have not noticed. MSAA was genuinely costly, which is where the reputation comes from. TAA, FXAA and the anti-aliasing built into upscalers are cheap: measured benchmarks have put FXAA at under two percent. Turning anti-aliasing off to find frames is solving a 2013 problem.
The two dials that are not really settings
Resolution scale is the highest-leverage control in the menu and it is not in the quality list. Rendering at 80 percent of your display resolution and upscaling removes roughly a third of the pixel work. The modern version of this is DLSS, FSR or XeSS, which do the same thing with far better reconstruction: quality mode typically returns 30 to 40 percent more frames at a visual cost most people cannot identify in motion. If you need frames, start here, not with shadows.
Resolution itself. The arithmetic is worth stating because it is routinely got wrong: 1440p is 1.78 times the pixels of 1080p, not 1.6, and 4K is 4 times 1080p, not 8. Going from 1440p to 1080p removes 44 percent of the pixel work, which is more than any individual quality setting will give you.
The settings that do not show up in your average frame rate
Every ranked list of settings, including the one above, is implicitly about average frame rate. Two of the most important settings do not appear in an average at all, which is why people follow a guide, see their average unchanged, and conclude the guide was wrong.
Texture quality, when you are short of video memory. Inside your memory budget, textures are free. Past it, the behaviour changes completely: the game starts moving data across the PCIe bus mid-frame and you get hitches as you turn or as new areas stream in. Your average might still read 70 fps while the experience is unpleasant. Rough thresholds at the time of writing: 8 GB is workable at 1080p and tight at 1440p on high textures, 12 GB is comfortable at 1440p, 16 GB is the figure to want for 4K with ray tracing. If dropping textures one step smooths the game out without changing the average, that was the problem.
Streaming and shader settings. Asset streaming quality and shader cache behaviour show up as stutter on first encounter rather than as lower frames. On a mechanical drive they show up far more.
Both of these are 1% low problems. If you are judging settings by an fps counter you are blind to them; a frame-time graph makes them obvious, and telling stutter apart from a low average is worth doing before you change anything else.
The settings everyone argues about
Motion blur. It costs very little on modern hardware. Per-object blur costs more than camera blur and neither is a meaningful frame rate saving. Turn it off because you dislike it, which many people do, not because you are hunting frames.
Depth of field. Cheap, and a matter of taste.
Film grain, chromatic aberration, vignette. Free, and off is usually the better-looking option anyway.
Shadow quality versus shadow distance. These are frequently one slider and should not be. Quality is a graphics card cost and distance is a processor cost, so on a processor-bound machine lowering the quality half achieves nothing while lowering the distance half can be worth double digits.
Why one list of percentages cannot be right for every game
Every guide to this, including the figures above, prints one set of costs. That is a simplification worth naming, because the same setting behaves very differently depending on how a game is built, and knowing which kind of game you are in tells you where to look first.
| Kind of game | Where the frames actually go | Where to look first |
|---|---|---|
| Open world with dense traffic and crowds | Heavily processor-bound in towns, graphics-bound in open country, so the answer changes as you drive | The right column: population density, draw distance, level of detail |
| Corridor shooter or linear single-player | Almost entirely graphics-bound. The scene is authored and the draw call count is controlled | The left column, and upscaling before anything else |
| Competitive shooter at low settings | Processor-bound by a wide margin, because the card has very little left to do | Nothing in the graphics menu will help. This is a chip and memory question |
| Simulation or strategy with a large world state | Processor-bound and often single-thread-bound, with the graphics card idling | Unit and object counts, simulation detail, view distance |
| Anything with ray tracing enabled | Graphics-bound immediately, whatever else is true of the game | Ray tracing itself, then the reflections tier |
This is the practical reason the generic guides disappoint people. A reader with a competitive shooter running at 110 fps follows a list written for a graphics-bound single-player game, turns off every effect, and gains nothing, because none of those settings were costing them anything in the first place. The organising question is always which chip is waiting, and the answer is not a property of the setting, it is a property of the game you are in.
A method rather than a preset
- Measure which chip you are waiting on. Graphics card at 95 percent and above, work the left column. Below 80 percent with a pegged thread, work the right column.
- Take the free wins first. Upscaling at quality mode, resolution scale, then the three or four effects in the first list above.
- Leave textures and anisotropic filtering at maximum unless the frame-time graph tells you memory is the problem.
- Change one thing at a time and watch frame times, not frame rate. A preset moves twenty settings at once and tells you nothing about which of them mattered on your hardware.
- Stop when it is smooth, not when it is high. A locked 90 is better than an average 110 that swings between 70 and 140.
If after all of that the number will not move, the settings were not the problem. The wider list of things that actually change frame rate covers what sits outside the options menu, and most of it is not what the optimisation guides tell you.