NVIDIA Reflex vs AMD Anti-Lag: Why Low Latency Sometimes Helps and Sometimes Does Nothing — Render Queue, CPU/GPU Sync & On/Off Comparison Guide

｜ HK GPU Fix

重點摘要

Low-latency modes shrink the render queue so the frame your graphics card draws reflects a more recent input, but they cannot fix network lag, a CPU bottleneck, or a monitor's own limits. This guide explains how NVIDIA Reflex and AMD Anti-Lag work, when each one helps, and how to run a fair on/off comparison before you switch it on.

What Low Latency Actually Means

When players talk about a game feeling responsive, they are usually talking about system latency: the time between an input, such as moving a mouse or pressing a key, and the moment the result of that input appears on screen as changed pixels. That interval is not a single number produced by one component. It is a chain, and every link in the chain adds time.

The chain runs in roughly this order. An input device reports a movement or a key press. The game logic and the processor prepare the next frame: where objects are, what the camera sees, and what needs to be drawn. The graphics card renders that frame. The finished frame is then shown by the display. If any link is slow, the whole chain feels slow, and fixing a different link will not help.

This distinction matters because low-latency modes only act on one part of the chain. They change how the processor and the graphics card coordinate the handover of frames. They do not speed up a network connection, they do not make a processor faster, and they do not change a display's own response behaviour. A player who turns on a low-latency mode and feels nothing has not necessarily done anything wrong. The delay they feel may simply live somewhere the mode cannot reach.

It also helps to separate latency from frame rate. Frame rate counts how many frames appear per second. Latency measures how old the information in a given frame is. A game can run at a high frame rate and still feel sluggish if each frame reflects an input made a long time earlier. Conversely, a game at a modest frame rate can feel direct if each frame is fresh. Low-latency modes target freshness, not frame count. Turning one on should not be expected to raise the frame rate, and a frame-rate counter that stays flat after switching the mode on is not evidence that the mode failed.

The Render Queue: The Heart of the Matter

To understand Reflex and Anti-Lag, start with the render queue. In a traditional rendering pipeline, the processor is allowed to run ahead of the graphics card. While the graphics card is busy drawing one frame, the processor prepares the next frame, and sometimes the frame after that, placing finished work into a queue for the graphics card to pick up.

Queuing has a genuine benefit: it keeps the graphics card fed. A graphics card that finishes a frame and finds no new work waiting must sit idle until the processor catches up, and idle time wastes performance. A queue smooths out small timing differences between the two processors and keeps utilisation high.

The cost of that queue is staleness. A frame sitting in the queue was built from input sampled when the processor prepared it. By the time the graphics card gets around to drawing it, the player may have moved the mouse further, and the game world may have moved on. The deeper the queue, the older the input embodied in the frame being drawn. In a fast-moving competitive game, that age is exactly what the player feels as a vague sense of the crosshair lagging behind the hand.

The queue tends to grow deepest when the graphics card is the slower partner, a situation called GPU-bound. The processor finishes frames faster than the graphics card can draw them, so finished frames pile up. When the processor is the slower partner, a situation called CPU-bound, the graphics card is often waiting rather than queuing, and the queue is not the main source of delay. That is why the same low-latency switch can feel dramatic on one system and do nothing on another: it only helps when a queue is actually part of the problem.

There is a second, subtler point. A queue also hides small hitches. If the processor occasionally takes longer than usual to prepare a frame, queued work lets the graphics card keep drawing in the meantime. Shallowing the queue removes some of that cushion. In most games this trade is worth it, because the latency gain is real and the lost cushion is small, but it explains why a low-latency mode is a trade-off rather than a free improvement, and why testing matters more than assuming.

NVIDIA Reflex: The Engine-Integrated Route

NVIDIA Reflex takes the engine-integrated route. For Reflex to work, the game itself must include Reflex support. The driver and the game engine cooperate: instead of the processor running freely ahead, the processor times its work so that it begins the next frame just as the graphics card is nearly finished with the previous one. The queue stays shallow, and input can be sampled at the latest possible moment before the frame is built.

Because the coordination happens inside the game, Reflex can be precise about timing in a way a driver alone cannot. The game knows when a frame's input sampling matters, when rendering will begin, and how long each stage should take. That inside knowledge is the core advantage of the engine-integrated approach, and it is also its core limitation: a game that has not integrated Reflex simply has no Reflex option, and no driver toggle can conjure one.

Most games that support Reflex offer three settings: Off, On, and On plus Boost. Off leaves the traditional queue behaviour in place. On enables the shallow-queue coordination described above. On plus Boost adds one further behaviour: in situations where the processor is the limiting factor and the graphics card would otherwise drop to a lower clock speed while waiting, Boost keeps the graphics card clock higher so that it is ready the instant work arrives. The cost is somewhat higher power consumption, since the graphics card is held ready rather than allowed to idle down. Boost does not make the processor faster; it only removes the extra waiting that a sleeping graphics card would add on top of a processor-limited scene.

A common misunderstanding is that Reflex is a frame-rate feature. It is not. Reflex does not render more frames, and it does not make the graphics card draw faster. If a game runs at the same frame rate with Reflex on and off, that is the expected outcome, not a fault. What should change, when the queue was contributing delay, is how fresh each frame feels: the crosshair tracks the hand more tightly, and the delay between a click and a visible shot shortens.

Reflex also includes measurement support in some games and hardware, where a compatible display and mouse can time the interval from click to pixel change. That kind of end-to-end measurement is the honest way to put a number on system latency, but it needs specific compatible equipment and a game that supports the measurement flash indicator. Most players do not have that setup at home, and this guide does not assume they do. A careful subjective comparison, done properly, is a reasonable substitute for everyday purposes, as described later.

AMD Anti-Lag: Three Generations That Must Not Be Confused

AMD's low-latency story is harder to follow because the name Anti-Lag has covered three quite different implementations across the years. Treating them as one feature leads to most of the confusion players report, so they are worth separating carefully.

The original Anti-Lag, introduced in 2019, works at the driver level. The driver throttles the processor so that it does not run too far ahead of the graphics card, keeping the queue shallow without any cooperation from the game. Because no game integration is needed, it can be switched on for practically any game. The trade-off is precision: the driver cannot see inside the game engine, so the effect varies with the game and the graphics API. In particular, its effect in DirectX 12 titles is notably limited, because that API gives the game itself more control over submission timing and leaves less room for the driver to intervene.

Anti-Lag+, introduced in 2023, took the opposite approach. Instead of throttling from the driver, it aligned frames from inside the game process, which produced a stronger effect. The problem was how that alignment was achieved: it worked by injecting code into the game, and anti-cheat systems in competitive titles could not distinguish that injection from tampering. After reports of players being flagged, AMD withdrew Anti-Lag+ from its drivers entirely. It is no longer a setting players should expect to find, and any guide still telling readers to enable it is out of date.

Anti-Lag 2, introduced from 2024 onward, returns to the engine-integrated route, now in the same family as Reflex. The game developer integrates an Anti-Lag 2 software kit, and the game and driver coordinate frame timing from inside the engine. There is no injection, so the anti-cheat problem of the previous generation does not apply. The limitation is coverage: whether a given game offers Anti-Lag 2 depends on whether its developer integrated the kit. Switching something on in the driver does not make Anti-Lag 2 active in a game that has not integrated it, and players should check for the option inside the game itself rather than assuming driver-level enablement is enough.

The practical consequence is simple to state and easy to get wrong: the words "Anti-Lag" on a driver panel and the words "Anti-Lag 2" inside a game refer to different mechanisms, with different integration levels and different game coverage. Comparing a driver-level Anti-Lag result with an engine-integrated Reflex result is not a fair comparison of two equivalent features, and any conclusion drawn from such a comparison will mislead.

Side-by-Side Comparison

The table below summarises the three AMD generations alongside NVIDIA Reflex, using only the distinctions established above. It is a map of mechanisms, not a ranking: which technology is available to a given player is decided by their graphics card and by what their games support, not by a universal winner.

A second table helps with the most common decision a player actually faces: whether a low-latency mode is even the right tool for the delay they feel. Each row names a source of delay, whether a low-latency mode addresses it, and what does address it instead.

How Low-Latency Modes Interact With Other Settings

Low-latency modes do not exist in isolation. Several neighbouring settings change the same timing chain, and combining them without understanding the interaction is how players end up with worse results than they started with.

Vertical sync is the first interaction. VSync makes the graphics card wait for the display's refresh moment before showing a finished frame, which removes tearing but adds a layer of waiting. A low-latency mode does not cancel that waiting: it shallows the queue before rendering, while VSync adds waiting after rendering. The two effects stack rather than cancel. A player who enables VSync and a low-latency mode together should expect the result to feel better than VSync alone, but not as direct as either mode with VSync off. Readers who want the full tearing and stutter story should read article #155; this article does not repeat that discussion.

Variable refresh rate, in the form of G-Sync or FreeSync, is a different story. A variable refresh display adjusts its refresh moment to match frame delivery, removing tearing without the fixed waiting of VSync. Low-latency modes generally coexist well with variable refresh, because they act on different parts of the chain: one on queue depth, the other on display timing. They are not the same thing, however, and enabling one is not a substitute for the other. A player who wants both low delay and tear-free presentation will typically want both, and should test the combination rather than assume either alone covers both goals.

Frame-rate caps are the third interaction, and the most commonly mishandled. A cap set very close to the display's maximum refresh rate, or above what the graphics card can sustain, changes queue and waiting behaviour in ways that can mask or exaggerate a low-latency mode's effect. During any comparison, the cap should be fixed and identical for both runs. A cap that is itself the thing being tested should be tested separately, one variable at a time.

Frame generation deserves its own warning. Generated frames insert an additional frame between rendered frames, which makes motion look smoother while adding to the perceived delay, because the displayed sequence now depends on work that takes extra time to produce. A low-latency mode can reduce part of that added delay by keeping the underlying rendered frames fresh, but it cannot erase the generation step itself. Any claim that switching Reflex on fully cancels frame-generation delay overstates what the mode does. Readers who want the full picture on frame generation should read article #175, which covers it in depth; this article does not repeat that discussion.

The general rule across all these interactions is one variable at a time. Change the low-latency setting, and nothing else, then compare. Change the cap, and nothing else, then compare. Stacking three changes and crediting the result to one of them is how myths are born.

Diagnosis: Finding Where Your Delay Actually Lives

Before touching any low-latency setting, it is worth spending a few minutes locating the delay. The steps below are diagnostic, not a prescription, and each step narrows the possibilities.

First, separate network delay from local delay. If the game shows a ping or network latency figure, note it. If the delay appears only in online matches and vanishes in offline practice or single-player modes, the dominant factor is likely network-related, and no local low-latency mode will fix it. If the sluggish feeling is present offline too, the cause is local, and the rest of the diagnosis applies.

Second, determine whether the system is GPU-bound or CPU-bound in the scene where the delay is felt. The practical test is simple: in the same scene, lower the resolution or reduce demanding graphics settings. If the frame rate rises clearly, the graphics card was the limiting factor, and a queue is likely to have been part of the delay. If the frame rate barely moves when the load is reduced, the processor or another bottleneck is limiting, and the queue is unlikely to be the main culprit. This test should be done in a repeatable scene, not in the middle of chaotic gameplay, or the numbers will not be comparable.

Third, check what the game itself offers. Open the game's graphics or gameplay settings and look for a low-latency option by name. If the game offers Reflex or Anti-Lag 2 inside its menus, the engine-integrated route is available. If it offers neither, only the driver-level route, where available, applies, and its effect will be the more modest and variable one. Do not assume a driver toggle has enabled an engine feature; the in-game menu is the authority.

Fourth, note the display's role. A display with a low refresh rate ceiling or high inherent input delay sets a floor under the total latency that no graphics setting can go below. If the display is the floor, the honest answer is that settings tweaks will yield small gains, and expectations should be set accordingly. This is not a fault in the graphics card or the mode; it is the chain's slowest link asserting itself.

Only after these four checks does it make sense to run an on-off comparison. Otherwise the comparison measures everything at once and proves nothing.

The On/Off Comparison: A Fair Test You Can Actually Do

A fair comparison does not require laboratory equipment. It requires discipline: the same game, the same scene, the same settings, and only the low-latency switch changing between runs.

Pick a repeatable scene. A training range, a fixed time-trial route, or a quiet stretch of a map that can be replayed the same way is ideal. Free-form multiplayer matches are poor test scenes because no two matches play identically, and the difference between opponents will swamp the difference between settings.

Fix every setting first. Resolution, graphics quality, VSync state, variable refresh state, and frame-rate cap must be identical in both runs. Write them down if needed. If anything else changes between runs, the comparison is void, and the honest response is to redo it rather than to report a result.

Run each condition more than once. Human perception varies with fatigue, attention, and warm-up. Two or three short runs per condition, alternating the order, gives a more trustworthy impression than one long run each. Note two things per run: the subjective feeling of directness, such as whether the crosshair feels attached to the hand, and whether the frame rate stayed roughly the same. Remember that a flat frame rate is expected; a change in frame rate would suggest something else changed too.

Be honest about a null result. If the two conditions feel indistinguishable after a careful comparison, that is a finding, not a failure. It most likely means the queue was not the dominant delay in that scene, or that the scene was already running with a shallow queue. The correct response is not to stack additional low-latency settings in the hope that more will eventually register. Stacking settings that act on the same chain can interfere with each other, and the comparison that matters is the one already done: this setting, in this game, on this system, made no difference the player could feel.

One further caution applies to measurement software. Some games show an in-software latency figure, and some hardware tools measure the click-to-pixel interval. Those figures are useful as relative indicators within one setup, but figures produced by different tools, different games, or different test conditions should not be compared directly. A number without its method is not evidence, and vendor or third-party figures quoted for a specific game and card under specific test conditions should be read as claims about those conditions, not as promises about any other system.

Boundary Cases: When the Mode Does Nothing, or Should Stay Off

Several situations reliably produce a null result or argue against enabling the mode, and recognising them saves a great deal of wasted fiddling.

If the game does not support the engine-integrated mode, there is no engine option to enable. The driver-level option may still be available, but its effect will be the modest and variable one described earlier, and in some APIs it will be barely noticeable. That is a property of the game's integration, not a defect to be worked around. In particular, no method that modifies game files or injects code to force a low-latency feature into a game should be used, especially in titles with anti-cheat systems. The Anti-Lag+ episode is the cautionary tale: a technique that tampers with the game process can be indistinguishable from cheating to an automated system, and the consequences fall on the player's account.

If the game already runs at a high frame rate with a light load, the queue is already shallow and the remaining delay is small. The improvement available to any queue-shallowing mode is bounded by the delay that actually exists. Feeling nothing in this situation is normal and does not mean the setting is broken.

If the system is severely processor-limited, the Boost variant can keep the graphics card clock ready, but it cannot make the processor complete game logic faster. A scene limited by single-threaded game logic or by a heavily loaded processor will stay limited after the switch. The bottleneck discussion belongs with processor and game performance, not with latency modes, and pretending otherwise only delays the real fix.

Older graphics cards, older drivers, and older games add a final wrinkle: support depends on what the game menu and the driver panel currently show, not on what a remembered article claimed. Support lists change over time as drivers and games update. Players should trust the options visible on their own system today rather than a list copied from elsewhere, and should be suspicious of any guide that publishes a definitive support list without a date.

Common Myths, Corrected

Several myths circulate about low-latency modes, and each one leads to a concrete mistake.

"Switching Reflex or Anti-Lag on raises the frame rate." It does not. The target is delay, not frame count. A player who enables the mode and watches the frame-rate counter for a rise will conclude, wrongly, that the mode is broken, and may then change other settings in search of a rise that was never on offer.

"Switching Anti-Lag on in the driver is the same as Anti-Lag 2." It is not. The driver-level original and the engine-integrated second generation differ in integration, in precision, and in game coverage. Results from one cannot be transferred to the other.

"High delay must mean a graphics card problem." It need not. Network delay, processor limits, display behaviour, and game settings are all common contributors, and the diagnostic steps above exist precisely to avoid blaming the wrong component.

"More low-latency settings switched on must be better." Stacking settings that act on the same chain can interfere, and each additional change makes the comparison harder to interpret. One mode, tested properly, beats three modes enabled on hope.

"Only competitive players need to care about delay." That overstates the case. Single-player and story-driven games tolerate delay more generously, because their demands on reaction speed are lower, but sluggish controls still harm the experience. The difference is one of priority, not of relevance: a competitive player should sort latency out first, while a story-driven player may reasonably prioritise image quality, but neither should be told delay does not matter at all.

When to Leave It On, When to Leave It Off

A reasonable default emerges from everything above. If the game offers an engine-integrated option and the system is graphics-card-limited in that game, switching it on is sensible, and the On plus Boost variant is sensible where processor-limited moments occur and the extra power use is acceptable. If the game offers only the driver-level option, switching it on is harmless to try, but expectations should be modest and the comparison should be done honestly. If a careful comparison shows no difference, leaving the mode off is a perfectly respectable outcome, not a missed optimisation. Settings exist to serve the experience, and a setting that changes nothing the player can feel is not serving anything.

The deeper lesson is about method rather than any single switch. Latency is a chain, and the chain rewards the player who finds the slowest link before reaching for a toggle. A few minutes of diagnosis, a fair on-off comparison, and a willingness to accept a null result will do more for responsiveness than any collection of toggles enabled on faith.

Frequently Asked Questions

Does NVIDIA Reflex increase FPS?

No. Reflex targets the freshness of each frame, not the number of frames per second. The frame rate should stay roughly the same with Reflex on and off. If the game feels more direct while the frame rate is unchanged, that is the expected signature of the mode working.

Is AMD Anti-Lag the same as Anti-Lag 2?

No. The original Anti-Lag works at the driver level and needs no game integration, while Anti-Lag 2 is integrated by the game developer and works from inside the engine. They differ in mechanism, precision, and game coverage, and a driver toggle does not activate Anti-Lag 2 in a game that has not integrated it.

Why does Anti-Lag+ no longer appear in drivers?

Anti-Lag+ worked by injecting code into the game process, and anti-cheat systems could flag that injection. AMD withdrew the feature from its drivers after those conflicts. Players should not seek out old methods to restore it, particularly in games with anti-cheat protection.

Can a low-latency mode fix high ping?

No. Ping is network delay between the player's machine and the game server. Reflex and Anti-Lag act on the local rendering chain only. If delay appears in online play but not offline, the cause is network-related and local graphics settings will not address it.

Why do I feel no difference with Reflex switched on?

The most likely reasons are that the queue was not the dominant delay in that scene, that the system is processor-limited rather than graphics-card-limited, or that the game was already running with a shallow queue at a high frame rate. A careful on-off comparison that finds no difference is a valid finding, not a sign that the setting is broken.

Should I use a low-latency mode together with VSync?

The two can be combined, but VSync adds its own waiting after rendering, which a low-latency mode does not cancel. Expect the combination to feel better than VSync alone but less direct than the low-latency mode with VSync off. Test the combination in the game concerned rather than assuming either setting covers the other's cost.

Does On + Boost make my processor faster?

No. Boost keeps the graphics card clock higher while the processor is the limiting factor, so the graphics card is ready the moment work arrives. It does not accelerate game logic or processor performance. A genuinely processor-limited scene stays processor-limited with Boost on.

A Final Note

Low-latency technology is at its best when it is understood as queue management rather than magic. The render queue exists for a reason, and shallowing it trades a little smoothing for fresher frames. That trade is worth making in the situations where a queue is genuinely adding delay, and it is worth skipping where it is not. The test is simple, the diagnosis takes minutes, and the honest answer, including "no difference here," is more useful than any toggle switched on faith.

For help choosing settings for a particular game or system, or for a check of whether delay is coming from the graphics card, the processor, or the display, readers are welcome to contact the shop at Room 1209, Dali Plaza, Shek Mun, by phone on 2186 8010, or by WhatsApp on 6558 6806.