FPS Drops With Game Overlay On and Returns When Off: Discord, Steam and GeForce Overlay Injection, Render Cost and Display Mode Diagnosis Guide

2026-11-19 | GPU Repair | 52 minutes | HK GPU Fix Editorial Team

FPS Drops With Game Overlay On and Returns When Off: Discord, Steam and GeForce Overlay Injection, Render Cost and Display Mode Diagnosis Guide

Key Takeaway

Your frame rate falls when an overlay is on and returns the moment all overlays are off. Run one on/off comparison: play the same sixty seconds of the same game with the overlays on, then again with them all off, and see whether the frame rate returns. That result separates an overlay fault from a driver, settings, or hardware fault.

The Shape of the Problem: Slow Only While an Overlay Is On

This piece covers one narrow pattern and nothing beyond it. Take one machine and one game: while an overlay is running — Discord, Steam, GeForce, AMD, Xbox Game Bar, or several of them together — the frame rate drops clearly, or the frame time goes ragged and play turns jerky. Turn every overlay off and play the same game in the same place again, and the frame rate is back to normal straight away. The desktop still comes up, no blue screen shows, the machine does not keep rebooting, and the game itself keeps running. Nothing has broken outright — only the speed has dropped, and it drops with the overlay switch alone.

Three conditions pin this pattern down, and all three must hold. First, the trigger is the overlay itself: the same game, the same scene, the same options — slower with the overlay running, normal with it switched off. Second, the test stays on one machine, one title, one spot, with the same sixty seconds played in both states. Third, the machine is otherwise steady: it gets to the desktop and stays running, and a slower game with an overlay on is the only new symptom.

That boundary is worth drawing, because the usual trap with this pattern is to stop at "the overlay was on" and call the job done. It is not done. An overlay is several things at once: an injection into the game's rendering, a fresh layer of drawing on every frame, a program in its own right that may draw on the graphics card, a stream of pop-up notifications, and a visitor in the game's display mode. Any one of those parts can slow a game, and each part has to be tested on its own. What needs answering is not whether the overlay caused it in some vague sense. It is which part of the overlay's work is doing the slowing — the injection, the drawing cost, the overlay program's hardware acceleration, the display mode, or several overlays running together. The on/off comparison is the move that splits the first answer off from the rest.

A game that crashes or goes black sits outside this piece altogether. A title that quits on its own, shows a black screen, or refuses to start is a different kind of trouble, with a different set of causes. This guide follows the opposite picture: a game that keeps running, keeps painting its picture, and merely runs slower while an overlay is on, then runs normally the instant the overlay is off. The checks below set two settled states side by side — overlay on against overlay off, one overlay at a time against all of them together, one display mode against another — because a fault tied to one switch is found by moving that one factor alone and watching whether the settled state moves back.

One more boundary belongs at the front, ahead of any check running. Choosing how the game displays in general is not covered here. A single move is taken from that topic — playing the same game once in each mode as a test — and nothing more. Whether an overlay can be called up at all is not covered here either. A single move is taken — noting whether the overlay opens — and nothing more. Background recording and live streaming are not covered here either. A single move is taken — noting whether instant replay or a recording encoder is running while the game slows — and nothing more. The piece keeps to its own pattern by taking a single move from each nearby topic and going no further into those topics.

Why an Overlay Costs Performance at All

An overlay has to put its own picture on top of the game's picture. To do that it injects itself into the game's rendering pipeline: after the game has painted each frame, the overlay paints its own layer on top — who is speaking, a notification, a frame-rate figure, a recording indicator. That injection has a cost of its own. Every frame gains one more drawing pass and one more synchronisation. The overlay is not a still picture sitting on the screen. It draws afresh on every frame: the speaking indicator moves, notifications slide in, the frame-rate figure changes. Those drawings compete with the game for the same graphics resources. When the graphics card still has spare capacity, the overlay's cost is invisible. When the card is already close to full load, one more drawing pass on every frame can make frame times uneven and the frame rate fall. The symptom is therefore not always a lower average frame rate. The average can look acceptable while the game still plays in jerks, because the trouble sits in uneven frame times and in the hitches that arrive when a notification pops up.

Anti-cheat software adds a second mechanism, and it must be kept separate from the drawing cost. Systems such as Easy Anti-Cheat and BattlEye can treat an injection as suspicious, and some of them switch an overlay off by themselves. Injection is not the same thing as wrongdoing: it is the normal way an overlay works, and the overlay is not a hostile program. But the interaction between the injection, the anti-cheat system, the driver, and the game engine can turn a small cost into a large one. Two shapes must never be confused here. An overlay that has been switched off by anti-cheat software is not the same as an overlay that slows a game down. One is the overlay being blocked; the other is the overlay running and costing performance. The checks below tell them apart by watching what happens when each overlay is switched on again, one at a time.

A third mechanism lives inside the overlay program. A program such as Discord carries its own hardware acceleration switch: switched on, the program draws its own window with the graphics card; switched off, it draws with the processor instead. Hardware acceleration on means the overlay program is one more graphics customer competing with the game. Hardware acceleration off means that competition falls away, while the program itself moves onto the processor. Published guides list switching the hardware acceleration off as a step with a large effect, and this guide treats that step as a comparison test, not as a repair. Switching it off and watching the frame rate return proves that the competition layer is involved. It does not prove that the option ought to stay off, and it does not prove that the option was the whole cause, because the injection layer and the notification layer can exist at the same time.

The display mode forms a fourth mechanism. A newer Discord overlay, released from April 2024 onward, works only in windowed mode and borderless full screen, and does not support true exclusive full screen; only the older legacy overlay option supports exclusive full screen. An overlay that wants to show its content on top of a game can push the game out of its exclusive state and onto the composited path, where the desktop window manager adds one more stage and the frame rhythm can worsen. This mechanism comes from the guide on choosing a display mode, and only one move is taken from it: the same game played in exclusive full screen on one run and in a borderless window on the next, watching which of the two runs slows down. The composition model itself is not rewritten here.

One further layer is worth a note, even though it falls outside this piece's topic. A GeForce overlay with Instant Replay switched on keeps encoding the last few minutes of play in the background, using encoder time and disk writes, even when no clip is ever saved. That layer belongs to the guides on background recording and encoder overload, and this piece takes only one note from it: if replay is on, it must be switched off as a separate comparison, because its cost sits on top of the overlay's own cost. A game that slows only while replay is on has a recording cost, not necessarily an overlay cost, and the two must be tested apart.

Five Tests That Narrow the Field Down

Five comparisons do the diagnosing here, and each one has two possible outcomes. They run in order: each outcome cuts down what comes next, and the opening comparison — overlay on against overlay off — counts for more than any other check here. It does not depend on how large any single reading is. It depends only on what shifts when a single variable shifts, in one stretch of play, on one machine.

The opening comparison is the overlay on/off test, and it is the central check of this piece. One game, one scene, one sixty-second stretch of play, run twice: once with the overlays on, once with every overlay switched off — Discord, Steam, GeForce, AMD, and Xbox Game Bar, all of them. Write down the mean frame rate, the frame time, and the 1% low figure for both runs. If the frame rate returns to normal the moment the overlays are off, the fault follows the overlay layer, and the diagnosis stays here. If the game is still slow with every overlay off, the overlay is not the cause, and the next step is the set of guides for sudden stutter, progressive slowing, or low card usage. What makes this test strong is that one thing only changes — whether overlays are on or off — and everything else is left alone. Switching the overlays off in this way is a test, not advice to stop using overlays for good. It proves the trouble follows the overlay. It does not name which overlay, and it does not say the overlay is a bad program.

The second test sets one overlay against the rest. With the overlays all off and the game running normally, each overlay is switched back on, one at a time, and the game is tried after every single one: Discord first, then Steam, then GeForce, then AMD, then Xbox Game Bar. Whichever overlay brings the trouble back when it is switched on again is the one at fault. Switching them all back on in one go gives no answer, because once several change together there is no way to tell which one brought the trouble back. This step is the one readers most often rush, and rushing it is the commonest way to lose the answer the first test just gave. Patience here is not a virtue; it is the test itself.

The third test sets a hidden overlay against a visible one. An overlay program can be running, and its injection still active, even while nothing of it is showing on screen. If the game slows while the overlay shows nothing at all, the injection and hook layer is the suspect: the overlay draws on every frame whether it is visible or not. If the game only hitches when a notification pops up, when someone joins a voice channel, or when someone speaks, the notification rendering layer is the suspect. The two are pulled apart by switching all overlay notifications off and playing again. Smooth with notifications off points at the notification layer. Still slow with notifications off points at the injection layer. The two layers call for different follow-up, and lumping them together under "the overlay slows the game" hides which one it is.

The fourth comparison is the display-mode test. The same game is run in exclusive full screen first, then again in a borderless window, with the overlay on for both runs. Slow in exclusive full screen only, and normal borderless, puts the blame on the display-mode layer: the overlay is pushing the game out of its exclusive state and onto the composited path. Slow in both modes rules the display mode out, and the hunt moves back to the injection, drawing-cost, and hardware-acceleration layers. This test takes the exclusive-against-composited mechanism from the display-mode guide, and leaves the rest of that guide alone.

The fifth comparison sets a light game against a heavy game. A light game shows no change with the overlay on, while a heavy game slows plainly — that points at the resource layer: the card is already close to full load in the heavy game, and the overlay takes the last of the spare capacity. A light game that slows too, even though the card has plenty of spare capacity, points away from resources and towards the injection-conflict or anti-cheat layer: the cost is an interaction, not a lack of capacity. This test picks which road the diagnosis follows from here. It cannot be left out, because each road leads to different checks.

TestWhat is comparedResult pointing at the overlay layerResult pointing elsewhere
Overlay on against overlay offSame game, same scene, same sixty seconds, overlays on, then all offFrame rate returns the moment the overlays are offStill slow with every overlay off; not an overlay fault
One overlay at a timeEach overlay switched back on singly, game tried after eachOne overlay brings the trouble back when switched on againNo single overlay brings it back; several overlays or background programs together
Hidden against visibleOverlay running with nothing showing, against notifications popping upSlow even with nothing showing; injection layerHitch only when a notification pops up; notification layer
Exclusive full screen vs borderless windowSame game, overlay running, played once in each modeSlows in exclusive full screen only; display-mode layerSlows in both modes; display mode is not the layer
Light game against heavy gameSame overlay on, light title, then heavy titleHeavy game only slows; card close to full load, overlay takes the remainderLight game slows too; injection conflict or anti-cheat layer

How Injection, Drawing Cost, and Hardware Acceleration Work

The injection mechanism deserves a closer look, because it is the part readers most often misunderstand. An overlay cannot simply paint over a game the way a window paints over a desktop. The game owns its picture, frame by frame, and the only way for the overlay to appear on top is to step into the game's rendering pipeline and add its own drawing after the game has finished each frame. That step is called injection, or hooking, and it is the normal working method of every overlay. The cost is real even when nothing is visible: the overlay still hooks the pipeline, still synchronises with the frame, and still stands ready to draw. That is why a game can slow down with an overlay that shows nothing at all. The overlay is not idle. It is present in the pipeline on every frame, and its presence has a price.

The drawing cost is the second mechanism, and it is the one that grows with what the overlay shows. A speaking indicator, a notification sliding in, a frame-rate figure updating, a recording marker — each of these is drawn afresh on every frame, and each drawing competes with the game for the same graphics resources. The cost is therefore not fixed. It rises when notifications arrive and falls when the screen is quiet, which is why a game can run smoothly for minutes and then hitch exactly when a message arrives. Published guides describe the symptom in exactly these terms: a fall in frame rate, uneven frame times, and a hitch when a notification pops up. The average frame rate on its own can hide this shape, because the average smooths the hitches away. That is why this guide asks for the 1% low figure and the frame-time graph alongside the average: the average says how fast the game runs on the whole, and the 1% low and the frame time say how unevenly it runs. A game with a good average and a bad 1% low is a game that hitches, and hitching is the shape the overlay layers produce.

The third mechanism is hardware acceleration, and readers mistake it for a repair more often than any other. Discord and programs like it can draw their own windows either with the graphics card or with the processor, and the choice is a setting inside the program. Acceleration on makes the overlay program a second graphics customer beside the game. Acceleration off moves the program onto the processor, and the graphics competition falls away. Published guides describe switching the hardware acceleration off as a step with a large effect; this guide treats it strictly as a comparison — switch it off once, play the same sixty seconds, and note whether the frame rate returns. A return proves the competition layer is involved. It does not prove the option should stay off, and it does not prove the option was the only cause. The injection layer can sit underneath the acceleration layer, and switching the acceleration off can hide the injection cost without removing it. That is why the acceleration test comes after the on/off test and the one-at-a-time test, not ahead of them. If the on/off test already has the frame rate returning with every overlay off, the acceleration layer is only one part of the answer, and changing it adds nothing to the diagnosis until the guilty overlay is named.

One further link is worth attention: how several overlays act on each other. Discord, Steam, GeForce, AMD, and Xbox Game Bar can all be on at once, each injecting into the same pipeline, each drawing on every frame, each with its own notifications. No single one of them may be at fault, and the game may still slow, because the combined cost of several small injections is larger than any one of them alone. That is why the second test brings the overlays back singly: when no single overlay brings the trouble back, that outcome itself names the combined layer, not one guilty program. The next step is not to blame all overlays at once, but to leave switched on only the overlays the owner really plays with, and to run the game again on that smaller set. Recording software and monitoring tools sit in the same combined layer: a frame-rate monitor, a recording encoder, and an overlay can each be small on their own and heavy together, and the test that pulls them apart is the same one-change-at-a-time rule.

The Five Suspicion Layers

Every case of a game slowing with an overlay on belongs under one of the five layers set out in this section. Those layers show up in the order the evidence tends to suggest first, but that order settles nothing on its own: naming a layer takes a test behind it, backing it or clearing it. No source agrees with another about which layer appears most often, so this guide ranks none of them first. Each layer carries its own pointing evidence, and the evidence decides.

LayerWhat it meansEvidence that points at it
First layer: injection-conflict layerOne overlay does not sit well with the game, the anti-cheat system, or the driverSwitching that one overlay back on brings the trouble back; only one game slows, others run normally
Second layer: render-cost layerThe overlay draws on every frame, and notifications pop up on topHitch only when a notification pops up; smooth with notifications off; heavy games slow, light games do not
Third layer: hardware-acceleration layerThe overlay program draws its own window using the graphics cardFrame rate comes back once the program's hardware acceleration is off; still slow with the overlay hidden while the program runs
Fourth layer: display-mode layerThe overlay pushes the game out of exclusive full screen onto the composited pathSlow only in exclusive full screen; normal in borderless windowed mode
Fifth layer: combined overlays and background programs layerSeveral overlays, recording, and monitoring tools togetherSmooth only with every overlay off; no single overlay brings the trouble back on its own; worse while recording

The first layer is the injection-conflict layer, and the one-at-a-time test exists to find it. One overlay does not sit well with one game, with its anti-cheat system, or with the driver, and the game slows while that overlay is on. The evidence is the second test itself: switching that single overlay back on brings the trouble back, and switching it off takes the trouble away. Supporting evidence is the range test: only one game slows, while the others run normally with the same overlay on. The follow-up is to switch that overlay off for that one game only, not for the whole machine. A per-game switch exists in the overlay programs for exactly this reason, and using it keeps the overlay available everywhere else. That is a conflict between one overlay and one game, not a broken overlay. The same overlay can run perfectly well in another game on the same machine.

The second layer is the render-cost layer: the overlay draws on every frame, and its notifications pop up on top of the game. The evidence is narrow and specific: the game hitches only when a notification pops up, when someone joins a voice channel, or when someone speaks; switching notifications off makes the game smooth again; heavy games slow while light games do not. This layer sits beside the guide on uneven frame times, which treats frame-time diagnosis in full; this piece uses only the frame-time comparison, and does not redo that guide. The follow-up is to switch the notifications off as a test and note the result. A return to smooth play proves the notification layer is at work. It does not prove the overlay should be used without notifications for good, and it does not prove the injection layer is innocent, because the two layers can exist at the same time.

The third layer is the hardware-acceleration layer: the overlay program draws its own window with the graphics card, and that drawing competes with the game. The evidence is that the frame rate returns once the program's hardware acceleration is off, and that the game is still slow with the overlay hidden but the program running. Nothing changed in this layer is a fix — each change is a trial: switch the option off once, write the outcome down, and put the option back. A change that brings the frame rate back shows the trouble sits in the acceleration layer. It does not mean the option should stay changed, and it does not show the acceleration was the whole cause.

The fourth layer is the display-mode layer. The overlay pushes the game out of its exclusive state and onto the composited path, where the desktop window manager adds one more stage and the frame rhythm can worsen. The evidence is that the game slows only in exclusive full screen and runs normally in a borderless window. The display-mode guide lends this layer its exclusive-against-composited mechanism; nothing else is taken from that guide. The next step is to write down which mode ran normally and which one slowed: that note names the layer, and an inspection starts from notes of that kind.

The fifth layer is the combined overlays and background programs layer, and it is an exclusion layer. Several overlays can be on at once, together with recording software, a frame-rate monitor, and other background programs, and the combined cost can slow a game even when no single program is at fault. The evidence is that the game is smooth only with every overlay off, and that no single overlay brings the trouble back on its own, or that the game is plainly worse while recording is running. That is not proof that background programs are the culprit. It only shows that too many things are on at once, and they must be separated one at a time. The follow-up is to leave on only the overlays the owner really plays with, switch the rest off, and run the game again on the smaller set.

The Seven Diagnostic Steps

Work through the seven checks one by one, in the order printed. Each check returns a result that decides which check follows. Jumping straight to a driver reinstall, or to swapping hardware, ahead of the opening three checks usually wastes the whole attempt: no baseline, no on/off comparison, and no one-at-a-time test in hand means a later change proves nothing at all, even if the frame-rate figure moves.

CheckWhat to doHow to read the result
Check 1Fix the scene and write down the baseline: one game, one scene, one sixty-second run, overlays on; note mean frame rate, frame time, 1% low, card usage, and clockWithout a fixed scene, no later test means anything; different scenes load the machine differently
Check 2Run the overlay on/off test, the deciding one: switch every overlay off — Discord, Steam, GeForce, AMD, Xbox Game Bar — and play the same sixty seconds againFrame rate back to normal means the overlay layer — stop at this check; still slow means move on to other guides
Check 3Bring the overlays back one by one: switch each overlay on alone, then try the game after each oneThe overlay that brings the trouble back names the layer; none of them alone means the combined layer
Check 4Run the notification comparison: switch all overlay notifications off and play againSmooth with notifications off means the notification layer; still slow means the injection layer
Check 5Run the hardware-acceleration test: switch the overlay program's hardware acceleration off and play againFrame rate returns means the acceleration layer; no change means it is not that layer
Check 6Run the display-mode test: the same game with the overlay on, first in exclusive full screen, then in a borderless windowSlow in exclusive full screen only means the display-mode layer; slow in both means it is not the mode
Check 7Stop line: every overlay off and the game still slow, with corruption in the picture, a black screen, or a crash alongside the slownessCarry the records — the on/off frame-rate figures, which overlay brought the trouble back, the display mode, and the checks run so far — to bring along to the inspection; do not install old overlay programs found online one by one

The first check prepares the ground for every check that follows, and it runs ahead of any test. Fix the scene: one game, one saved spot, one view, one sixty-second stretch of play. Put the baseline down in full: the mean frame rate, the clock, the card usage, the 1% low figure, and the frame-time graph, all with the overlays on. Do not compare a quiet indoor scene with a busy outdoor scene: the loads differ, so any test between them proves nothing. If the earlier frame rate was never written down anywhere, say so plainly: nobody remembers a number exactly, and a remembered figure is not a baseline. Every check after this one is read against the baseline. Without it, a later figure cannot be read as a shift or as no shift — there is nothing to set it beside.

The second check is the overlay on/off comparison — the one this guide turns on. Every overlay is switched off — Discord, Steam, GeForce, AMD, and Xbox Game Bar — and the same scene is played a second time, for another sixty seconds. If the frame rate is back to normal, the overlay layer is confirmed, and the work stops at this check apart from the one-at-a-time test that names the guilty overlay. If the game is still slow, the overlay is not the cause, and the work moves to the guides for sudden stutter, progressive slowing, or low card usage. Switching the overlays off in this way is a test. It is not advice to stop using overlays for good. Overlays are not always harmful: the overlay programs exist because players use them, and a test that switches them off proves only that the trouble follows them.

The third check is the one-at-a-time test, and it comes straight after the on/off comparison for a reason: the on/off test proves the trouble follows the overlays as a group, and only the one-at-a-time test can name which one. Each overlay is switched back on alone, and the game is tried after every single one. Whichever overlay brings the trouble back when it is switched on again is the one at fault. The follow-up then is to switch that overlay off for that one game only. If no single overlay brings the trouble back, the combined layer is the answer. The follow-up then is to keep only the overlays the owner actually uses. Hold to changing one variable only: switching several overlays back on at once and then seeing the trouble return gives no usable answer, because several changes at once leave no way to tell which overlay mattered.

The fourth check is the notification test. Every overlay notification is switched off — Discord notifications, Steam notifications, and any others — and the game is played again. Smooth with notifications off means the notification rendering layer: the game hitches only when something pops up on top of it. Still slow with notifications off means the injection layer: the overlay costs performance even while it shows nothing. Each outcome points a different way, and no other test separates them. Notifications switched off for the test go back on afterwards when the test clears them; note down what changed either way.

The fifth check is the hardware-acceleration test. The overlay program's hardware acceleration is turned off — in Discord's settings, or in the matching setting of whichever program is under test — and the game is played again. If the frame rate returns, the acceleration layer is confirmed, and the result is written down. If nothing changes, the acceleration layer is ruled out, and the work moves on. The option is put back after the test where it was not the cause, leaving the machine as it was. One caution belongs here: a change made while the game is still running may not take effect until the game is restarted, because the overlay may not unload until the game closes. Published guides say so outright, and this guide follows it: restart the game after changing an overlay setting, or the test may measure the old state and mislead the reader.

The sixth check is the display-mode test. With the overlay on, the same game is played first in exclusive full screen, then in a borderless window. Slow in exclusive full screen only means the display-mode layer: the overlay is pushing the game out of its exclusive state and onto the composited path. Slow in both modes rules the display mode out, and the hunt moves back to the injection, drawing-cost, and acceleration layers. This check sits deliberately after the one-at-a-time and notification checks, because a mode result by itself cannot name the guilty overlay — it can only show where to look next once that overlay is named.

The seventh check is the stop line. Every overlay is off and the game is still slow, and on top of that the picture is corrupt, the screen is black, or the game has crashed. At that point, stop testing, and bring the records in for inspection: the frame-rate figures with the overlays on and off, which overlay if any brought the trouble back, the display mode in use, and the checks already completed. Do not keep installing and removing overlay programs again and again, and do not install old overlay programs found online one by one. Trying overlay programs one after another at random piles up changes and no insight, and the cause stays unidentified.

Six Misleading Ideas to Set Aside

Misleading idea one: an overlay always slows a game down. This is wrong. With a light game, or with a card that still has spare capacity, the overlay's cost stays invisible. Only when the card is already close to full load does the cost show plainly. An overlay has a cost, and that cost is not a slowdown in every case. The evidence can only come from an on/off comparison on one machine, in one game, in one scene. "Overlays always slow games" and "overlays never slow games" are both wrong, and the on/off test replaces both.

Misleading idea two: a small dip in the frame rate proves the overlay is at fault. That is not right. One frame-rate figure, read one time, proves nothing on its own. The call must rest on whether the same scene, tested twice — overlays on, then overlays off — steadily reproduces the difference, and whether switching one overlay back on brings the difference back. One number, read once, cannot separate an overlay fault from ordinary variation between runs.

Misleading idea three: turning the overlay off has fixed it. This is wrong. Switching an overlay off is a test of comparison. It proves the trouble follows the overlay layer, and it keeps the trouble away while the overlay stays off. It does not name the root cause, and it does not say the overlay should never be used again. To keep using the overlay, the guilty one must be named by the one-at-a-time test, and switched off for that one game only, not for the whole machine. Calling the test a repair ends the work before the layer has been properly found.

Misleading idea four: switching all the overlays off at once is enough. This is wrong. Switching them all off at once proves the trouble follows the overlays as a group, and it finds nothing about which one is guilty. Only switching them back on one at a time, with a game test after each one, can name the guilty overlay. A reader who switches everything off and stops there has an answer about the group and no answer about the member, and the next game that slows will need the whole test run again.

Misleading idea five: an overlay that will not open is the same fault as an overlay that slows a game. This is wrong. An overlay that cannot be summoned is a different shape: its switch, its shortcut, or its permissions are at fault, and that shape belongs to its own guide. A game slowing while an overlay is on is a performance shape, and the two shapes have different checks. Lumping them together pushes the reader towards the wrong set of checks, because the test for a missing overlay is not the test for a slow game.

Misleading idea six: updating to the newest driver will always fix an overlay problem. This is wrong. An overlay problem is not a driver-version problem. The driver, the overlay, the anti-cheat system, and the game engine interact, and swapping the driver before the on/off comparison has run throws away the material the test depends on — the overlay state, the baseline figures, and the scene exactly as it was. The on/off comparison comes first. A driver swap before the diagnosis is done ruins the test; a driver swap after the diagnosis is a different decision, judged on its own checks.

Results Table: What Each Outcome Means

The table below pulls the checks together in one place. Each row opens with something visible — a frame-rate figure, an overlay switched on, a mode changed — and closes by naming the layer that outcome points to. It follows the seven checks, so the reader can cross-check the diagnosis without it standing in for the checks themselves.

Outcome observedLayer indicatedNext move
Every overlay off brings the frame rate back at onceOverlay layer in generalBring the overlays back one at a time to name the guilty one
One overlay switched back on brings the trouble backFirst layer: injection-conflict layerSwitch that overlay off for that one game only
No single overlay brings the trouble backFifth layer: combined overlays and background programs layerKeep only the overlays actually used; test the combination that remains
Smooth with notifications off, hitching with them onSecond layer: render-cost layer, notification partNote the notification result; the injection layer still needs its own test
Still slow with notifications offSecond layer: render-cost layer, injection part, or first layerMove to the hardware-acceleration and display-mode comparisons
Frame rate comes back once hardware acceleration is offThird layer: hardware-acceleration layerNote the result; put the option back where it was not the cause
Slows in exclusive full screen only, runs normally borderlessFourth layer: display-mode layerNote which mode runs normally and which one slows
Slow in both display modesDisplay mode ruled outReturn to the injection, drawing-cost, and acceleration layers
Heavy game only slows, light game does notRender-cost layer; card close to full loadThe overlay takes the last of the spare capacity
Light game slows tooFirst layer: injection conflict or anti-cheat layerThe cost is an interaction, not a lack of capacity
Still slow with every overlay off, and the picture has gone corrupt, the screen has gone black, or the game has crashedOutside this guideStop testing; bring the records in for inspection

Those last two rows deserve a careful read, because they lay out what each test can and cannot prove. An on/off comparison that brings the frame rate back shows the trouble follows the overlays as a group — but it does not show that one overlay is the only cause, because the notification layer, the acceleration layer, and the combined layer can exist at the same time as the injection layer. The one-at-a-time test, the notification test, and the acceleration test still need to run before the layers can be pulled apart, and this guide says that limit outright. A result of "no single overlay brings it back" proves nothing about any one overlay: it proves that the combined cost is the shape, and the follow-up is to cut the combination down, not to blame one program.

Where the Sources Part Ways

The sources behind this guide split apart, and the places where they part need to be set out plainly rather than smoothed away. The first split is over how large an overlay's cost is. Published guides put unlike figures on that cost, each measured on different hardware, in a different game, by a different method, and no figure can be written down here as a general truth. This piece adopts none of them. It asks the reader to judge by the on/off comparison on one machine, in one game, in one scene, instead of by a figure from another machine that the sources do not agree on. The disagreement itself is the finding: the cost varies with the game, the graphics API, the card, and the combination of overlays, and only a test on the machine in question can say what it is there.

The second split is over which cause comes up most often. One published guide puts hardware acceleration first. Another puts the need to separate the overlay itself first. The two are slicing through the same trouble in unlike ways, not rival orderings of a single list, and the sources never settle on one view. This guide ranks none of them first. Five layers follow, each with its own backing evidence, and none of them is ever picked out as the most frequent cause.

The third split is over whether switching an overlay off counts as a repair or a test. Every published guide consulted here presents it as a repair step. The view taken here is different: switching an overlay off is a test. It shows the trouble stays with the overlay layer, and it does not advise never using overlays again. The split counts, because a reader who mistakes the test for a repair ends the work before learning which overlay is guilty, or whether the trouble sits in the notifications, the acceleration, or the display mode.

The fourth split concerns small frame-rate changes. A threshold for run-to-run variation is put forward by one published guide alone: a change under that threshold, the guide says, may be nothing more than ordinary variation between runs. No other source consulted here backs that threshold as a general standard, so this piece does not treat it as a rule. The judgement here rests instead on repeatability — whether the same scene, tested twice with the overlays on and then off, shows the same difference again — rather than on measuring one number against a threshold the sources do not agree on.

The fifth parting is over the new Discord overlay and exclusive full screen. One published guide states that the new overlay works only in windowed and borderless modes, and does not support true exclusive full screen, with the legacy option as the one that supports it, and that the legacy option may be removed in future. That is that guide's statement about the version of its day. This piece states it as the position at the time of writing, and does not write it as a promise about future versions, because the overlay programs change their behaviour between releases.

Outside This Guide: Patterns Other Guides Cover

The table below collects lookalike shapes that sit with other guides. Each row gives the shape and its guide, and why that guide is the right home for it. The table belongs with the diagnosis: if what the reader sees lines up with a row here better than with the shape in the opening section, another guide is the right one to move to.

PatternThe guide that covers itWhy it sits outside this piece
One game crashes or goes black, overlay one step among several#290, crash and black screenThe premise here is a game that keeps running and only slows, not a crash
Overlay cannot be summoned at all#460, overlay switch and shortcutThe premise here is an overlay that is on and slowing the game, not one that will not open
Interface flickers inside the game while the 3D picture is normal#473, interface flickerThe premise here is a frame-rate fall, not a flickering interface
Picking a display mode in general — exclusive full screen or a borderless window#209, display-mode choiceThis piece takes only the one mode comparison, and does not teach display-mode choice
Windows Game Mode on makes games slower#400, Game ModeThe trigger here is an overlay, not Game Mode
Instant Replay background buffering slows the game#270, background recordingThis piece notes replay only as a separate comparison, and does not teach recording diagnosis
Streaming or recording encoder under load#171, encoder overloadHere the trigger is an overlay being on, not an encoder under load
Frame rate looks good but play feels jerky#363, frame timeThis piece uses only the frame-time comparison, and does not redo frame-time diagnosis
Sudden stutter and a large frame-rate fall#136, sudden stutterHere the fault follows the overlay switch, not a sudden onset
Game grows slower the longer it runs#356, progressive slowingThe fault here is a step tied to the overlay switch, not a slow shift over time
Card usage low and frame rate low all along#509, low card usageThe premise here is a game that runs normally with overlays off
Card usage reads as no usage at all while gaming#408, usage readingThe premise here is a usage figure that is present and normal with overlays off
Game closes back to the desktop mid-play#157, closing to desktopThe premise here is a game that stays running and only slows
Full-screen game jumps back to the desktop mid-play#265, focus lossThe premise here is a frame-rate fall, not a loss of focus
Anti-cheat system blocks the overlay outrightAnti-cheat blockingThe overlay itself is switched off, not slowing the frame rate

Two rows in that table repay a closer look, since readers confuse them with this shape more than any others. The first is the crash and black-screen guide. A game that closes itself or goes black has a crash shape, and the overlay is only one step among several in that guide. The shape here is a game that keeps running and only slows, and the overlay is the whole subject. The two shapes share an overlay and little else, and their checks go in different directions. After that comes the overlay that will not open. An overlay that cannot be summoned has a switch, shortcut, or permissions shape, and its checks are about the switch and the shortcut. The shape here is an overlay that is on, working, and slowing the game. Mixing the two up sends the reader off to the wrong checks, because the test for a missing overlay cannot say anything about a slow game.

When to Stop Testing

Stop when all seven checks are done and no layer is named. A game that has been fixed to one scene, weighed with the overlays on and off, run through a one-at-a-time test, a notification test, a hardware-acceleration test, and a display-mode test — and still leaves no clear answer — has used up what home testing can offer. Going further from there is guesswork, and blaming programs that were never at fault muddies the picture further. Put the changed options back before stopping, so the machine ends up as it was.

Stop when every overlay is off and the game is still slow, with the temperature or fan figures off, or the picture has gone corrupt, the screen has gone black, or the game has crashed. Those signs belong to another family of guides, and more gaming tests from here risk turning the question into a hardware fault. Note the new sign, when it showed up, and which test was running then, and keep that note with the records. That note counts as evidence, and it says more than a list of changed options ever could.

Stop too when one overlay has been switched off and nothing moved, and then a second option is changed and nothing moves either. Two factors tried and an outcome that has not shifted already answers the question: neither factor governs the layer. A third, fourth, and fifth overlay program tried one after another gains nothing. A pile of changes is left behind to undo before an inspection can even start.

Hold on to the baseline notes once testing ends. The frame-rate figures with the overlays on and off, the 1% low figures, the frame-time graph, the card usage, and the clock, written out in the same scene on both sides of each test, count for more together than any one of them does alone. An inspection starts with those notes — that is why the baseline comes first, written down before any test runs.

What to Bring for Inspection

Bring the notes from the seven checks: the fixed scene and the baseline figures; the frame-rate figures with the overlays on and with them all off; which overlay, if any, brought the trouble back when it was switched on again; the notification test result; the hardware-acceleration test result; the display-mode test result; and whether one game slows or every game does.

Bring the game name and the scene that was fixed for the tests, and the options in use at the time, because the overlay programs in use and the display mode are evidence too, and so is the outcome of switching them. Bring the card model, driver release, and processor model, since that set of three belongs in the evidence. Keep the options as they stand now, with every test option put back, and do not bring a list of the options changed during testing. What counts as evidence is the options as they stand plus the seven test outcomes. A past change counts only where it moved the frame rate, since that change points at the layer.

For inspection, HK GPU Fix is at Dali Plaza 1209, Shek Mun. The phone number is 2186 8010 and WhatsApp is 6558 6806. The published inspection fee range is HK$400 to HK$2,800, and no price is estimated before inspection. There is no Mong Kok shop and no on-site service.

Frequently Asked Questions

My frame rate falls when an overlay is on and returns when I switch it off. Is the overlay definitely at fault?

The on/off comparison proves the trouble follows the overlay layer, and that is a strong result — but it does not yet name which overlay, or which part of its work. The overlays as a group are implicated, and the next step is the one-at-a-time test: switch each overlay back on singly, with a game test after each one, until the one that brings the trouble back is found. It also does not prove the overlay is a bad program. An overlay has a real cost — injection, drawing on every frame, hardware acceleration, display-mode interaction — and the cost only shows when the card is close to full load, or when one overlay conflicts with one game. Name the guilty overlay first, then decide what to change.

The game only hitches when a notification pops up. Is that the injection layer or the notification layer?

It is the notification layer, and the two must not be lumped together. The injection layer costs performance even while the overlay shows nothing, because the overlay still hooks the pipeline on every frame. The notification layer costs performance only when something pops up — a message arriving, someone joining a voice channel, someone speaking — because the overlay must draw that new content on top of the game. The test is to switch all overlay notifications off and play again: smooth with notifications off points at the notification layer; still slow with notifications off points at the injection layer. The two layers call for different follow-up, so the test matters.

How do I find out which overlay is the culprit?

Switch all the overlays off until the game runs normally again, then switch them back on singly, with a game test after each one. Whichever overlay brings the trouble back when it is switched on again is the culprit. Switching them all back on in one go gives no answer, because once several change together there is no way to tell which one brought the trouble back. This step is the one readers most often rush — and rushing it loses the answer the on/off test just gave. A change made while the game is still running may not take effect until the game is restarted, so restart the game after each change, or the test may measure the old state.

Does switching the overlay off fix the problem?

No. Switching it off is a test of comparison, not a repair. It shows the trouble follows the overlay layer, and it keeps the trouble away while the overlay stays off. It does not name the root cause, nor does it mean the overlay must never be used again. To keep using overlays, the guilty one must be found by the one-at-a-time test and switched off for that one game only, not for the whole machine. Treating the test as a repair ends the work before the layer has been properly found, and the reader never learns whether the trouble sat in the injection, the notifications, the acceleration, or the display mode.

A light game runs fine with the overlay on, but a heavy game slows down. What does that tell me?

A light game showing no change with the overlay on while a heavy game slows points at the resource layer: the card is already close to full load in the heavy game, and the overlay takes the last of the spare capacity. A light game that slows too points away from resources and towards the injection-conflict or anti-cheat layer: the cost is an interaction, not a lack of capacity. That split picks the road the diagnosis follows from here: the resource layer calls for the usage and frame-time readings, and the conflict layer calls for switching the overlays off one game at a time.

Does switching off hardware acceleration in Discord fix it?

It can prove the acceleration layer is involved, and that is not the same as fixing the problem. With hardware acceleration on, the overlay program uses the graphics card to draw its own window, and that use competes with the game. Switch it off once, play the same sixty seconds, and note whether the frame rate returns. A return proves the competition layer is involved. It does not prove the option should stay off, and it does not prove the option was the only cause, because the injection layer can sit underneath the acceleration layer. The option is a trial, not a repair, and it is restored after the test where it was not the cause.

Why does exclusive full screen matter? The game slows there but not in a borderless window.

Because an overlay that wants to show its content on top of a game can push the game out of its exclusive state and onto the composited path, where the desktop window manager adds one more stage and the frame rhythm can worsen. The new Discord overlay works only in windowed and borderless modes, and does not support true exclusive full screen; only the older legacy overlay option supports it. So a game that slows only in exclusive full screen, and runs normally borderless, points at the display-mode layer. A game that slows in both modes clears the display mode, and the hunt moves back to the injection, drawing-cost, and acceleration layers. The mode comparison comes from the display-mode guide, and only that one move is used here.

My overlay will not open at all. Is that the same fault as a game slowing down?

No. An overlay that cannot be summoned is a different shape: its switch, its shortcut, or its permissions are at fault, and that shape belongs to its own guide. A game slowing while an overlay is on is a performance shape, and the two shapes have different checks. Lumping them together pushes the reader towards the wrong set of checks, because the test for a missing overlay cannot say anything about a slow game. Note whether the overlay opens, and then run the on/off comparison for the performance question on its own.

Should I update to the newest driver to fix an overlay slowdown?

No — not before the on/off comparison runs. An overlay problem is not a driver-version problem. The driver, the overlay, the anti-cheat system, and the game engine interact, and swapping the driver first throws away the material the test depends on: the overlay state, the baseline figures, and the scene exactly as it was. Run the on/off comparison first, then the one-at-a-time test, then the notification and acceleration comparisons. Swapping the driver ahead of the diagnosis ruins the test; once the diagnosis is done, a driver swap is a different decision, judged on its own checks.

When should I stop home testing and arrange an inspection?

Stop when every overlay is off and the game is still slow, with temperature or fan figures off, or the picture has gone corrupt, the screen has gone black, or the game has crashed. Bring the notes from the seven checks: the fixed scene and the baseline figures, the frame-rate figures with the overlays on and with them all off, which overlay if any brought the trouble back, the notification, acceleration, and display-mode results, and whether one game slows or every game does — together with the game name, the card model, driver release, and processor model. Those notes, along with the options as they stand now, are the starting point for an inspection.

Only one game slows down with the overlay on. What does that mean?

It points at the injection-conflict layer first: one overlay sits badly with that one game, its anti-cheat system, or its engine, while other games run normally with the same overlay on. Try other games the same way. If only one game slows and the rest run normally, the fault lies with that game or its engine, and the next move is to switch the guilty overlay off for that one game only. If every game slows, the fault sits with the overlay or the system, and the diagnosis returns to the notification, acceleration, and combined layers. The range test picks the road the diagnosis takes next, and skipping it is not an option.

Sources

The sources below fed into this guide. Repair guides are used here to explain what the fault does and how to test for it, not as things to follow. Numbers from them are given as source claims wherever the text says so. No figure from any source is treated here as a finding of this site.

  1. techtroduce.com, "Discord Overlay FPS Drops and Stutters - Test These Settings First" — https://www.techtroduce.com/discord-overlay-fps-drops-stutters-settings/ — an overlay draws its interface on every frame and competes with the game for graphics resources; the cost shows plainly when the card is close to full load; the symptom is a frame-rate fall, uneven frame times, and a hitch when a notification pops up; the test order is to switch the overlay off for all games first, then per game, then switch hardware acceleration off; a game that still hitches with Discord fully off is not an overlay problem.
  2. benchium.com, "Discord Causing FPS Drops? How to Fix" — https://benchium.com/discord-fps-drops-fix.html — Discord affects frame rate through four mechanisms: hardware acceleration, overlay rendering, voice processing, and notifications popping up; that site gives figures for the cost of each mechanism, which are that site's claims and are not repeated here as findings of this site; a notification popping up is said by that site to cause a frame pause, which is that site's claim and is not repeated here as a figure.
  3. tier1settings.com, "How to Optimize Discord for Gaming (Stop Lag & FPS Drops)" — https://tier1settings.com/best-discord-settings-to-reduce-lag/ — the Discord support position is to switch the overlay off per game when there is lag, a crash, or another problem; the new overlay is on by default for all games, works only in windowed and borderless full screen, and does not support true exclusive full screen, with the legacy overlay option as the one that supports it; switching the hardware acceleration off ends the program's competition with the game for graphics resources.
  4. usekudu.com, "How to Fix Discord Overlay Causing FPS Drops and Stuttering" — https://usekudu.com/guides/gaming/fix-discord-overlay-causing-fps-drops-and-stuttering — an overlay adds one more drawing layer on top of the game, and the cost shows most plainly on lower-end systems, older drivers, and games already pushed to their limit; hardware acceleration plus recording plus other overlays together make it worse; the test must switch overlays off one at a time, not all at once.
  5. thetechgorilla.com, "How To Disable Nvidia Overlay" — https://thetechgorilla.com/how-to-disable-nvidia-overlay/ — the NVIDIA overlay injects into the game's rendering pipeline in order to draw on top of it; that site's author gives a figure from a personal test on one card, which is a personal test and is not repeated here as a general figure; Instant Replay is the heaviest setting, encoding in the background; the NVIDIA App has replaced GeForce Experience, and the switch sits in a different place.
  6. blog.tempered.to, "Disable Overlays on Windows: Fast Steps for PC Gamers" — https://blog.tempered.to/blog/disable-overlays — switch overlays off one program at a time, for Steam, NVIDIA, Discord, AMD, and Xbox Game Bar, then stop leftover overlay processes in Task Manager; changing a setting while a game is running may not take effect until the game is restarted; several overlays on at once slow performance.
  7. rottenwifi.com, "How To Get Rid Of CPU Gpu Overlay" — https://rottenwifi.com/how-to-get-rid-of-cpu-gpu-overlay/ — the quickest way to tell overlays apart is the shortcut that hides each one; many overlays need the game restarted before a setting change fully unloads; shortcuts should not be pressed at random, because many overlays use global shortcuts.

A game that slows with an overlay on and runs normally the moment every overlay is off is not a judgement against overlays. Fix one scene, play the same sixty seconds with the overlays on and then with them all off, bring them back one at a time, test the notifications, test the hardware acceleration, and test the display mode. Those checks show whether the injection, the drawing cost, the acceleration, the display mode, or the combination is slowing the game — and they show it before any hardware is touched.

Related articles

Related services

If you need Chip-level GPU repair, see these service pages: