Game FPS Doesn't Increase After Enabling DLSS / FSR? Upscaling Not Working, Bottleneck & Frame Cap Diagnosis Guide

2026-11-14 | GPU Repair | 36 minutes | HK GPU Fix Editorial Team

Game FPS Doesn't Increase After Enabling DLSS / FSR? Upscaling Not Working, Bottleneck & Frame Cap Diagnosis Guide

Key Takeaway

Enabling DLSS or FSR does not always raise FPS: upscaling only lightens the pixel-shading work, so if the processor, asset streaming, or a frame cap is the real limit, the frame rate stays flat. Check the internal render resolution and GPU usage first — they show whether upscaling is running and where the bottleneck actually sits.

What This Problem Is — and What It Is Not

The same game, the same machine, the same scene. Before DLSS or FSR is turned on, the frame rate is a certain number. After it is turned on, the frame rate is almost exactly the same number — or it moves by one or two percent, which is inside measurement error. That is the whole problem this guide covers, and its shape is the most important piece of evidence in the diagnosis. The problem is defined by a before-and-after comparison, not by a low frame rate on its own.

A game that runs slowly with DLSS off and slowly with DLSS on is a different question from a game that runs slowly in general. A game whose picture turns blurry, smeared, or ghosted after upscaling is turned on is a different problem too, and it belongs to the guide on upscaling image quality. A game that crashes when ray tracing, DLSS, or frame generation is switched on, and runs normally when those features are switched off, is a crash-on-toggle problem, and it belongs to its own guide. A frame rate that falls after ray tracing is turned on is a ray tracing cost problem. A frame rate that falls after HDR is turned on is an HDR cost problem. A frame rate that suddenly drops sharply, with no before-and-after comparison at all, is a sudden FPS drop problem. None of those is this guide.

Four conditions must all be true before the steps below apply:

  1. The same game, the same scene, and the same graphics settings are used for both measurements, with DLSS or FSR off for one and on for the other, and the frame rate is measured in both cases rather than judged by feel.
  2. The frame rate after turning upscaling on is essentially unchanged from before — flat, or moving by only a couple of percent.
  3. The game is running and playable in both cases. If the game crashes, freezes, or shows a black screen, a different guide applies.
  4. The comparison is between upscaling off and upscaling on. A complaint that the frame rate is low in absolute terms, without that comparison, is not this problem.

The counter-intuitive core of this guide is that a flat frame rate does not mean upscaling is broken. Upscaling works by drawing the game at a lower internal render resolution and then reconstructing that smaller picture back up to the display's output resolution. The work it saves is the graphics card's pixel-shading work. But one frame of a game is more than pixel shading: the processor has to run the game logic, the physics, the AI, submit draw commands, and prepare the frame; assets have to be streamed in from the drive; and a frame cap or vertical sync can set a ceiling that nothing can rise above. None of that other work gets smaller because the render resolution went down. So a flat frame rate after enabling DLSS or FSR is, most often, a message about which part of the machine is actually setting the pace — and the job of this guide is to find which part that is, layer by layer, instead of dismissing the whole machine with a single sentence about it not being powerful enough.

The Before-and-After Comparison Is the Whole Diagnosis

Nothing in this guide works without a proper comparison, and a proper comparison has rules. Measure in the same scene, because different scenes put different loads on the processor and the graphics card: a quiet corridor and a crowded battle are not the same test, and comparing one against the other proves nothing. Keep every other graphics setting identical, because changing a quality preset at the same time as turning upscaling on mixes two changes together, and the result cannot be attributed to either one. Use the game's built-in frame rate counter or a monitoring tool, and write both numbers down, together with the graphics card usage in each case. A note that says "off: high FPS, GPU near full; Quality: same FPS, GPU lower" is a diagnosis in progress. A note that says "it did not feel faster" is a starting point with no direction.

The comparison also needs to be repeated in the same way each time a setting is changed later in this guide. Change one thing, measure again, write the result down. Change two things at once and see the frame rate move, and the reader will not know which change mattered — and will not know what to do when the problem returns. That discipline is the difference between a diagnosis and a series of guesses, and it applies to every step below.

One more point about measurement error deserves stating plainly. Frame rates fluctuate from second to second even in a fixed scene, as the game loads assets, as background activity changes, and as the scene itself moves. A difference of one or two percent between two measurements is noise, not an improvement and not a failure. Only a sustained, repeatable difference, measured in the same scene under the same settings, counts as a real change. If the numbers are within that noise band in both directions, the honest reading is that the frame rate did not move.

The Five Suspicion Layers

A flat frame rate after enabling upscaling can come from five layers, and they must be kept separate. Mixing them together is how a frame cap problem gets treated by dropping the upscaling preset again and again, or a game that never actually engaged DLSS gets treated as a processor bottleneck, and how the real cause survives because the wrong layer was treated. The five layers below are ordered the way the diagnosis should work through them, and each one has its own deciding test.

Layer One: The Bottleneck Layer — Upscaling Only Lightens the Pixel-Shading Work

The first layer is the bottleneck, and it is the layer most often misunderstood. Super resolution upscaling — DLSS Super Resolution and FSR Super Resolution — works by rendering at a lower internal resolution and reconstructing the picture back to the output resolution. The saving is in the graphics card's pixel-shading work: fewer pixels to shade means less work for the graphics card per frame. But a frame is not only pixel shading. The processor must run the game logic, the physics, the artificial intelligence, submit the draw commands, and prepare each frame. Assets must be streamed from the drive. A frame cap or vertical sync can set a ceiling. None of those jobs shrinks when the render resolution drops.

That is why a flat frame rate is not, on its own, evidence that upscaling failed. If the graphics card was never the part setting the pace — if the processor was already the slowest step in the chain, or the game was already hitting a frame cap — then lightening the graphics card's work changes nothing about the pace, because the pace was never set by the graphics card. The frame rate stays where it was, and the graphics card simply finishes its lighter work sooner and waits longer. The deciding evidence for this layer is graphics card usage, dealt with in Step 4: a graphics card running at 95–99% is the bottleneck, and upscaling should help; a graphics card sitting well below that while the frame rate is flat means the graphics card is waiting on something else.

A published report by XDA Developers illustrates the point with a concrete case: a machine with a high-end graphics card and an older processor, running at 1440p, showed graphics card usage of only around 80%, and enabling DLSS barely moved the frame rate. The report's conclusion, in its own words, was that DLSS only helps when the graphics card is the bottleneck, and that enabling DLSS when the graphics card already had headroom only shifted the bottleneck further toward the processor. That is the same mechanism described here, seen from the other side: the upscaling worked, in the sense that it reduced the graphics card's work, but the frame rate could not rise because the graphics card was not the part holding it back. A second XDA Developers report makes the same point from a different angle: the same high-end card at 4K with DLSS Quality, where the graphics card was running near 99%, showed a clear gain, while at 1440p, where the internal render resolution dropped to 960p and the graphics card was no longer busy, the gain was small. The difference was not the upscaling. The difference was which part of the machine was setting the pace at each resolution.

Layer Two: The Verification Layer — "Upscaling Is On" Does Not Mean Upscaling Is Running

The second layer is verification, and it is the layer that must be cleared before any bottleneck conclusion is drawn. Having a tick in a menu is not the same as the upscaling actually running, and there are several ways the two can come apart. The game may actually be running DLAA, which is native-resolution anti-aliasing and is designed not to raise the frame rate at all. The game may not support DLSS at all — NVIDIA's own documentation notes that for games without NVAPI support, the DLSS options simply do not appear in the game — so an option that looks present may belong to a different feature, or the setting may have been changed by a driver-level override. Dynamic resolution scaling may be running at the same time, pulling the render resolution back up and cancelling the reduction that upscaling was supposed to provide. Any one of these leaves the menu saying one thing and the renderer doing another.

Verification therefore cannot rest on the menu. It must rest on two measurements: the internal render resolution and the graphics card usage. After upscaling is turned on, the internal render resolution should fall by the amount the chosen preset specifies, and the graphics card usage should fall with it. The preset scale, as documented in the upscaling guides and in the build-pc-guide reference on DLSS, is roughly two-thirds of each side at Quality — about 66.7% of the width and height — and about half of each side at Performance, with Ultra Performance at about one-third of each side. Those are per-side figures, not total pixel counts, and they matter because they set the expectation: if the internal render resolution has not fallen, upscaling is not running, whatever the menu says. If the internal render resolution has fallen and the graphics card usage has fallen but the frame rate has not moved, then upscaling is running and the problem is a bottleneck or a cap, which is Layer One or Layer Three. That fork is the single most important branch in this guide, and it is decided in Step 2.

There is a developer diagnostic switch published by NVIDIA that can display on screen whether DLSS is active. It exists, and it is a developer tool. It is mentioned here only so that its existence is not a mystery; the details of enabling it are not part of this guide, and the verification in this guide uses only the game's own options and monitoring tools — internal render resolution, graphics card usage, and the on-versus-off frame rate comparison.

Layer Three: The Cap Layer — Frame Caps and Vertical Sync Eat the Gain

The third layer is the cap. If the game is already running at a frame cap — a built-in game cap, a driver-level cap, or a cap set by a frame-limiting tool such as RTSS — or if vertical sync has locked the frame rate to the display's refresh rate, then the frame rate was already at its ceiling before upscaling was turned on, and it will still be at its ceiling afterwards. Upscaling has not failed. Its gain has been eaten by the ceiling. The evidence for this layer is a frame rate that sits exactly at a round number — the refresh rate, or a cap value — combined with graphics card usage that falls when upscaling is turned on while the frame rate does not move. A falling graphics card usage with a flat frame rate is the signature of this layer: the graphics card is doing less work per frame, but the frame rate cannot rise because something else has set the ceiling.

This layer is easy to miss precisely because a capped frame rate looks like a healthy one. A game locked at 120 FPS on a 120 Hz display looks smooth and finished, and there is nothing in the picture to suggest that upscaling did anything at all. But the graphics card usage tells the story: if it fell when upscaling was turned on, the upscaling did its job, and the ceiling is the reason the number did not move. The cap must be checked separately from the bottleneck, because a cap and a bottleneck can exist at the same time, and each must be confirmed on its own before either is treated. Checking the cap is Step 5, and it deliberately comes after the bottleneck judgement, not before it.

Layer Four: The Frame Generation Confusion Layer — Super Resolution and Frame Generation Are Different Functions

The fourth layer is a confusion between two features that share a name and do very different things. DLSS Super Resolution and DLSS Frame Generation are separate functions, and NVIDIA's own documentation lists them as such, alongside Ray Reconstruction and DLAA. Super Resolution lightens the rendering load, and the frames it produces are real rendered frames, prepared by the processor and the graphics card together. Frame Generation synthesises intermediate frames between already-rendered frames, and the frames it adds are displayed frames, not rendered ones. It raises the number on the frame rate counter, but it does not speed up the game's own simulation, and the responsiveness of the controls still follows the base frame rate — the rate at which real frames are actually being prepared.

That distinction matters in both directions. Turning on frame generation and seeing the number rise does not mean the upscaling problem is solved, because the number that rose is a different number: the displayed frame count, not the rendered frame rate, and the underlying stutter and uneven frame pacing remain where they were. Conversely, turning on super resolution and seeing the number stay flat cannot be "fixed" by turning on frame generation to make the number move, because the two features are not measuring the same thing, and papering over a flat rendered frame rate with generated display frames leaves the real bottleneck untouched. Frame generation also needs a reasonable base frame rate to work well; when the base rate is low or unstable, generated frames only make the situation worse, because the gaps they are filling are themselves uneven. The correct question is always about the rendered frame rate first, and frame generation is a display-smoothing tool, not a cure for a bottleneck.

Layer Five: The Scene Layer — The Gain Varies with Resolution, Scene, and Game, and There Is No Universal Number

The fifth layer is the scene, and it is the layer that forbids any universal claim about how much upscaling should gain. The same graphics card behaves differently at different output resolutions. At 4K output, DLSS Quality still renders internally at 1440p, which is a heavy load, so the saving from upscaling is large and the gain is usually clear. At 1080p output, the Performance preset renders internally at only 960 by 540, and the graphics card was already far from busy, so the saving is small and the gain is naturally small. Competitive games, open scenes with little geometry, and processor-led scenes show smaller gains for the same reason: there was never much pixel-shading work to save. A game that is processor-led will show almost no gain at any resolution, because the processor, not the pixel shading, sets the pace.

No universal frame rate gain percentage or multiplier can be written for upscaling, because the gain depends on the bottleneck, the resolution, the scene, and the game. A marketing-style figure that promises a fixed percentage for every game on every card is not a measurement; it has no method, no named game, no named resolution, and no named card behind it, and it contradicts published reports in which the same card gained only slightly at 1440p while gaining clearly at 4K. The honest statement is that the gain varies, and the only way to know the gain for a particular game on a particular machine is to measure it, in that game, in that scene, with the method in Step 1. Any source that quotes a single figure for all games should be set aside.

LayerWhat can be wrongDeciding evidenceWhat resolves it
1. BottleneckThe processor, simulation, or asset streaming sets the pace, not pixel shadingGraphics card usage well below 95% while FPS is flatIdentifying the real bottleneck; upscaling is not the fix
2. VerificationDLAA selected, dynamic resolution cancelling the reduction, or no DLSS supportInternal render resolution and GPU usage have not fallenCorrecting the mode, turning off dynamic resolution, confirming support
3. CapA frame cap or vertical sync has set a ceilingFPS pinned at a round number; GPU usage falls but FPS does not moveRecognising the ceiling; the gain was eaten, not absent
4. Frame generation confusionFrame generation mistaken for super resolutionThe displayed frame count rises while responsiveness follows the base rateJudging by the rendered frame rate, not the display counter
5. SceneLow output resolution, a light scene, or a processor-led game leaves little to saveThe same card gains clearly at 4K and barely at 1080pMeasuring per game and per scene; no universal gain figure exists

The Seven Diagnostic Steps

The seven steps below are ordered so that each one tests one layer, and so that no step destroys the evidence a later step needs. Work through them in order, one variable at a time, and write down the result of each step before moving on. Changing two things at once and seeing the frame rate move leaves the reader not knowing which change mattered — and not knowing what to do when the problem returns.

Step 1: Measure FPS in a Fixed Scene, Upscaling Off Versus On

Use the same game, the same scene, and the same graphics settings, and measure the frame rate once with DLSS or FSR off and once with it on, using the game's built-in counter or a monitoring tool. Record the graphics card usage alongside each frame rate figure. Do not change the scene between the two measurements, and do not change any other setting. Do not compare one measurement taken in a menu against another taken in a battle scene.

This step proves that the problem has been narrowed to a flat before-and-after comparison, and it records the only direct evidence of how flat the frame rate really is. It proves nothing about why the frame rate is flat — that is what the later steps are for. But without two numbers taken under identical conditions, nothing else can be checked against them. A difference of one or two percent between the two measurements is measurement noise, not a result, and it should be read as "the frame rate did not move". If the two numbers differ by a large, repeatable margin, then the problem described in this guide is not present, and the frame rate did move after all.

Step 2: Confirm Upscaling Is Actually Running — Internal Render Resolution and GPU Usage

After upscaling is turned on, check the internal render resolution and the graphics card usage. At the Quality preset, the internal render resolution should fall to roughly two-thirds of each side; at Performance, to roughly half of each side. The graphics card usage should fall with it. If the internal render resolution has not fallen and the graphics card usage has not fallen, upscaling is not running at all — go to Step 3, and do not jump to a bottleneck conclusion. If both have fallen but the frame rate has not moved, upscaling is running and the problem is a bottleneck or a cap — go to Step 4.

This step is the fork the whole diagnosis turns on, and it must not be skipped. A flat frame rate with a fallen internal render resolution and a fallen graphics card usage is a completely different problem from a flat frame rate with an unchanged internal render resolution, and the two have opposite treatments. The first means the upscaling did its job and something else is setting the pace. The second means the upscaling never engaged, and the settings must be corrected before any bottleneck is discussed. Confusing the two is the most common error in this diagnosis, and it is the error this step exists to prevent.

Step 3: Split DLAA, Dynamic Resolution, and Game Support

Look at what the game option actually selected. If the option is DLAA, the game is running native-resolution anti-aliasing, which is designed not to raise the frame rate — DLAA uses upscaling technology for anti-aliasing at native resolution, as NVIDIA's documentation puts it, and a native-resolution render cannot gain from a lower render resolution that never happened. Switch it back to a DLSS Super Resolution preset and measure again. If dynamic resolution scaling is on at the same time, turn it off and measure again: dynamic resolution decides how much resolution to compute with, and upscaling decides how to reconstruct the computed picture, and the two can stack — but dynamic resolution can pull the render resolution back up and cancel the reduction that upscaling was supposed to provide. If the game genuinely does not support DLSS, the option will not appear at all, and no driver setting can add it to a game that lacks support.

Change one thing at a time in this step. Turn dynamic resolution off and measure. Switch DLAA back to DLSS Quality and measure. Each change gets its own measurement, and each measurement is written down. If any one of these changes makes the frame rate move, the cause was the setting, not the hardware, and the diagnosis ends here.

Step 4: Read the GPU Usage — Which Part Is the Bottleneck

This is the most important fork in the guide, and its judgement standard is graphics card usage, not processor total usage. If the graphics card usage stays at 95–99% for a sustained period, the graphics card is the bottleneck, and upscaling should help — so go back to Step 2 and re-verify, because a graphics-card bottleneck with a flat frame rate usually means upscaling never actually engaged. If the graphics card usage sits clearly below 95% — hovering around 80%, for example — while the frame rate stays flat, the graphics card is waiting on something else: the processor, the simulation, or asset streaming is setting the pace.

Processor total usage is not the deciding evidence here, and it must not be used as a substitute. A processor bottleneck does not mean the processor is running at 100%: a single core can be fully loaded, with the game's main thread pinned, while the total across all cores looks modest. That is why the graphics card usage is the test that matters. A graphics card below 95% with a flat frame rate is the direct evidence of a bottleneck elsewhere, whatever the processor total says. The figure.rocks guide on this subject makes the same point in its own terms: lowering graphics settings does nothing when the game is not graphics-card-limited, and a processor-bound game does not mean the processor is at 100%. The same logic applies to upscaling, which is, in effect, a way of lowering the render resolution: if lowering the render load does not move the frame rate, the game was not graphics-card-limited in the first place, and the test itself has delivered the answer.

Step 5: Rule Out the Other Caps One by One — Frame Cap, Vertical Sync, Background Programs

Check the game's built-in frame cap, the driver-level cap, and any cap set by a frame-limiting tool, and check whether vertical sync has locked the frame rate to the display's refresh rate. Check also whether background programs are consuming resources: recording, streaming, or a browser with hardware acceleration running alongside the game. Turn one thing off at a time and measure again after each change.

This step comes after the bottleneck judgement, not before it, and it must not be used as the opening move. A cap and a bottleneck can exist at the same time, and they must be confirmed separately. A frame rate that sits exactly at the refresh rate or at a round cap value, combined with graphics card usage that falls when upscaling is turned on, is the signature of a cap eating the gain. A background program consuming processor time can look like a processor bottleneck, and it must be ruled out before the bottleneck is treated as a property of the game itself. Each exclusion gets its own measurement, so that when the frame rate finally moves, the change that moved it is known.

Step 6: Split One Game from Every Game

If only one game shows no improvement from upscaling, the cause belongs to that game: its upscaling integration, its dynamic resolution setting, or the fact that the game itself is processor-led. The next step for a single-game problem is to check that game's upscaling version and settings, and to repeat Steps 1 to 3 inside that game alone. If every game shows the same flat result, the cause is at the system level: a processor bottleneck, a frame cap, or a driver setting that applies across games. The next step for a system-level problem is the bottleneck method — the same graphics card usage test, applied across games, to find which part of the machine is setting the pace everywhere.

This step decides the direction of everything that follows it. A single-game problem that is treated as a system problem wastes effort on the system, and a system problem that is treated as a single-game problem wastes effort on one game's settings. The split must be made with measurements, not with a guess: the same before-and-after comparison, run in at least two different games, one of which is known to be graphics-card-led at the resolution being tested.

Step 7: Decide — and Record What Happened

The decision follows from the steps, and there are three directions.

If upscaling was never actually running — the cause found in Steps 2 and 3 — deal with the cause: switch DLAA back to a DLSS Super Resolution preset, turn dynamic resolution off, and confirm that the game supports DLSS. Then repeat Step 1 and confirm that the frame rate now moves as expected.

If the graphics card usage is low and the bottleneck is in the processor, upscaling is not the fix. The correct direction is to reduce settings that load the processor — object density and view distance are the usual examples — or to accept the current frame rate, and not to keep dropping the upscaling preset: dropping from Quality to Performance only reduces the pixel-shading work further, and it will not make the processor faster; it will only make the picture softer. Whether to change hardware is the reader's own decision, and this guide does not make it for them. A diagnosis that ends with "the bottleneck is in the processor" is a complete diagnosis, and the next step, if any, belongs to the reader.

If a frame cap has eaten the gain, upscaling is not ineffective — its gain was absorbed by the ceiling, and the frame rate was already at its limit before upscaling was turned on. The decision there is about the cap itself: whether the cap is wanted, and whether the lower graphics card usage that upscaling delivered is itself useful, for example as cooler and quieter running at the same frame rate.

Bring the machine in with the complete record when the steps have been worked through and the answer is still unclear, or when the question has become which part to change rather than which layer is at fault: the on-versus-off frame rate comparison, the graphics card usage, the internal render resolution, the game, and the preset used. Inspection takes place at Room 1209, Dali Plaza, Shek Mun. You can call 2186 8010, or reach the shop on WhatsApp at 6558 6806. Inspection is charged at the site's published three-tier fees of HK$400–2,800, quoted after inspection. There is no home-visit service and no Mong Kok shop; inspection is carried out in the shop.

StepWhat to doWhat it provesWhat it does not prove
1. Measure in a fixed sceneSame game, scene, and settings; FPS off versus on, with GPU usageWhether the frame rate really is flatWhy it is flat
2. Confirm upscaling is runningInternal render resolution and GPU usage after turning it onWhether upscaling engaged at allWhere the bottleneck is, on its own
3. Split DLAA, dynamic resolution, supportSwitch DLAA to DLSS, turn dynamic resolution off, confirm support; one change at a timeWhether a setting prevented upscaling from runningAnything about the bottleneck
4. Read the GPU usageSustained GPU usage: 95–99% versus clearly below 95%Which part sets the paceWhich processor setting to change
5. Rule out the capsGame cap, driver cap, frame-limiting tool, vertical sync, background programs; one at a timeWhether a ceiling ate the gainAnything about the game itself
6. Split one game from every gameRepeat the comparison in a second gameWhether the cause is per-game or system-levelWhich setting, on its own
7. DecideFix the setting, accept a processor bottleneck, or recognise a cap; bring the record inWhich direction follows from the evidenceAnything beyond the test that was actually run

Six Common Misconceptions

Misconception 1: Turning on DLSS means the FPS must rise

No. Upscaling only lightens the pixel-shading work. When the processor, the simulation, asset streaming, or a frame cap is what sets the pace, the frame rate will not move. The gain varies with resolution, scene, and game, and there is no universal number for it. A flat frame rate after enabling upscaling is a measurement result, not a verdict on upscaling, and it must be read together with the graphics card usage before any conclusion is drawn.

Misconception 2: A flat FPS means DLSS never turned on

No. A flat frame rate can mean upscaling is running but the bottleneck is elsewhere — a fallen graphics card usage is the evidence for that — or it can mean upscaling never actually engaged, with the internal render resolution unchanged. The two must be separated by Step 2. A single frame rate number, on its own, cannot tell them apart, and a conclusion drawn from that number alone is a guess.

Misconception 3: The processor usage is not high, so it cannot be a processor bottleneck

No. A processor bottleneck does not mean the processor is running at 100%. A single core can be fully loaded, with the game's main thread pinned, while the total across all cores looks modest. The direct evidence is the graphics card usage sitting below 95% while the frame rate stays flat. Judging the processor by its total usage is judging it by the wrong number, and it is the error that keeps a processor bottleneck hidden behind a calm-looking total.

Misconception 4: Dropping one more preset will make the FPS rise

No. Dropping from Quality to Performance only reduces the pixel-shading work further. When the bottleneck is in the processor, dropping the preset will not make the processor faster; it will only make the picture softer. Before dropping a preset, confirm where the bottleneck is. A preset change is a graphics card change, and it can only help a graphics card problem.

Misconception 5: Turning on frame generation and seeing the FPS rise means the problem is solved

No. Frame generation raises the displayed frame count, not the real rendered frame rate. The responsiveness of the controls still follows the base frame rate, and uneven frame pacing and stutter remain where they were. Frame generation is a display-smoothing tool, not a cure for a bottleneck, and it needs a reasonable base frame rate to work well — when the base rate is low or unstable, generated frames only make the situation worse.

Misconception 6: DLAA and DLSS Quality are different presets of the same function

No. DLAA is native-resolution anti-aliasing, and it is designed not to raise the frame rate. DLSS Quality renders at a lower internal resolution and reconstructs the picture back up. Selecting DLAA by mistake and then concluding that DLSS does not work is a verification error, not a failure of DLSS, and it is corrected in Step 3 by switching back to a Super Resolution preset and measuring again.

Boundaries: Problems That Look Similar but Are Not This One

Not every upscaling problem is a flat frame rate, and the similar problems each belong to their own guide. If the symptom is one of the following, follow that guide, not this one.

A picture that turns blurry, smeared, or ghosted after upscaling is turned on is an upscaling image quality problem. A game that crashes when ray tracing, DLSS, or frame generation is switched on, and runs normally when they are switched off, is a crash-on-toggle problem. A frame rate that falls after ray tracing is turned on is a ray tracing cost problem. A frame rate that falls after HDR is turned on is an HDR cost problem. A frame rate that suddenly drops sharply, with no before-and-after comparison, is a sudden FPS drop problem. A picture that is sometimes sharp and sometimes blurry is a dynamic resolution problem, and dynamic resolution appears in this guide only as one branch of Step 3. High latency or ghosting after frame generation is turned on is a frame generation problem, and frame generation appears in this guide only as a confusion to be cleared in Layer Four. The general question of whether the bottleneck is in the processor or the graphics card, without any upscaling comparison, is a bottleneck guide in its own right; this guide applies that method to the upscaling on-versus-off comparison, and it does not repeat the full bottleneck method.

Finally, there are things this guide will never teach, and a source that teaches them should be closed. Never download or replace DLSS DLL files from the internet. Never install an unofficial upscaling upgrade tool. Never edit the registry to force a diagnostic display. None of these is a diagnosis, and each one trades a working machine for a less safe one. The verification in this guide uses only the game's own options and monitoring tools.

When to Stop and Bring the Machine In

Stop when all seven steps have been worked through, one variable at a time, and the same flat result appears every time — or when the question becomes which part to change, rather than which layer is at fault. Stop sooner if the diagnosis points at a processor bottleneck and the next question is a hardware decision, because that decision is the reader's own, and it should be made with the full record in hand rather than on a single frame rate number.

Bring the machine in with the notes from every step: the on-versus-off frame rate comparison in a fixed scene, the graphics card usage, the internal render resolution, the game, and the preset used. In-store inspection takes place at Room 1209, Dali Plaza, Shek Mun. You can call 2186 8010, or reach the shop on WhatsApp at 6558 6806. Inspection is charged at the site's published three-tier fees of HK$400–2,800, quoted after inspection. There is no home-visit service and no Mong Kok shop; inspection is carried out in the shop.

A diagnosis that ends at the stop line is not a failed diagnosis. A note that says "off: high FPS, GPU near full; Quality: same FPS, GPU lower; internal render resolution fell as expected; cap ruled out; same result in two games" is a precise description of where the fault lives, and it is exactly what an in-store inspection needs to start from the right place instead of from the beginning.

Frequently Asked Questions

Q1: I enabled DLSS and the FPS did not move at all — does that mean DLSS is broken?

Not necessarily. A flat frame rate does not mean upscaling failed. Upscaling only lightens the pixel-shading work, so if the processor, the simulation, asset streaming, or a frame cap is what sets the pace, the frame rate stays flat even while upscaling is running. Check the internal render resolution and the graphics card usage: if both fell when upscaling was turned on, upscaling is running and the bottleneck is elsewhere. Only if neither fell is upscaling not running at all.

Q2: How can I tell whether DLSS is actually running?

Do not judge by the menu tick. Check two measurements after turning it on: the internal render resolution and the graphics card usage. At the Quality preset the internal render resolution should fall to roughly two-thirds of each side, and at Performance to roughly half of each side, and the graphics card usage should fall with it. If the internal render resolution has not fallen and the graphics card usage has not fallen, upscaling is not running, whatever the menu says — and the cause is usually DLAA, dynamic resolution, or a lack of game support, dealt with in Step 3.

Q3: What GPU usage figure tells me where the bottleneck is?

Sustained graphics card usage of 95–99% means the graphics card is the bottleneck, and upscaling should help — so a flat frame rate at that usage usually means upscaling never engaged, and Step 2 should be repeated. Graphics card usage clearly below 95%, hovering around 80% for example, while the frame rate stays flat, means the graphics card is waiting on something else: the processor, the simulation, or asset streaming is setting the pace. That usage figure is the deciding test, and it is more reliable than the processor total usage.

Q4: My CPU usage is not high — why could it still be a CPU bottleneck?

Because a processor bottleneck does not mean the processor is running at 100%. A single core can be fully loaded, with the game's main thread pinned, while the total across all cores looks modest. The direct evidence is the graphics card usage sitting below 95% while the frame rate stays flat. Judge the bottleneck by the graphics card usage, not by the processor total, and do not rule out a processor bottleneck just because the total looks calm.

Q5: What is the difference between DLAA and DLSS Quality? What happens if I pick the wrong one?

DLAA is native-resolution anti-aliasing: it renders at the full output resolution and uses upscaling technology only to smooth edges, so it is designed not to raise the frame rate. DLSS Quality renders at a lower internal resolution — roughly two-thirds of each side — and reconstructs the picture back up, so it reduces the graphics card's work and can raise the frame rate when the graphics card is the bottleneck. Picking DLAA by mistake and then concluding that DLSS does not work is a verification error: switch back to a DLSS Super Resolution preset and measure again.

Q6: If dynamic resolution is on at the same time, can it cancel out DLSS?

Yes. Dynamic resolution decides how much resolution to compute with, and upscaling decides how to reconstruct the computed picture, and the two can stack — but dynamic resolution can pull the render resolution back up and cancel the reduction that upscaling was supposed to provide. If both are on, turn dynamic resolution off, change nothing else, and measure again. That single change, with its own measurement, shows whether dynamic resolution was the cause.

Q7: If a frame cap or VSync has locked the FPS, is enabling DLSS still useful?

It can be, even though the frame rate number will not move. If the frame rate is already at its ceiling, upscaling cannot raise it above the ceiling — but the graphics card usage will fall, because the graphics card is doing less work per frame to hold the same frame rate. That lighter load can mean cooler and quieter running at the same frame rate. The upscaling did its job; its gain was absorbed by the ceiling, and the ceiling, not the upscaling, is what holds the number still.

Q8: Frame generation shows a higher FPS — does that mean the problem is solved?

No. Frame generation raises the displayed frame count by synthesising intermediate frames between rendered ones; it does not speed up the game's own simulation, and the responsiveness of the controls still follows the base frame rate — the rate at which real frames are prepared. Uneven frame pacing and stutter remain where they were. Frame generation is a display-smoothing tool, not a cure for a bottleneck, and it needs a reasonable base frame rate to work well.

Q9: If I drop one more quality preset (Quality to Performance), will the FPS rise?

Only if the graphics card is the bottleneck. Dropping from Quality to Performance reduces the pixel-shading work further, so it helps when the graphics card sets the pace. When the bottleneck is in the processor, dropping the preset will not make the processor faster; it will only make the picture softer. Confirm where the bottleneck is first, with the graphics card usage test in Step 4, before dropping a preset — a preset change is a graphics card change, and it can only help a graphics card problem.

Q10: What is the difference between only one game not improving and every game not improving?

If only one game shows no improvement, the cause belongs to that game: its upscaling integration, its dynamic resolution setting, or the fact that the game itself is processor-led. Check that game's upscaling version and settings, and repeat the before-and-after comparison inside that game. If every game shows the same flat result, the cause is at the system level: a processor bottleneck, a frame cap, or a driver setting that applies across games. That split decides the direction of everything that follows, so it should be made with measurements in at least two games, not with a guess.

Related articles

Related services

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