GPU Shows 0% Utilization While Gaming? Task Manager Engine Readings, Wrong-GPU & False-Zero Diagnosis Guide

2026-11-10 | GPU Repair | 20 minutes | HK GPU Fix Editorial Team

GPU Shows 0% Utilization While Gaming? Task Manager Engine Readings, Wrong-GPU & False-Zero Diagnosis Guide

GPU Shows 0% Utilization While Gaming? Task Manager Engine Readings, Wrong-GPU & False-Zero Diagnosis Guide

Key Takeaways

A 0% GPU reading while a game runs smoothly is usually a reading problem, not a faulty card. Task Manager's figure is the busiest engine, not an average, so a game on the 3D engine and a video on the decode engine give different numbers. Check the right card, the right engine, and the right moment before suspecting the hardware.

First, Check Which Problem This Is

The game is running. The picture is moving, the frame rate looks normal, and nothing feels wrong — except that Task Manager, or a monitoring program, says the GPU is at 0%. The reading and the reality do not match: the picture is clearly being drawn, yet the number says the card is doing nothing.

That mismatch is the whole subject of this guide, and the entry condition has to be stated plainly at the start, because there is a second problem that looks similar and needs a different guide. The table below draws the line.

SituationWhat is happeningWhich guide
The game runs smoothly, the frame rate is normal, but the reading is 0%A reading problem: the wrong card, the wrong engine, a frame cap, or the wrong momentThis guide
The game runs slowly, the picture stutters, the dedicated GPU sits near 0% while the integrated graphics carries loadThe game is running on the wrong graphics processor — a card-selection problem on a laptopThe laptop card-selection guide (#150), not this one
Nothing is open, yet the reading is highA background-load problem, the reverse of this oneThe idle-load guide (#255), not this one
The utilisation is high but the frame rate is low, or the clock speed does not drop back to idleA bottleneck or a utilisation-versus-clock problem — the reverse of this guideThe bottleneck guide and the utilisation-versus-clock guide, not this one
The reading is 0% and the game cannot draw its first frameA device-initialisation failureThe game initialisation guide (#393), not this one

If the game is slow at the same time as the reading is 0%, this guide is the wrong one, and every step that follows will point in the wrong direction. The rest of this guide assumes the game runs fine and only the number is wrong. Its aim is to show that utilisation is not a single number but a set of readings, and that 0% has several completely different meanings. Without separating them, a normal reading gets mistaken for a dead card.

Utilisation Is Not One Number: What Task Manager Actually Measures

The facts in this section come from the Microsoft DirectX engineering team's official blog post, "GPUs in the task manager". Task Manager's GPU data comes directly from the Windows graphics core — the GPU scheduler (VidSch) and the video memory manager (VidMm). The numbers are not estimates. But they are also not a single "how busy is the card" figure, and that distinction is where most confusion starts.

First, the driver has to support WDDM 2.0 or a newer version before Task Manager will show GPU data at all. On an older WDDM 1.x driver, Task Manager shows no GPU data. So a missing GPU section in Task Manager can itself be a driver-version clue, not a card fault.

Second, a GPU contains several engines that can work in parallel. They are independent computing units: the 3D engine, the Copy engine, Video Decode, Video Encode, and Compute. The Copy engine can be moving data while the 3D engine is drawing a scene, and neither waits for the other.

Third, and most importantly, the overall percentage Task Manager shows is the busiest engine, not an average. Microsoft's own reasoning is straightforward: if the figure were an average, then on a card with 10 engines, a game that fully loads the 3D engine would still show only 10% overall, which would make the card look idle when it is not. So Microsoft chose the busiest engine's figure as the overall percentage.

That single rule explains the core phenomenon of this guide. When a video is playing, the Video Decode engine can be busy while the 3D engine reads 0% — Microsoft's blog uses exactly this as its example, noting that different video formats can use different engines for decoding. Staring at the 3D chart while a video plays, and concluding the GPU is at 0%, is reading the wrong engine.

The Performance page shows only 4 engines by default, but clicking an engine name switches the view to the other engines. The Processes page has a GPU engine column, which states which card and which engine a program is using — for example, "GPU 1 – 3D". That column is the direct evidence for which program is using which card, and it matters more than the overall percentage.

Different Jobs Use Different Engines

The table below is the engine map for this guide. A game normally runs on the 3D engine, but not every job that uses the GPU runs on 3D, and that is the point.

EngineMain workWhich everyday scene uses it
3DGraphics rendering, drawing scenesGame picture, three-dimensional scenes, desktop effects
Compute / CUDAGeneral-purpose computingAI computing, image processing, local compute programs
CopyMoving data between memoryLarge data transfers, capture and preview generation
Video DecodeDecoding videoPlaying video, video in browser tabs, stream previews
Video EncodeEncoding videoRecording, live streaming, screen capture

The key distinction is simple: a game normally runs on the 3D engine, but a video, a compute job, or a recording does not. Judging a video by the 3D chart, or a compute job by the 3D chart, produces a 0% answer that is wrong. Public guides agree on the remedy: when the number looks unusually low, click the engine name and switch to the engine the program is actually using.

A Frame Cap: Finished the Frame, Then Waiting

When a game has a frame-rate cap — VSync switched on, an in-game frame-rate limit, or a driver-level frame cap — the GPU finishes a frame and then has to wait for the next display cycle. During that wait, the GPU genuinely has nothing to do, and a reading taken at that moment can be 0%, or close to it. That is not a fault. It is the normal side effect of a frame cap.

The test is direct: switch the cap off and look again. If the reading rises when the cap is off, the earlier 0% was waiting time, not a card doing nothing. The comparison should also note the frame rate at the same time: frame rate rising together with the reading is consistent. This step is a comparison only, not an instruction to leave VSync off or to run without a cap permanently. Once the comparison is done, the original setting should be restored.

The Game Is Not Busy: Menus, Pauses, and the Background

A game sitting on its main menu, on a pause screen, or on a loading screen does not need to draw a complex scene every frame, so the GPU load is genuinely low and a low reading is normal. The same applies after Alt-Tabbing out of the game, shrinking the game window, or leaving the game in the background: Windows pauses or sharply reduces the rendering work of a background program. A reading taken in that state measures the backgrounded game, not the game itself.

The rule is to make the program do the work that is being checked before reading the number: load into an actual game scene, start playing a video, start a render. An idle program will not show meaningful GPU activity, and a reading taken from an idle program proves nothing about the card. Public guides state this plainly, and it is the cheapest step in this guide — and the one most often skipped. Many "0%" readings disappear on their own the moment the reader steps into an actual game scene.

Different Tools, Different Numbers — Neither One Is Broken

NVIDIA's `nvidia-smi` defines its `utilization.gpu` figure as the percentage of time, over the sampling period, during which a kernel was executing. Task Manager defines its overall figure as the busiest engine. Those are different definitions, over different sampling windows, so the two tools can show different numbers for the same machine at the same time without either one being broken. A difference between tools is expected, not evidence of a fault.

Monitoring software also samples, and different sampling frequencies draw different-looking curves from the same activity. The practical rule is to fix one tool and one way of observing, and use it for before-and-after comparisons. Jumping between tools and comparing absolute numbers across them is not a valid comparison. The numbers are useful for showing change, not for chasing one absolute "correct" reading.

There is a related possibility worth one line: the driver is responsible for reporting GPU state to the system, and if the driver version and the monitoring tool do not line up, the reported reading can drift away from the actual work. The way to tell is to cross-check with a different monitoring tool and to look at the physical signs — temperature, fan speed, and power draw — to see whether they move with the game load. That cross-check is step five below.

Utilisation and Clock Speed Are Two Separate Readings

Utilisation says how busy the graphics card is at the moment. Clock speed says which working state the card is running in at the moment. The two have to be read separately and never mixed together. A 0% utilisation reading with a normal clock speed points in a different direction from a 0% utilisation reading with a clock speed locked low: the first leans toward a reading problem, the second toward a state-change problem, which is the subject of a different guide. Keeping the two readings apart is part of reading the numbers correctly.

A Six-Step Diagnosis

Step one: confirm the game is actually running — do not read the number from a menu

Step into an actual game scene before reading the number. Do not read it from the main menu, a pause screen, or a loading screen. After Alt-Tabbing back into the game, wait a few seconds for the game to resume rendering before reading. This is the same requirement the laptop card-selection guide uses: the reading only means something when the game is doing the work being measured.

Step two: look at the right card — GPU 0 and GPU 1 are not "integrated" and "dedicated"

On the Performance page, the left side lists GPU 0, GPU 1, and sometimes GPU 2. The number itself does not say which one is the integrated graphics and which one is the dedicated card: the numbering follows hardware position, and it can differ between machines. The way to tell is to click into each one and read the model name at the top right — Intel UHD or Iris Xe means integrated graphics, NVIDIA GeForce or AMD Radeon means the dedicated card.

If the card being watched is the integrated graphics while the game is running on the dedicated card, a 0% reading on the integrated card is normal, not a fault. The 3D engine of the dedicated card is the one to watch.

If the dedicated card stays near 0% while the integrated graphics carries load, and the game is also running slowly, that is not this guide's problem — it is the card-selection problem, and the diagnosis follows the laptop card-selection guide. On a desktop, check where the monitor cable is plugged in: if it is connected to the motherboard output, the picture is running on the integrated graphics.

Step three: look at the right engine — the chart's default is not the whole story

On the Performance page, click the engine name and look through the engines one by one: 3D, Copy, Video Decode, Video Encode, Compute. A game normally runs on 3D, but a video, a compute job, or a recording does not.

On the Processes page, add the GPU engine column — right-click the column headings and tick the GPU and GPU engine columns — and look at what the game program itself is using, for example "GPU 1 – 3D". That column is the direct evidence for which program is using which card and which engine. If the game program shows 3D activity in that column while the overall chart shows 0%, the chart is showing the wrong engine, not a game that is failing to use the GPU.

Microsoft's own example belongs here, because it is the strongest evidence in this guide that 0% can be normal: while a video plays, the 3D reading can be 0% because the work is on the Video Decode engine.

Step four: the frame-cap comparison

Switch off VSync, switch off the in-game frame-rate limit, and switch off any driver-level frame cap, one at a time, checking the reading after each change. If the reading rises when a cap is switched off, the earlier 0% was waiting time. Note the frame rate at the same time: if the frame rate rises and the reading rises together, the two are consistent. If the frame rate does not change but the reading rises, the reading itself may be unreliable, and the investigation should go back to step three.

This step is a comparison only. It is not a recommendation to run without VSync or without a frame cap permanently. Once the comparison is done, restore the original settings.

Step five: cross-check with another tool, plus the physical signs

Open a different monitoring tool — GPU-Z, HWiNFO, or the card maker's own monitoring software — and look at the same period. If the other tool shows normal load, the Task Manager reading is a reporting problem, not a GPU problem.

At the same time, look at the physical signs: temperature, fan speed, and power draw. If the number says 0% but the temperature and fan speed rise with the game load, the reading and the reality have come apart, which points toward a reporting error. If the temperature, fan speed, and power draw are all calm, and the game is running smoothly, the game itself is lightly loaded — go back to step one. Do not jump between tools comparing absolute numbers: fix one tool for the before-and-after comparison.

Step six: the driver and WDDM layer

Task Manager needs a driver supporting WDDM 2.0 or a newer version before it will show GPU data at all. A driver that is too old or damaged can leave Task Manager with no GPU data, or with abnormal readings.

The Device Manager comparison is the screen for this step: if the card appears as "Microsoft Basic Display Adapter", the system is running on the generic display driver, and the monitoring layer is not normal either. That situation follows the driver-layer guide.

Reinstalling the driver is the last resort, not the first step. Use the driver program's own clean-install option first, rather than starting with a thorough removal utility. Before reinstalling, note the existing driver version and the baseline readings; after reinstalling, use the same way of observing to compare.

Boundary Cases: What This Guide Is Not About

A laptop that runs games slowly with the dedicated GPU near 0% is the card-selection guide's subject. Its symptoms are slowness; its process covers confirming which card is running, graphics settings assignment, the control-panel performance setting, and the MUX switch. Two points from that guide also apply here — matching GPU numbers to model names, and reading the number only inside an actual game scene.

High utilisation with nothing open is the reverse symptom, covered by the idle-load guide. The engine table, the sampling principle, and the driver-reporting point are covered in detail in that guide.

Utilisation and clock speed falling out of step — for example, clock speed and fan speed not dropping back to idle after a game, or not rising when entering a game — is the subject of the utilisation-versus-clock guide. The point that the two are separate readings is covered in detail in that guide.

A frame rate that cannot rise, and whether the CPU or the GPU is the limit, is the bottleneck guide's subject. It deals with the reverse situation — high utilisation but low frame rate — and this guide does not repeat its reasoning.

Using Task Manager as a flicker test — whether Task Manager itself flickers, to separate a driver problem from an application problem — is a different use of the same tool, covered by the flicker guide. This guide uses Task Manager to read GPU engine numbers, which is a different purpose.

A game that fails to initialise at all, because it cannot create a Direct3D device, is the opposite entry condition from this guide: the first frame is never drawn, while here the picture is running normally. The two should never be mixed together.

Common Mistakes

MistakeCorrection
A GPU reading of 0% means the card is faulty or doing nothingWrong. 0% has several meanings: the wrong card, the wrong engine, a game that is not busy, a frame cap making the GPU wait, or a reporting error
GPU 0 is always the integrated graphics and GPU 1 is always the dedicated cardWrong. The numbering follows hardware position and can differ between machines; click into each one and read the model name
Task Manager's overall percentage is the card's total loadWrong. The overall percentage is the busiest engine, under Microsoft's official mechanism — not an average, and not all the engines added together
Two monitoring tools show different numbers, so one of them must be brokenWrong. Sampling windows and calculation methods differ; fix one tool and use it for before-and-after comparison
A slow game with a dedicated GPU at 0% is the same problem as this guideWrong. That is the card-selection problem: the symptom is slowness. This guide is about a game that runs smoothly while the reading is 0%
Switching off VSync made the reading rise, so the problem is fixedWrong. That was only a comparison, proving the earlier 0% was waiting time; restore the original setting afterwards
Reinstalling the driver is the first stepWrong. Work through the six steps first; the driver is the last resort, and reinstalling without separating the layers wipes out the comparison evidence
High utilisation means the card is working hard, and low utilisation means it is idleWrong. Utilisation and clock speed are two separate readings and have to be read apart

When to Seek a Professional Check

If the six steps have been worked through and the reading is still 0%, and the following can all be stated clearly, then a professional check is the reasonable next step: what the frame rate is inside an actual game scene, and whether the picture is smooth; which card — by model name — and which engine shows activity in Task Manager, and what the game program's GPU engine column says; whether the frame cap is on or off, and whether the reading changed with it; which other tool was used for the cross-check, and what it showed; whether temperature, fan speed, and power draw moved with the game load; and which driver version is installed and which WDDM version it reports.

HK GPU Fix is a graphics card repair specialist in Shek Mun, Hong Kong, at Room 1209, Dali Plaza. The published three-tier inspection and repair pricing range is HK$400–2,800, with the actual quote confirmed after inspection. For enquiries, call 2186 8010 or WhatsApp 6558 6806. There is no Mong Kok shop and no on-site service; inspection is carried out at the Shek Mun workshop. Bringing the record described above gives the inspection the most useful evidence to work from.

FAQ

The game runs fine, but Task Manager shows 0% GPU utilisation. Is the card faulty?

Usually not. A game that runs smoothly while the reading is 0% points toward a reading problem: the wrong card, the wrong engine, a frame cap, or the wrong moment. A faulty card normally shows other symptoms — crashes, artefacts, or no picture — not a smooth game with a single number reading zero.

Why does the overall percentage not match the engine charts?

Because the overall percentage is the busiest engine, not an average. On a card with 10 engines, a game that fully loads the 3D engine would show only 10% if the figure were an average, so Microsoft's design uses the busiest engine instead. The overall figure and any single engine chart are measuring different things.

I am watching the 3D engine while a video plays, and it reads 0%. Is that normal?

Yes. A video is decoded on the Video Decode engine, not the 3D engine. Microsoft's official blog uses this exact example: while a video plays, the 3D reading can be 0% because the work is happening on a different engine. Switch the chart to Video Decode to see the activity.

GPU 0 shows 0% but GPU 1 is busy. Which one is my dedicated card?

The number does not say. Click into each GPU on the Performance page and read the model name: Intel UHD or Iris Xe means integrated graphics, NVIDIA GeForce or AMD Radeon means the dedicated card. The numbering follows hardware position and can differ between machines.

Two monitoring tools show different GPU numbers. Which one is right?

Neither is automatically wrong. Task Manager's overall figure is the busiest engine, while `nvidia-smi` measures the share of time a kernel was executing over its sampling period. Different definitions and different sampling windows give different numbers. Fix one tool and use it for before-and-after comparison.

I switched off VSync and the reading rose. Does that mean the problem is fixed?

No. That was a comparison, not a repair. It shows the earlier 0% was waiting time caused by the frame cap. Restore the original setting afterwards, and treat the result as evidence about the reading, not as a fix for the card.

Should I reinstall the driver to fix a 0% reading?

Not as a first step. Work through the six steps — the right scene, the right card, the right engine, the frame-cap comparison, and a cross-check with another tool — first. The driver is the last resort, and the driver program's own clean-install option comes before any thorough removal utility.

The reading is 0% in the game menu but normal during play. Is that a fault?

No. A game on its menu or pause screen is not drawing a complex scene every frame, so the GPU load is genuinely low. Read the number only inside an actual game scene, after waiting a few seconds for rendering to resume.

Related articles

Related services

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