By Walter Jobs, Technology writer and editor · Published 9 October 2026
Short answer: a graphics card below about 95% during a demanding game is waiting for something. Nine things can make it wait, and only three of them are a processor limit. The other six are settings, caps and configuration faults that cost you nothing to fix. Work through them in the order below, because that is roughly the order of how often each one turns out to be the culprit.
One piece of framing first, because it saves a lot of wasted effort. A card at 99% is the healthy state, not the worrying one. It means you are getting everything you paid for. People arrive at this problem convinced that high usage is bad and low usage is efficient, which is exactly backwards.
Before you diagnose anything, read the number correctly
Two measurement mistakes produce a low reading on a perfectly healthy machine.
The first is averaging. Task Manager samples slowly and smooths, so a card swinging between 100% and 40% in a busy scene displays as something unremarkable in the middle. Use an in game overlay that samples continuously instead, and watch it during actual play rather than in a menu, a loading screen or a paused game, all of which legitimately idle the card.
The second is reading the wrong chip. On a laptop, Windows lists the integrated graphics inside the processor and the separate card as two devices. If you are watching the one that is not running the game, you will see a low number forever. Expand the graphics section of your overlay or Task Manager and confirm which device is doing the work.
The nine causes, in order of likelihood
1. A frame rate cap you forgot about
This is the most common answer by a distance, and the easiest to miss because caps hide in four separate places: the game’s own limiter, V-Sync, the driver control panel, and a platform wide overlay or power feature. Any one of them will hold the card at exactly the level needed to hit the cap and no higher.
How to confirm: look at the frame rate, not the usage. If it is pinned at 60, 120, 144 or exactly your monitor’s refresh rate and barely moves regardless of what is happening on screen, it is capped. A real bottleneck produces a wandering number, not a flat line.
If this is it: nothing is wrong. Your card is idling because you asked it to.
2. The game is limited by its own engine
Plenty of games have an internal ceiling unrelated to your hardware, whether a hard limit written into the engine, a physics loop tied to frame rate, or simply a title old enough that it was never built to run faster. Competitive titles are the common case in the other direction: they are so light on graphics that no modern card is meaningfully loaded by them.
How to confirm: run a different, graphically heavier game. If the card loads to 99% there, your hardware is fine and the first game is the limit.
3. Your settings are too light for the card
At 1080p on medium, a strong modern card finishes each frame so quickly that it spends most of its time waiting for the processor to describe the next one. This is a real processor limit, but it is one you created with the settings rather than one the hardware imposed.
How to confirm: raise the resolution or the preset. Usage should climb toward 99% immediately. If it does, you have found free visual quality: you were leaving it on the table at no cost in frames.
4. The processor genuinely cannot keep up
Now we reach an actual bottleneck. The card is idle because the processor cannot describe frames fast enough to keep it busy, and raising settings does not help because the card had spare capacity all along.
How to confirm: watch per core load rather than overall processor usage. A total figure of 45% can hide one core pinned at 100% while the rest idle, and that single saturated core is the ceiling. Overall usage is the wrong number to look at here, which is why so many people conclude their processor is fine when it is not.
If this is it: put both parts into the bottleneck calculator to see how large the gap actually is and what the pair should return at your resolution. The CPU side of this symptom is covered in detail in 100% CPU usage while gaming.
5. Memory is the constraint, not either processor
A single memory stick halves the bandwidth feeding your processor, and the result looks exactly like a weak processor: the card waits, usage sits low, and no graphics setting changes anything. Running out of capacity produces a different and more obvious version, where the frame rate collapses periodically as Windows swaps to disk.
How to confirm: Task Manager, Performance, Memory. Check the slots used line for single channel, and watch whether committed memory is pressing against your installed total during play.
6. The card is thermally or power throttling
A card that starts a session at 99% and settles to 70% after ten minutes is not being starved of work, it is reducing its own clocks to stay within a temperature or power limit. The distinguishing feature is time: this one appears gradually, while every other cause on this list is there from the first second.
How to confirm: log core temperature and clock speed for fifteen minutes of play. Falling clocks alongside rising temperature is the signature. Dust, an aged thermal pad, a restricted case or an undersized power supply all produce it.
7. The card is in the wrong slot or running at reduced link width
Boards have one slot wired for full bandwidth and others wired for far less. A card in a secondary slot, or in a primary slot sharing lanes with a populated fast storage drive, runs at a fraction of the link width it expects. Modern cards with narrow interfaces are considerably more sensitive to this than older ones.
How to confirm: your driver control panel reports the current link speed and width. Compare it against what the card is rated for. A four lane link on a card built for eight or sixteen is the finding.
8. A background process is stealing processor time
Shader compilation, a cloud sync, an antivirus scan, a game launcher updating something in the background, or a browser with a heavy tab open all take processor time away from the game. The effect is indistinguishable from a weak processor while it lasts, and it stops when the task finishes, which is what makes people describe the problem as random.
How to confirm: sort Task Manager by CPU while the game runs windowed and look at what is above the game.
9. The driver or power profile is wrong
Least likely, and the one people try first. A Windows power plan set to a balanced or battery saving profile will hold clocks down on a laptop, and a graphics driver that failed to install cleanly over an old one can leave the card running on a generic display driver. Both are real, both are rare compared with the eight causes above.
How to confirm: check that Windows reports your actual card rather than a basic display adapter, and set the power mode to best performance while plugged in.
A table to narrow it down fast
| What you observe | Most likely cause |
|---|---|
| Frame rate pinned to one flat number | A cap, cause 1 |
| Fine in one game, low in another | Engine limit or light settings, causes 2 and 3 |
| Usage climbs when you raise resolution | Settings too light, cause 3 |
| One processor core pinned, others idle | Processor limit, cause 4 |
| One memory stick fitted | Single channel memory, cause 5 |
| Starts high, degrades after ten minutes | Throttling, cause 6 |
| Low everywhere, including synthetic tests | Link width or driver, causes 7 and 9 |
| Comes and goes unpredictably | Background process, cause 8 |
When low usage is the right answer
Worth saying plainly, because a lot of people spend money fixing something that was never broken. If you are hitting the frame rate your monitor can display, a card below 99% is doing precisely what it should. There is no prize for loading it fully, and a card held at 60% by a frame cap runs cooler, draws less power and lasts longer than one you have pushed for no visible gain.
The number that matters is the frame rate against what your screen can show. Usage is a diagnostic clue, not a score.
Frequently asked questions
What GPU usage should I see while gaming?
In a demanding game with no frame cap, 95% or above. That means the card is the part setting your frame rate, which is the correct arrangement on a well matched machine. Anything meaningfully below that in a heavy scene means the card is waiting, and the nine causes above are the reasons it waits.
Why is my GPU usage low but my CPU usage is also low?
Because the limit is probably a cap or a single saturated core. Overall processor usage averages across every core, so a four core limit on a sixteen thread chip reads as roughly 25% even while that one core is the ceiling. Switch your monitoring to per core and look for the one pinned near 100%.
Does low GPU usage mean I need a better processor?
Only if you have ruled out the six causes above that are not processor related. Caps, engine limits and light settings account for more of these cases than an actual processor limit does. Confirm a single saturated core first, then check the size of the gap before buying anything.
My GPU usage drops to 0% for a moment and the game freezes. Is that the same problem?
No, that is a different fault. Brief drops to zero with a visible hitch usually point at the driver recovering from a hang, a power delivery problem, or a storage stall while the game streams data. Low but steady usage is a feeding problem; usage that falls to nothing and returns is a stability problem, and it is worth checking reliability history and drive health rather than working through this list.
Will overclocking my processor fix low GPU usage?
It can help when a single saturated core is genuinely the ceiling, because that is the one scenario where a few hundred megahertz translates directly into frames. It does nothing for any of the other eight causes. Establish which one you have before changing voltages, since the gain is modest even in the best case.
The diagnostic steps here are things you can verify on your own machine. Where this article refers to the size of a performance gap between two parts, that comes from our own model rather than a measurement of your system, and how we calculate this explains the method.