By Walter Jobs, Technology writer and editor · Published 9 October 2026
Short answer: a bottleneck by itself lowers your frame rate smoothly; it does not stutter. What people call bottleneck stutter is usually one of four separate faults that happen to travel alongside a processor limit. Getting the distinction right matters, because a genuine bottleneck is fixed with a faster part and stutter is almost never fixed that way.
Average frame rate and frame time are different measurements
Your frame counter reports an average over the last second. It is a summary, and summaries hide exactly the thing that makes a game feel bad.
Frame time is how long each individual frame took, in milliseconds. At a steady 60 frames per second every frame takes about 16.7 ms and the line on a frame time graph is flat. Now imagine 59 frames arriving in 14 ms each and one taking 120 ms. The counter still reads about 60 and you saw a visible hitch, because your eye responds to the gap, not to the average.
This is why two machines reporting the same number can feel completely different, and why the single most useful thing you can do when diagnosing this is to switch your overlay from frames per second to a frame time graph.
| What the graph looks like | What it is | What fixes it |
|---|---|---|
| Flat line, higher than you want | A genuine bottleneck | A faster part |
| Flat line with occasional tall spikes | Stutter | Not a faster part. See below |
| Constantly jagged, small variation | Micro stutter | Frame pacing, caps, driver |
| Flat, then a long plateau, then flat | A hitch, usually streaming or compilation | Storage, shader cache, memory |
What a bottleneck actually feels like
A processor limited machine is not jerky. It is consistently slower than you want, with the frame rate wandering as scene complexity changes. Walking into a busy area takes you from 120 to 75 over a second or two and it stays there while you are in the crowd. That is a slope, not a spike. Annoying, visible on the counter, and smooth.
If what you are experiencing is a momentary freeze, a hitch every few seconds, or a half second lockup when you turn a corner, that is not the bottleneck. Something else is interrupting the pipeline.
The four faults that actually cause stutter
Shader compilation
The most common cause in modern titles by a wide margin. When the game meets a visual effect it has not prepared, it compiles the shader on the spot, and the frame waits. The signature is unmistakable: severe hitching during your first hour in a new area, which then improves and never returns in that location.
A faster processor compiles faster and shortens each hitch, which is the grain of truth behind blaming the bottleneck. It does not remove them. What removes them is letting the game finish its precompilation step before playing, and not clearing the shader cache.
Running out of video memory
When textures exceed what is on the card, the driver starts moving data across the PCIe bus to system memory mid frame. That transfer is slow enough to be visible as a spike. Running out of video memory produces stutter rather than a lower average, which is why it gets misdiagnosed as a weak card when the card is fast and simply short of memory.
The test is direct: drop texture quality by one step. If the stutter disappears while the average stays roughly the same, memory was the problem.
Storage and asset streaming
Open world games load the world as you move through it. If the drive cannot supply data as fast as you travel, the frame that needs a missing asset waits for it. The pattern is positional: it happens when you move quickly, when you fast travel, or when you cross a region boundary, and never when you stand still.
This is covered in full in can storage bottleneck gaming.
Background processes taking the core the game needs
A game’s main thread needs one core to itself. When Windows schedules something else onto that core, the frame it was preparing arrives late. An antivirus scan starting, a cloud sync, a launcher updating, or a browser tab doing something expensive all produce intermittent spikes that correlate with nothing you are doing in the game, which is what makes this one so frustrating to chase.
Where the bottleneck genuinely does contribute
Having separated them, there is one real connection worth stating fairly.
A processor running close to its ceiling has no slack. When anything interrupts it, a background task, a compilation, a sudden spike in entities, there is no spare capacity to absorb the interruption, so the frame is late rather than merely slower. The same interruption on a processor with 40% headroom is invisible.
This shows up in your 1% lows. Those are the slowest one percent of frames, and they are the single best number for how a game actually feels. A machine with a comfortable processor shows 1% lows reasonably close to its average. A machine against its processor ceiling shows them far below it, because every small interruption becomes a visible event.
| Average | 1% low | What it tells you |
|---|---|---|
| 120 fps | 95 fps | Healthy. Plenty of headroom |
| 120 fps | 70 fps | Running close to a limit, feels less smooth than the average suggests |
| 120 fps | 35 fps | Something is interrupting. Not a bottleneck, find the fault |
| 60 fps | 52 fps | Smooth, simply slower than you want. A genuine bottleneck |
So a bottleneck does not cause stutter, but it does remove the cushion that was hiding it.
What to do, in order
- Switch your overlay to frame times. Spikes and slopes need different answers and the counter cannot distinguish them.
- If it improves after an hour in one area, it was shader compilation. Nothing to buy.
- Drop texture quality one step. If the spikes vanish, you were short of video memory.
- Check whether it correlates with movement. Positional hitching points at storage.
- Watch Task Manager during a spike. Something above the game in the CPU column is your answer.
- Only then consider the hardware gap. If frame times are flat and simply too high, that is a real bottleneck, and the bottleneck calculator will tell you which part to change.
Frequently asked questions
Why does my frame rate look fine but the game feels bad?
Because the average is hiding the frames that went wrong. A single 120 ms frame among sixty normal ones leaves the counter almost unchanged and is clearly visible to you. Look at your 1% lows or a frame time graph instead; if the 1% low is far below the average, that gap is what you are feeling.
Will a better CPU stop my stuttering?
Only if the stutter comes from the processor having no headroom to absorb interruptions, which is one of several causes. If it is shader compilation, video memory, storage or a background task, a faster chip shortens the symptom at best and changes nothing at worst. Identify the cause from the frame time pattern before spending.
What are 1% lows and why do people care about them?
They are the slowest one percent of frames in a run, reported as a frame rate. They matter because smoothness is set by your worst frames rather than your typical ones. Two machines averaging 100 frames per second, one with 1% lows at 85 and one at 40, feel nothing alike, and only the second one will be described as stuttering.
Can a frame rate cap reduce stutter?
Often yes, which surprises people. Capping slightly below what your machine can sustain leaves headroom for interruptions and produces more consistent frame pacing. Trading a few frames for a flatter frame time graph usually feels better than taking every frame available and getting an uneven delivery of them.
Is micro stutter the same thing?
No. Micro stutter is small, constant variation in frame delivery rather than occasional large spikes, and it usually comes from frame pacing rather than from anything running out of capacity. It commonly responds to a frame cap, to enabling or disabling V-Sync, or to a variable refresh rate display, none of which help with the spike type.
The frame time and 1% low figures in the tables are illustrative, chosen to show the relationships rather than measured from one machine. Where this article refers to performance gaps between parts, those come from our own model, described on how we calculate this.