How to read a frame time graph

An average frame rate is one number trying to summarize a minute of uneven work. It can say 80 while a few frames took so long that the game felt like 30. A frame time graph shows those individual waits. The line is not a scoreboard. It is a diary of how long each frame took, and learning to read the diary stops you from chasing the wrong fix.

What the line means

Frame time is how long one frame took, usually drawn in milliseconds. A lower point is a quicker frame. A perfectly flat line would mean every frame took the same time, which real games rarely do. A gentle thickness in the line is normal. A thin spike is a single late frame, the kind your eye reads as a hitch. A line that rises and stays high means the game got heavier and remained heavier, which is a different problem from a spike. Why Your Frame Rate Stutters Even When the Average Is High is the symptom. The graph is how you prove the symptom is a spike rather than a low average you misread.

Scales that lie

The vertical scale can make a calm graph look violent or a bad one look flat. Note the units before you react. A spike that reaches a few extra milliseconds may be invisible. A spike that reaches a large multiple of the surrounding line is the hitch you felt. Comparing two screenshots with different scales will invent a winner. Set the same scale when you compare a settings change. How to Benchmark Your Own Games Properly is the habit that keeps the run fair. The graph only helps if the two runs met the same camera path.

Spikes versus a high plateau

Spikes that hit once, the first time an effect appears, often belong to shader compilation or to a texture being loaded. They may fade on a second pass. A plateau that rises whenever you look at a city is a cost you can tune with settings or a limit of the hardware. A regular zigzag, almost rhythmic, can be a sync interaction or a cap fighting the game. How to tell a CPU limit from a GPU limit tells you which chip is busy during that plateau. A spike with both chips bored may be the drive or a background task waking up. The shape points at the family of causes. It does not name the setting by itself.

Percentages people quote

Tools sometimes report a low percentile frame time, a way of saying almost all frames were at least this quick. It is useful because it ignores one freak spike and still catches a bad stretch. It is not a second average to brag about. If the average is excellent and the low percentile is poor, you are in hitch country. If both are poor, the game is simply heavy. Fixing a heavy game means settings or hardware. Fixing hitch country means finding the repeating spike. Do not use a single percentile from a stranger's PC as a target for yours. Different overlays measure differently.

A reading you can act on

Capture a short run, look for spikes that match what you felt, and change one thing. If the spikes vanish, keep the change. If the whole line drops lower, you bought general speed, which is also a win. If the graph is identical, the change was irrelevant and you can revert it without nostalgia. Save the image with a filename that says what you changed. A folder of unlabeled graphs is a mood, not a record. The point of the line is a decision you can repeat next week.

Frequently asked questions

Is a flat line the goal? A line without tall spikes is the goal. Microscopic flatness is not something you will feel. Chase hitches you can notice in play.

Can the graph itself hurt performance? A heavy overlay can. Capture a short sample, then turn the graph off and play. Do not leave three monitoring tools drawing charts during a match you care about.

Why does the graph look worse than the game feels? The scale may be zoomed in on tiny differences. Check the millisecond marks. If you cannot feel it in a real session, you do not have to fix the picture of it.