Frame pacing

On this page 7

Frame pacing is the consistency of the intervals at which a game presents finished frames to the display, measured as the variation in frame time from one frame to the next. In gaming performance, it explains why two systems with the same average frame rate can feel very different: a game running at 60 fps with perfectly even 16.7 ms intervals looks smooth, while one that alternates between 8 ms and 25 ms frames at the same average looks jerky and uneven.

Good frame pacing means each frame stays on screen for roughly the same duration and represents a matching step of simulated game time. Poor frame pacing produces judder, microstutter and visible hitches even when a frame counter reports a healthy number, which is why reviewers and engine developers now treat it as a separate quality metric alongside raw frame rate.

How frame pacing works in the rendering pipeline

Frame pacing results from how the CPU, GPU and display hand work to one another. The CPU runs game logic and records draw calls, the GPU renders the frame, and the swap chain presents it at a refresh boundary or, with VRR, at the moment it is ready. Any stage that finishes irregularly passes its irregularity downstream.

Two timelines must stay aligned. The simulation timeline decides which moment of game time a frame depicts, and the presentation timeline decides when the player actually sees it. When a frame showing 16 ms of motion is displayed for 33 ms, or the reverse, objects appear to speed up and slow down even if every frame is rendered correctly. Engines therefore try to sample game time at a predicted display moment rather than at the moment work begins.

The following table lists the main pipeline stages and how each one can disturb pacing.

StageRoleTypical pacing disturbance
Game thread (CPU)Simulation, input, scriptingGarbage collection, streaming, AI spikes
Render thread (CPU)Draw call submissionShader compilation, driver overhead
GPURasterization, shading, post-processingLoad spikes from effects or scene changes
Render queueBuffers frames between CPU and GPUUneven queue depth, added latency
DisplayScans out frames at refresh intervalsMissed vsync deadlines, repeated frames
Pipeline stages that influence frame pacing

Measuring frame pacing with frame time data

Frame pacing is measured with per-frame timing logs, not with an fps average. Capture tools record the time between consecutive presents, which yields a frame time graph; a flat line indicates good pacing and a sawtooth or spiky line indicates poor pacing. Percentile metrics such as 1% lows and 0.1% lows summarize the worst frames in a run.

Different metrics answer different questions, so analysts combine several of them, as the table below shows.

MetricWhat it capturesLimitation
Average fpsOverall throughputHides uneven intervals entirely
1% and 0.1% lowsSeverity of the slowest framesIgnores alternating patterns at high fps
Frame time varianceSpread of intervals around the meanSensitive to run length and scene
Frame-to-frame deltaSize of consecutive changesRequires raw per-frame data
Displayed frame timeWhat the screen actually showedNeeds display-side or hardware capture
Common frame pacing metrics

The distinction between presented and displayed frames matters. Software tools see when the game called present, but with vsync or a compositor the screen may show the frame later. Hardware capture with a camera or an analysis card reveals dropped and runt frames that software logs miss, which was historically important for multi-GPU configurations.

Frame pacing overlaps with several neighboring terms but describes a specific property: regularity over time. Frame time is the raw measurement, stutter is a visible symptom, and input lag is a separate latency dimension that can worsen even while pacing improves.

ConceptDefinition in briefRelation to frame pacing
Frame timeDuration of one frame in millisecondsThe data from which pacing is judged
StutterVisible hitch or pause in motionA symptom of a severe pacing failure
Input lagDelay from input to visible responseDeeper queues can smooth pacing yet add lag
VRRDisplay refresh that follows the GPUMasks small variations, cannot fix large spikes
Frame rate capUpper limit on rendered fpsOften the simplest way to even out intervals
How frame pacing relates to neighboring terms

A frequent misconception holds that variable refresh rate solves pacing. G-SYNC and FreeSync remove tearing and the quantization of vsync, so a 50 fps frame is shown for 20 ms instead of alternating between 16.7 ms and 33.3 ms. However, if the game itself produces a 10 ms frame followed by a 40 ms frame, the display faithfully shows that unevenness in the source.

Causes of poor frame pacing and how to fix them

Poor frame pacing has identifiable causes, and each one leaves a characteristic pattern in the frame time graph. Diagnosing the pattern first prevents random changes to settings that address the wrong bottleneck.

SymptomLikely causeTypical remedy
Single large spikesShader compilation, asset streamingPrecompiled shader cache, faster storage
Regular alternating intervalsFrame rate just below refresh with vsyncCap at a divisor of refresh or enable VRR
Gradually rising frame timesMemory leak or thermal throttlingDriver or game patch, better cooling
Constant noisy jitterCPU bottleneck, background processesLower CPU-heavy settings, close tasks
Judder in cutscenes or 30 fps modesMismatched simulation and display timingEngine-side pacing fix, correct cap
Pacing symptoms, likely causes and remedies

The most effective user-side tool is a frame rate cap set slightly below the sustainable minimum. When the GPU and CPU always finish early, every frame waits for the same deadline and intervals become nearly identical. On VRR displays, capping a few frames below the maximum refresh keeps the game inside the variable refresh window and avoids the latency penalty of vsync.

  • In-engine limiters usually pace better than external ones because they control simulation timing directly.
  • Latency-reduction modes such as NVIDIA Reflex shorten the render queue and can tighten pacing in GPU-bound scenes.
  • Dynamic resolution scaling trades image sharpness for steadier GPU frame times under fluctuating load.

Frame pacing on consoles and fixed refresh displays

Console games commonly target 30, 40 or 60 fps on displays that refresh at 60 or 120 Hz, so correct pacing requires each frame to occupy an equal number of refresh cycles. A 30 fps game on a 60 Hz screen must show every frame for exactly two refreshes; a pattern of one, three, two, two refreshes keeps the 30 fps average but produces obvious judder.

The 40 fps mode on 120 Hz displays works because 120 divides evenly by 40, giving each frame three refreshes and a 25 ms interval. Platform holders provide pacing APIs and certification guidelines, but pacing bugs still ship, and they are often fixed in patches once players capture frame time data that exposes the irregular cadence.

Explained without jargon

Imagine a flip book. If every page is turned at exactly the same speed, the drawing moves smoothly. If some pages flip quickly and others linger, the motion looks shaky, even when the total number of pages per second stays the same. Frame pacing is that rhythm in a video game.

A game counter might proudly report 60 frames per second, but that number is only an average. What your eyes notice is whether the pictures arrive at a steady beat. A slightly lower but perfectly even frame rate often feels smoother than a higher, uneven one.

A typical use case

A player with a 144 Hz monitor without variable refresh notices that an open-world game feels rough despite an on-screen counter showing about 70 to 80 fps. A frame time capture reveals intervals jumping between 6.9 ms and 13.9 ms as vsync rounds each frame to whole refresh cycles, plus occasional spikes of 60 ms when new areas stream in.

The player sets an in-game limit of 72 fps, exactly half the refresh rate, so every frame occupies two refresh cycles and a constant 13.9 ms interval. After installing the game on an NVMe drive and letting the shader cache build during a first session, the large spikes become rare. The average frame rate is slightly lower than before, yet the frame time graph is almost flat, and traversal now looks consistently smooth.