How to Actually Measure Input Lag and Stuttering on Your PC

Most optimisation content has the same flaw: it tells you to apply twenty tweaks and never shows how to know whether any of them worked. In the end, the only proof is "I think it got better" — and expectation is a terrible measuring instrument.

This guide is the opposite: it shows what to measure, what each number means and from which value there is a real problem.

Why average FPS is useless here

Average FPS is the most publicised metric and the least useful for diagnosing stutter. A PC can show 200 average frames and still have isolated frames that took 50 milliseconds — and those are exactly what you feel.

Think of it this way: if ten frames take 5 ms and one takes 60 ms, the average still looks great. But that single frame is the hitch that made you miss the shot. What matters is consistency, not the average.

The metrics that actually matter

1. Micro-stutter (scheduling jitter)

Measures the "gaps" where the system held the processor and nothing was processed. It is the root of micro-stutter.

How to read it: peaks below 0.6 ms indicate a healthy system. Between 0.6 and 2 ms, moderate oscillation. Above 2 ms, you feel hitches in-game.

2. System timer resolution

The granularity of the clock Windows uses to schedule tasks.

How to read it: 15.6 ms is the Windows default and is coarse for games. 1 ms or less is what you expect on a prepared machine. Worth knowing: on recent Windows versions this resolution is per-process, and the value can vary with which programs are open.

3. Mouse polling rate

How many times per second the mouse reports its position.

How to read it: 125 Hz is factory default and can be improved; 500 Hz is good; 1000 Hz is the competitive standard. Careful with the measurement: if the test uses the browser's ordinary movement event, it saturates around 60 per second and shows a false number. Correct measurement uses raw device events.

4. Frequency stability under load

Here the question is not "how fast", but "does it hold". A continuous load is applied and you watch whether throughput stays constant or oscillates.

How to read it: above 92% stability is excellent. Between 75 and 92%, partial sustain. Below 75%, something in power plan or thermals is limiting.

5. Network jitter (not just ping)

Average ping matters less than its variation. A steady 40 ms ping is better than a 20 ms ping that jumps to 90.

How to read it: variation below 6 ms is a stable route. Above 15 ms, hit registration becomes inconsistent — that is the "died behind the wall" feeling.

How to measure without fooling yourself

Three rules separate measurement from guesswork:

1. Same conditions in both measurements. If you measured before with the browser open and after with everything closed, the comparison is worthless.
2. A single measurement says nothing. Every PC has natural variation. Repeat and look at the pattern, not the isolated number. A serious tool uses robust statistics — median, percentile — instead of a single instant value.
3. A small gain may be noise. A 1 or 2% difference between measurements is probably normal variation, not improvement. Be suspicious of anyone presenting that as a result.

How UPBOOST does it

We built our own tool, UPCHECK, that runs on the client's PC and measures exactly these five points — before any change and again after the optimisation. It uses the median of several samples precisely so two consecutive runs give the same result, and shows the side-by-side comparison at the end.

That changes the conversation: instead of "trust me, it improved", the client sees the number before, the number after, and a report they can keep. When an item does not improve, it shows as not improved — because a tool that only knows how to deliver good news is useless for diagnosis.