← All games

The latency of a thought

Time as the axis: how many clock ticks of silicon fit inside one step of a human

One axis instead of two columns

People and machines are usually compared by who won. This page uses a different axis: duration. Everything either side does has a length in time, and lengths can be laid against each other. While an impulse travels from your retina to your finger, a processor counts off hundreds of millions of clock ticks. That is not a figure of speech; it is one number divided by another, and both were measured rather than felt.

The page does three things. It lets you measure your own latency in five flashes. It reads that result out in the clock ticks of processors from 1971 onwards. And it runs the division backwards, putting human intervals into machine units and machine intervals into human ones. Until you measure anything, the laboratory figure of 195 ms stands in for you, and the page says so where it uses it.

Five flashes

Wait for green, then press. Five times. What counts is the median, not the best attempt: the best attempt is usually the one where you guessed the moment right, while a median survives one piece of luck and one lapse. A press before green is a false start — it is discarded rather than scored, and the count of them is printed beside the result so that discarding cannot flatter you quietly. Nothing is saved: the number lives in this tab and is sent nowhere.

Click to start
or press space
—
—
—
—
—
Nothing measured yet. Every figure below is standing on the laboratory 195 ms.

What is in that number besides you

Your browser and your display add tens of milliseconds that are not in the laboratory 190–200 ms. A 60 Hz screen shows the green on average 8 ms after the page decided on it; a USB keyboard is polled every 8 ms; the operating system is busy with something of its own throughout. Measured end-to-end lag on online experiment platforms runs from 30 to 100 ms. So if you got 250 ms, that does not mean you are 55 ms slower than a laboratory participant. It most likely means you are about the same and the difference is living in the wiring.

A. Anwyl-Irvine et al., Behavior Research Methods 53, 2021; R. Woodworth, H. Schlosberg, Experimental Psychology, 1954. Your median is ours in the weakest sense — one person, one room, one machine — and it does not stand beside either.

While you were pressing

Your latency divided by one clock period. The clock speeds come from a table of 124 processors; 123 of them record a speed and all 123 parsed cleanly. Where a part shipped across a range of speeds, the figure used is the top of that range — the most generous reading available to the machine.

In the laboratory 195 ms an Intel 4004 counts 144,300 clock ticks and an Apple M3 Max 789,750,000. Measure yourself above and these become your own.

Processor Year Clock Ticks in 195 ms
Intel 40041971740 kHz144,300
Intel 808619785–10 MHz1,950,000
Intel 80386198516–33 MHz6,435,000
Intel Pentium199360–300 MHz58,500,000
Intel Pentium III1999450 MHz – 1.13 GHz220,350,000
Intel Core 2 Duo20061.8–3.3 GHz643,500,000
Intel Core i7-2600K20113.4–3.8 GHz741,000,000
Apple M120203.2 GHz624,000,000
Apple M3 Max20234.05 GHz789,750,000
Apple M3 Max, 2023, 4.05 GHz: 789,750,000 clock ticks in 195 ms.
Clock speeds: the processor table of the cpu_history project, 124 entries from the Intel 4004 of 1971 to the Snapdragon X Elite of 2024. One entry, Google's TPU v4, records no clock at all and is not on this page. The duration is either your own median, marked as such, or the laboratory 190–200 ms of R. Woodworth and H. Schlosberg, Experimental Psychology, 1954.

The other direction

The same division, run backwards. On the left are the elementary steps from The Limits of a Human, with the same values and the same sources. Then how many ticks of the selected processor fit inside each, and what each becomes if one tick is stretched to one second — which is the only way the ratio becomes something a person can picture.

Step Duration Ticks of Apple M3 Max If one tick were a second
One addition in a CPU~1 ns44 s
A fetch from RAM~100 ns4056.8 min
One synaptic delay~150 µs607,5007 days
Retina → visual cortex~63 ms255,150,0008 years
The Stroop conflict cost~100 ms405,000,00013 years
Reaction to a flash (laboratory)~195 ms789,750,00025 years
One sum in the head (ours)~4 s16,200,000,000513 years
Synaptic delay: B. Sabatini, W. Regehr, Nature 384(6605), 1996 — at body temperature the postsynaptic response begins 150 µs after the presynaptic spike. Retina to primary visual cortex: M. Schmolesky et al., Journal of Neurophysiology 79(6), 1998. The Stroop cost is the modern single-trial 70–130 ms of C. MacLeod, Psychological Bulletin 109(2), 1991. Reaction is the laboratory 190–200 ms of R. Woodworth, H. Schlosberg, Experimental Psychology, 1954. A sum in the head is ours and does not stand on the same footing as the rest. Processor figures: 1 ns is a round ceiling for an integer addition, and a fetch from main memory is the standard 60–100 ns.

A quarter of a second, taken apart

Where the time between the flash and the press goes. The first four rows are you; the last two are the equipment standing between you and the timer. They cannot honestly be added down the column: some of it overlaps, and the deciding row was not measured at all — it is the remainder of the laboratory 195 ms after everything nameable had been named.

Retina → primary visual cortex63 ms
M. Schmolesky et al., Journal of Neurophysiology 79(6), 1998: V1 units answer a flash at 66 ± 10.7 ms
Deciding — whatever is left74 ms
not measured: the remainder of the 195 ms after the three named parts. Our arithmetic, not anyone's observation
Conduction from cortex to the hand18 ms
our arithmetic: about a metre of path at the 50–60 m/s conduction velocity of large motor fibres, the standard clinical figure
Muscle: from command to force40 ms
electromechanical delay of 30–50 ms: P. Cavanagh, P. Komi, European Journal of Applied Physiology 42(3), 1979
One frame of a 60 Hz display8 ms
arithmetic: a frame lasts 16.7 ms and the page's decision waits for the next one — half a frame on average
Polling the keyboard8 ms
USB HID polls a device every 8 ms at the standard 125 Hz; gaming keyboards get to 1 ms

The two equipment rows are why the median measured on this site lands near 250 ms rather than 195. They are not part of you, they are not part of the published figure, and they are the reason the two numbers differ.

Three caveats, without which this page is a lie

A clock tick is not an operation, and nowhere near a thought. A wide modern core issues several instructions per tick, so it does more work than the tick count suggests; a fetch from main memory costs around 100 ns, three or four hundred ticks of waiting, so it does far less. On the Intel 4004 one machine cycle took eight clock periods — 10.8 µs — and the honest 1971 number is the one in the table divided by eight. Ticks are the beats of a metronome, not the notes anybody played.

Reaction is not thinking. It measures the journey of a signal and says nothing about what was decided along the way — a fast press and a good decision are separate things, and this page compares only the first. That is why Reaction is left out of the score of humanity elsewhere on this site, and why this page keeps no score of its own, reports nothing to the shared counter and has no key in it.

Every number here carries a name. The laboratory values come from published work; the clock speeds come from the processor table; and whatever was measured by us — your median, and the sum in the head carried over from The Limits of a Human — is marked as ours and does not stand beside a publication. Anything obtained by subtracting or dividing is called arithmetic rather than a measurement, in the row where it appears.

And the distance is not a verdict. Eight orders of magnitude between a clock tick and a press describe how fast a signal moves, not what is done with it. A billion ticks do nothing on their own; somebody has to give them something to do, and what is easy to give them is not what a quarter of a second of a human is spent on.

Where to go from here
To cite this page: “The latency of a thought”, You vs Computer, https://youvscomputer.com/games/latency.html