Essential Latency Numbers Every Web Developer Must Know

In modern web development, understanding and optimizing latency is crucial for delivering an excellent user experience. Every millisecond can affect satisfaction, whether it’s page load, resource response, or interaction feedback.

Below are three well‑known latency benchmarks from abroad:

Peter Norvig – https://norvig.com/21-days.html#answers

Jeff Dean – https://www.cs.cornell.edu/projects/ladis2009/talks/dean-keynote-ladis2009.pdf

University of California, Berkeley – https://colin-scott.github.io/personal_website/research/interactive_latency.html

Typical latency figures (in nanoseconds unless otherwise noted):

Execute a typical instruction: 1 ns

Load from L1 cache: 0.5 ns

Branch misprediction penalty: 5 ns

Load from L2 cache: 7 ns

Mutex lock/unlock: 25 ns

Load from main memory: 100 ns

Send 2 KB over a 1 Gbps network: 20 000 ns

Sequentially read 1 MB from memory: 250 000 ns

Seek to a new disk location: 8 000 000 ns

Continuous read of 1 MB from disk: 20 000 000 ns

Round‑trip packet from the US to Europe and back: 150 ms (150 000 000 ns)

Note: 1 ns = 10⁻⁹ s, 1 ms = 10⁻³ s = 1 000 µs = 1 000 000 ns.

Interestingly, when network speed is sufficiently high, reading data can be faster than accessing memory. The performance gap between memory, SSD, and HDD is roughly an order of magnitude.

Vercel‑Provided Latency Numbers (2023 Android measurement)

1. Network Impact

Wi‑Fi latency: 1–4 ms (affects TTFB, FCP, LCP)

5G high‑frequency: 1–5 ms (affects TTFB, FCP, LCP)

5G mid‑frequency (poor signal or tower overload): 10–30 ms

4G LTE: 15–50 ms

3G: ~150 ms

Inter‑continental round‑trip (single region deployment): ~150 ms

Intra‑continental city‑to‑city (≈5 000 km): ~33 ms

2. Data Processing

Service or database within the same cloud region: ~10 ms

Parse 1 MB CSS: ~100 ms (affects FCP, LCP)

Parse 1 MB HTML: ~120 ms (affects FCP, LCP)

Parse 1 MB JavaScript: ~150 ms (affects INP, FCP, LCP)

3. User Experience

Perceived instantaneous response threshold: 40–80 ms (affects INP)

Perceived smooth frame rate (60 fps): 16 ms per frame; latency 5–10 ms for rendering

Perceived slow response threshold: 200 ms (INP “needs improvement”)

Original source: https://vercel.com/blog/latency-numbers-every-web-developer-should-know