We Timed 40 Video Sites: What Separates a 0.8-Second Start from an 8-Second One

Play-button-to-first-frame measurements across three ISPs and two devices. The gap between fast and slow platforms isn't bandwidth — it's decisions made before the video loads.

We pressed play on 40 video platforms, on the same devices, on the same connections, at the same times of day — and measured the gap between click and first frame. The spread was 0.8 to 8.4 seconds. Bandwidth explained almost none of it. Here’s what did.

The Setup

Two test devices (a mid-range Android phone and a five-year-old laptop), three Thai ISPs (AIS, True, 3BB), three measurement windows per day for a week. We defined start-up as: click → player initialization → first decoded frame rendered. Anything the user perceives as “waiting” counts.

The Results

Tier          │ Sites │ Median start │ Common traits
──────────────┼───────┼──────────────┼─────────────────────────────
Instant       │   7   │ 0.8–1.5 s    │ Tiny player, cached manifest
Good          │  14   │ 1.5–3.0 s    │ Moderate JS, warm CDN edge
Sluggish      │  12   │ 3.0–5.5 s    │ Ad scripts before player init
Painful       │   7   │ 5.5–8.4 s    │ Redirect chains, cold origin

The fastest tier shares a recognizable architecture: a small player bundle (under ~200 KB of JS before video logic), manifests served from a CDN edge with long TTLs, and — critically — ads and trackers deferred until after playback begins.

The Three Killers, Ranked

1. Third-party scripts before init. On the slowest tier, the median site loaded 9 distinct third-party domains before the player even started fetching a manifest. Each is a DNS lookup, a TLS handshake, a parse-and-execute — sequentially, on the critical path. One site loaded three separate analytics SDKs and two ad networks before the <video> tag existed in the DOM.

2. Redirect chains. Click → HTTP 302 → another 302 → a landing page → then the player. Every redirect is a full round trip. Three hops on a trans-Pacific route is a second and a half of pure geography.

3. Cold manifests. The .m3u8/.mpd is fetched on every play on some platforms — it’s tiny, but if it comes from an origin server rather than an edge cache, that’s 200+ ms before any segment request can begin.

What the Fastest Sites Do Differently

  • Player bootstrap inline or first-priority, ads strictly deferred
  • Manifest edge-cached with stale-while-revalidate
  • Preconnect hints (<link rel="preconnect">) to the segment CDN
  • Poster image is the first paint — perceived speed matters even before decode

The full dataset, per-ISP breakdowns, and the methodology JSON are in the Sextube Metrics benchmark archive.

The Takeaway

Fast start-up isn’t bought with bandwidth — it’s chosen with architecture. The 0.8-second sites didn’t have better servers; they had fewer things between the play button and the picture.