The millisecond figure next to each node is most people's only selection criterion. It is useful — but it does not measure what many assume it does. Understanding it makes the difference between picking a good node and picking a lucky one.

Clash Verge node list with green, amber and red latency results
The lightning icon tests the whole group at once

What the number measures

Hitting the lightning icon sends an HTTP request through each node to a test URL (Google's generate_204 by default) and times the full round trip. So the figure reflects the real path — you to the node, node onwards to the wider internet — which is far closer to actual experience than a plain ping to the server.

Reference ranges

  • Under 200 ms: excellent — browsing, video and calls feel smooth;
  • 200–400 ms: usable; pages feel slightly delayed, video is fine;
  • Over 400 ms: workable but interactive apps stutter;
  • Timeout: the node is down or temporarily blocked.

Low latency does not imply high bandwidth. A 50 ms node might cap at 10 Mbps while a 150 ms premium line saturates your connection. For 4K streaming, the only real test is playing the stream.

Letting the core choose: url-test groups

Most subscriptions include an "Auto" group (type url-test) that re-tests periodically and follows the fastest node. Leaving your selection on Auto is a fine default for browsing.

Two cases where manual selection beats auto:

  • Streaming: auto-switching can land on a node that a streaming service refuses; pin a node you know unlocks it;
  • Gaming: every automatic switch drops UDP sessions — in-game that is a disconnect. Pin a node; more in gaming optimization.

Everything times out?

  1. Confirm the underlying network works (turn the proxy off, load a local site);
  2. Update the subscription — providers rotate node addresses; see subscription management;
  3. Change the test URL to http://cp.cloudflare.com/generate_204 in Settings and re-test, to rule out the test target itself being blocked;
  4. Still all red: the subscription has likely expired or the protocol is being blocked locally — ask your provider.

A node that is fast, then slow, then fast

Evening congestion (roughly 20:00–23:00 local time) on shared international lines is normal and universal. If a node is consistently slow, check how many connections it is carrying on the Connections page before blaming it. Switching your main group to fallback type — primary node with automatic backup — is a sturdy compromise; group configuration lives in the custom rules guide.