Configured right, Clash Verge playing on foreign servers feels close to a commercial game booster. Configured wrong, it is 460 ms and rubber-banding. The difference lives in three places: TUN, UDP, and the node.

Game traffic through TUN and a UDP-capable node, comparing latency and jitter of two nodes
Three links in the chain: TUN captures, the node passes UDP, the line stays stable

First: TUN is non-negotiable

Game clients essentially never respect the system proxy. Without TUN mode, your game connects directly and the proxy might as well be off. TUN captures at the adapter level; the game has no say. This is the entry ticket.

Second: verify UDP end to end

Real-time game data — position sync, voice — is almost all UDP. Any link in the chain that drops UDP degrades you to lag or disconnects:

  1. Protocol choice: Hysteria2 and TUIC are UDP-native by design; SS/VMess/Trojan need the server to enable UDP relay (nodes marked udp: true in the config).
  2. How to check: with the game running, open Clash Verge's Connections page and filter by the game's process. You want entries typed UDP; TCP-only means UDP is not passing.
  3. If the node lacks UDP: bulk-force udp: true via Script as an experiment (useless if the server refuses), otherwise change node or provider.

Third: pick nodes by jitter, not just ping

An 80 ms node can play worse than a 120 ms one — what your hands feel is jitter (latency variance) and packet loss, not the average. Practical rules:

  • Physically closer exits first;
  • Nodes labeled IPLC/IEPL (dedicated lines) hold up during evening congestion when shared lines collapse;
  • Never game on an auto-test group: every url-test switch tears down UDP sessions — an in-game disconnect. Pin one node, or use a fallback-type group; interval tuning is covered in node selection.

Local-server games: keep them off the tunnel

Rule mode already sends domestic traffic direct, but some games' login servers use domains missing from public rule lists and get caught by the catch-all. Give local games explicit direct rules (syntax in the custom rules guide):

prepend-rules:
  - PROCESS-NAME,GenshinImpact.exe,DIRECT
  - DOMAIN-SUFFIX,hoyoverse.com,DIRECT

Pre-game checklist

  1. TUN on; game traffic visible on the Connections page;
  2. UDP entries present among those connections;
  3. Node pinned — not an auto-switching group;
  4. Downloaders excluded via direct rules (see performance tuning) so they do not fight the game for tunnel bandwidth.

All four pass and latency still disappoints? The bottleneck is the line itself — a dedicated-line node buys more than any client-side tweak.