Clash slow speedstroubleshooting by layer: node quality, network congestion, and local settingschecked one by one

Troubleshoot slow speeds in three layers: nodes, network paths, and local settings. Check node load and protocols, peak-hour congestion, DNS and TUN settings, and misrouted rules to narrow down the cause.

First, distinguish slow page loads, slow downloads, and stalls

If a page stays blank before its images and text load quickly, check DNS lookup, connection, and time to first byte. If a download starts quickly but then stalls, check sustained throughput and connection drops. A latency reading from the node list can't distinguish between these cases. Client latency tests typically send a small request to a specified URL. The result reflects only that probe address and the request time at that moment—not the download bandwidth available from your target site.

  1. Pause cloud sync, system updates, and other downloads on the affected device. In rule mode, open the client's Connections page and check which policy group matched the target domain and which node the traffic ultimately uses. If the match goes to an automatic selection group, switch to a group where you can pin a node for the test. After testing each group, verify the actual exit node.
  2. In Chrome, open Developer Tools → Network, check Disable cache, and leave the panel open. Reload the target page, select the main document request, and check DNS Lookup, Initial connection, and Waiting for server response in the Timing tab. If the request reused an existing connection, DNS or handshake timings may not appear; don't record them as zero milliseconds.
  3. Choose a file of known size that you can download repeatedly. Use the same method to record the full download time over three runs, noting the median and number of failures. Change only one condition per run. For example, 30 MB in 10 seconds is about 3 MB/s, or 24 Mb/s; 60 seconds is about 0.5 MB/s. These decimal conversions are examples to illustrate how to record results.

Layer 1: Identify slow nodes, node load, and protocol differences

Start with an A → B → A comparison

On the same device and network, at the same time of day, test node A, then node B, then switch back to A using the same target file. Start a fresh download each time and make sure the previous connection has ended; existing connections may not switch when you change the policy group selection. If A is slow on both runs, B is clearly faster, and a second download source shows the same difference, focus first on node A or its exit path.

If A is slow for only one domain and performs about the same as B with another download source, consider the route from the exit node to that site, CDN assignment, or site-side rate limiting. Client latency and speed readings can't tell you a server's CPU load or number of concurrent connections. If the provider publishes load metrics, maintenance records, or a bandwidth cap, compare those with your test times.

Confirm the connection works before comparing protocol speeds

Check the client's Logs for the target connection at the time of your test. If the configuration failed to load or the client reports an unsupported proxy type, first check which core is running rather than just the client version. With mihomo, record the full core version shown in the client, if available; for a standalone binary, run mihomo -v. Keep connection failures separate from cases where the connection works but throughput is low.

When comparing TCP-based nodes with nodes that use UDP transport, such as Hysteria2, start with two routes from the same provider and data center. Hotel or office networks may restrict UDP, and different routes may have different ingress and egress paths. If you change both the protocol and the data center, you can only conclude which available route is faster—not attribute the difference to the protocol alone.

Layer 2: Check whether peak-hour congestion is repeatable

Choose two fixed times, such as 08:30 and 21:30, and test the same node and download target for two consecutive days. Take three readings in the morning and evening, keep the connection method unchanged, and record failed attempts. If speeds drop in the evening, run two comparisons within a few minutes: use the same device to connect directly to a normally accessible download source, then test the original target through a proxy route with a clearly different ingress path. Different node names don't necessarily mean different ingress paths.

Comparison results (examples)Next step
A drops from 4.0 to 0.6 MB/s, while B stays near 3 MB/s and the direct connection is stableCheck route A's ingress or egress path first. Retest the next day to rule out a temporary fluctuation.
Multiple proxy routes and the direct connection all slow down in the eveningTry a wired connection instead of Wi-Fi, keeping other conditions unchanged. If speeds are still low, check router upload load and your broadband connection.
Two proxy routes slow down only for the same download source; other sources are fineTest another file from the same site to check for site-side rate limiting or an issue with a specific CDN path.

A direct-connection test using a different source only shows whether your local connection can reach normal speeds for that test; it doesn't prove the target site is working normally. If the same node recovers over a mobile hotspot, the issue is more likely with your original network or its route to the node. But switching to a hotspot also changes the carrier and routing, so it can't pinpoint a specific congested hop.

Layer 3: Check rules, DNS, and TUN routing

First, check where the slow request is routed

Check whether the slow resource is the main site, an image or video, or a download subdomain. Then look up the full domain on the client's Connections page and check the matched rule, policy chain, and final node. rules are matched from top to bottom; normally, the first match determines the action. If an earlier DOMAIN-SUFFIX rule or rule set sends the resource domain to DIRECT, adding a proxy rule later won't change the result. To fix this, adjust rule priority using a client-supported override or a copy of the configuration. Reload it and send the request again to verify; avoid changes that will be overwritten when the subscription updates.

Pages load slowly but downloads are fine: check DNS separately

Check dns.enable, nameserver, and any applicable nameserver-policy in the active configuration. Also note whether the browser's Secure DNS is enabled. System DNS, the browser's built-in resolver, and the core's DNS don't always use the same path; a fast system lookup doesn't mean the slow request uses a fast resolver. If you suspect the DNS configuration, keep the original settings and change only the upstream currently in use. Record the first visit and a repeat visit separately. With fake-ip, also check whether the application is routed through the core so you don't mistake a mapped address for the site's real IP.

Slow only with TUN enabled: compare using the same browser

First, disable TUN and enable the system proxy, then run one browser test. Next, disable the system proxy and enable TUN, test the same URL, and confirm in the connection logs that the rule and node are unchanged. System proxy mode covers only apps that follow the proxy settings, so this comparison can't tell you whether every app can use the system proxy. In Windows 11, find the manual proxy settings under Settings → Network & internet → Proxy. If the client uses mixed-port: 7890 and listens on the local loopback address, you can point the manual HTTP proxy to 127.0.0.1:7890. Use the port and listen address from your actual configuration.

If several sources are slow only in TUN mode, first check whether another VPN or traffic-filtering tool is also managing routes. Then check the TUN stack and auto-route settings supported by your client. If small requests work but long transfers repeatedly stall, MTU may be worth investigating—but rule out node and packet-loss issues first. Don't assume an arbitrary MTU value is a universal fix. Retest the same file after each change and restore the original value if there's no improvement.

Retest and record: keep results others can reproduce

Record the test time, connection method, client and core versions, current mode, matched rule, final node, file size, three download times, and number of failures. For a slow single node, share the A → B → A results. For peak-hour slowdowns, share morning and evening comparisons from both days. If only TUN is slow, include connection logs from both routing modes. Before sharing, remove subscription links, tokens, credentials, and private LAN addresses that don't need to be public. Use this evidence to identify which layer needs attention before deciding whether to switch nodes, contact your network provider, or change local settings.

Download the latest version of Clash