IceRiver Web Interface 404 Error / Can’t Access Miner IP

Clear browser site data for the miner IP. Chrome: chrome://settings/siteData → search miner IP → delete. Firefox: about:preferences#privacy → Manage Data → delete. Then close ALL browser windows (not just the tab) and reopen. This evicts the old SPA bundle that the new firmware no longer serves. Roughly 40% of `404 after firmware upgrade` reports clear with this single step.

Try a private/incognito window and a second device. Ctrl+Shift+N (Chrome), Ctrl+Shift+P (Firefox), or use your phone on WiFi. If the UI loads in private mode but not in your normal session, you have confirmed cache. The cleanup in step 1 then fixes it permanently.

Hard power-cycle the miner. Unplug the AC power cable from the PSU completely. Wait 30 seconds — long enough for capacitors to drain and the controller userland to fully die. Plug back in, wait 3 minutes for boot and DHCP renewal. Retry the IP. This restarts a hung `httpd` process and clears any leaked file descriptors. Don't skip the 30-second wait — a quick reboot does not drop controller RAM.

Verify network reachability. From a LAN host run `ping <miner-ip>`. If the ping succeeds, the miner is alive — your problem is the web UI specifically. If the ping fails, you have a separate network/DHCP problem and the 404 is a red herring; switch to the IceRiver DHCP / Detect-IP playbook before proceeding.

Check the IceRiver firmware portal at iceriver.io/firmware-download for known-good versions. Note the previous stable filename — you may need it in Tier 3 if reflash becomes necessary. The KS5L/KS5M and KS0 Ultra have a track record of rough firmware releases since 2024.

20-second factory reset. Locate the small recessed reset button on the controller end-plate (KS uses a pinhole — paperclip works). Press and hold for a full 20 seconds until the red LED beside it comes on solid. Release. Miner reboots itself. Wait 3 minutes. Use IceRiver Detect-IP from a Windows PC on the same LAN to re-find the miner. Re-enter pool config from scratch on the restored UI.

Use `curl` for a clean read of what the controller is actually serving: `curl -v http://<miner-ip>/ -m 10`. `Connection refused` = httpd is down (power-cycle). `HTTP/1.1 404` = httpd up but content broken (Tier 3). Empty body with 200 = SPA bundle missing. `Connection timed out` = httpd wedged mid-handshake. Each result routes you to a different fix — and crucially distinguishes a network problem from a web-server problem.

Watch the controller TCP stack with `nmap -p 22,80,443 <miner-ip>`. If port 80 is `closed` while ping succeeds, httpd failed to bind on boot. If port 80 is `open` but the response is 404, the server is up. If port 22 (SSH) is open on your KS variant — KS0 family historically allowed SSH with default creds — you can shell in for further diagnosis. Controller-level shell access is the difference between recovery and a $400 board swap.

Power-cycle then watch boot timing with a stopwatch. The KS controller takes ~90-120 s for full boot. If the web UI is reachable for the first 30 s and then 404s, httpd is crashing on startup-completion (often a config-read race — Tier 3 step 13). If 404 from the moment of network availability, the web partition is suspect (Tier 3 step 11).

Disconnect from the internet and retry locally. Some KS firmware versions phone home for a license/telemetry check on first web hit; if the WAN is unreachable the SPA can hang waiting for a remote endpoint. Disconnect WAN, leave the miner on LAN, retry the UI. If it loads when isolated, you have confirmed a stuck remote-call dependency in firmware — flag for upstream firmware update.

Reflash the SD card / eMMC with the official IceRiver image. Power off, remove the controller microSD (KS0/KS1/KS2/KS3 family — KS5 revisions vary, check before assuming), use Balena Etcher (verifies post-write) or `dd if=iceriver_ks*.img of=/dev/sdX bs=4M status=progress conv=fsync` on Linux. Reinsert, boot, wait 5 minutes for first-boot setup. This restores the `www/` partition and resets all userland config.

Replace the SD card if reflash fails verification. SD cards die — the IceRiver factory cards are entry-grade industrial parts and don't last forever under continuous read/write. If Balena Etcher reports a verify mismatch, the SD has bad sectors. Buy a Class-10 industrial-grade microSD (SanDisk High Endurance, Samsung PRO Endurance — 32 GB is plenty), write the official image, swap. New SD + reflash recovers about 70% of `post-reflash still 404` cases.

Roll back to the previous-stable firmware. If the 404 began immediately after a firmware upgrade and the new firmware is the prime suspect, write the previous stable image. You may need to ask IceRiver support or check community archives — the official portal sometimes only hosts the current version. Specifically watch for KS0 Ultra and KS5L/KS5M release notes — these models have shipped buggy firmware multiple times.

Avoid 3rd-party OC firmware on a recovery attempt. The community OC firmwares (`xyys`, `tswift`) for KS0 are excellent for performance but a poor choice for 404-recovery — if the OC firmware has its own web-UI bug, you have stacked two unknowns. Get back to a working stock firmware first; then re-apply OC firmware once the dashboard is healthy.

SSH into the controller (where available) and tail the web server log. On variants with SSH open: `ssh root@<miner-ip>` (default creds vary by FW version). Once in: `dmesg | tail -50` for kernel-level errors, `ps | grep httpd` to confirm the server is running, `cat /var/log/httpd.log` if present. A line like `bind: address already in use` or `cannot read /www/index.html: I/O error` tells you exactly which fix to jump to.

Stop DIY when UI is still 404 after browser cache clear, power-cycle, 20-second factory reset, AND SD card reflash with a known-good card. You have eliminated every software cause. The controller board is the failure — RAM, eMMC (where soldered), or main SoC. Book a D-Central ASIC Repair slot.

D-Central bench process: controller-level diagnostics with bench logic analyzer, eMMC raw-read with a programmer to confirm filesystem damage versus storage-side failure, controller swap with graded inventory (we keep KS-series controllers on hand), full post-repair burn-in with web UI accessibility verified at 24 hours. One of the few Western shops doing IceRiver controller-level work.

Ship safely — controller board and end-plate only, not the whole miner. Anti-static bag the controller, double-box with at least 5 cm of foam every side. Include a note with: KS model, firmware version observed before failure, what tiers you tried, your contact info. Per-unit diagnostic time is the biggest line item on a repair invoice — hand us the diagnostic context and you save real money.