Path of Exile 2 Settlers Act 4 Boss Desync and Crash Fix

What is the POE2 Act 4 Boss Desync and Crash?

The Path of Exile 2: Settlers of Kalguur Act 4 release (launched in Early Access on May 14, 2026 with patch 0.4.0) opened the new Kalguur continent and introduced four endgame bosses: Mektul, Forger of Sand, The Brine King Reforged, Olroth, Origin of the Fall, and Aukuna, the Black Sekhema. Players hitting these encounters report a consistent pattern: the boss freezes in place for 8–15 seconds while damage and projectiles still register from the server, followed by either a Disconnected from server modal or a hard client crash with Application crashed: Critical error encountered (E_GPU_HUNG) written to logs\Client.txt.

GGG acknowledged the issue in the official "Known Issues – 0.4.0" thread on pathofexile.com/forum/view-thread/3527890 on May 19, 2026, attributing it to a server-side action queue backlog when the boss enters its second phase combined with a client predictive movement mismatch on high-latency connections. A partial server fix shipped in hotfix 0.4.0c (May 23, 2026), but the crash variant still affects players on Vulkan with RTX 50-series GPUs and any player above 80 ms latency to the gateway.

When does it occur?

• During the second phase of Mektul, Forger of Sand (after his 50 % HP threshold)
• On Olroth's tornado attack when 4+ active minions are on screen
• During the Brine King Reforged's water-pillar phase when fighting at 60+ fps with screen-shake on
• When ping to the matched gateway exceeds 80 ms (especially OCE -> EU/US fallback)
• More frequently in 6-player parties than solo
• Reproducibly on Vulkan renderer + DLSS 4 + RTX 50-series GPUs

Common causes

• Server action-queue backlog when phase-2 boss spawns more than 12 entities simultaneously
• Client predictive-movement desync exceeding the 300 ms reconciliation window
• Vulkan command buffer overflow on Nvidia Blackwell during the screen-shake VFX
• Gateway routing the player to a non-optimal region (notably OCE players to Texas instead of Sydney)
• Outdated D3D12 / Vulkan driver below the 0.4.0 minimum (Nvidia 576.40, AMD Adrenalin 25.5.1)
• Particle quality set to High or Ultra exceeding the 12 GB VRAM budget on RTX 5070 / 9070 XT
• Background streaming software (OBS, Discord screen share) injecting into the present queue mid-phase

Step-by-step fixes

1. Lock the gateway to your nearest region. From the character select screen, click the gateway button at the bottom right and manually select Sydney / Singapore / Frankfurt / Washington DC / São Paulo (whichever is geographically closest, not "Auto"). The 0.4.0 auto-router is the root cause of half the desync reports. Confirm your latency in-game with /recheck_lag — anything over 80 ms means try a different gateway.

1. Switch from Vulkan to DirectX 12 as the verified fast fix on Nvidia 50-series. Quit POE2, open %USERPROFILE%\Documents\My Games\Path of Exile 2\production_Config.ini, set:

   renderer=dx12
   predictive_mode=conservative

Save and re-launch. The DX12 renderer does not trigger the command buffer overflow on the Brine King water-pillar.

1. Disable Predictive Mode if DX12 is not an option. In-game: Options -> Network -> Predictive Mode -> set to Lockstep. Lockstep adds ~50 ms input latency but eliminates the desync window that produces the freeze. GGG explicitly recommends this in their May 19 post for players above 60 ms.

1. Lower Particle Quality to Medium for Act 4 specifically. Options -> Graphics -> Particle Quality -> Medium. The Mektul sand-storm and Olroth tornado push the particle system past the VRAM budget on 12 GB cards; Medium reduces particle count by ~60 % without breaking the encounter's mechanical readability.

1. Update GPU drivers to the 0.4.0 minimum. Nvidia: install 577.41 (Game Ready, May 13, 2026) or newer using the "Custom -> perform a clean installation" path in the NVIDIA app. AMD: install Adrenalin 25.5.2 Preview (May 28, 2026) — the prior 25.5.1 branch has a known Vulkan crash on POE2 (and a separate Windows 11 WDDM bug, see KB5089549 articles).

1. Apply GGG's hotfix command-line flag. Right-click Path of Exile 2 in Steam -> Properties -> Launch Options. Add: --no-multithread-render --waitforpreload. This serializes the present queue and forces full asset preload before the boss room loads. Costs 4–6 seconds on the boss-arena transition but eliminates the GPU hang crash.

1. Clear the asset bundle cache if you previously fought the boss on patch 0.4.0a or 0.4.0b. Quit the game and delete <install dir>\Bundles2\_.index.bin and Bundles2\reactive\*.bundle. Re-launch — POE2 re-downloads the corrected bundles (~280 MB). Required because the May 23 hotfix changed boss skill timings server-side but did not invalidate the cached client predictions.

1. Reduce party size for Act 4 boss runs. The action-queue backlog scales linearly with party size; 6-player parties hit the limit 4× more often than solo. Run Act 4 bosses solo or in 2-stacks until GGG ships the server-side queue refactor (ETA: 0.4.1, late June 2026 per the dev manifesto).

If it still doesn't work

If you crash on the same boss attempt repeatedly, the encounter state may be corrupted server-side for your character — log out, wait 5 full minutes (the instance must be torn down), and re-enter a fresh instance. For repeat crashes specifically on Mektul phase 2, file a bug report on pathofexile.com/forum/forum/bug-reports/6 with the exact contents of Client.txt from <install dir>\logs\ (lines 200 before and 50 after the crash timestamp) and the gpuView trace if you can capture one — GGG's engineering team has been responsive on Blackwell crashes since the 0.4.0a hotfix. For repeat disconnects without a crash, check the GGG status page at pathofexile.com/status — gateway-level packet loss has flared multiple times since the Act 4 launch. As a last resort, switch to a standalone (non-Steam) install from pathofexile.com/download — the standalone client uses a different patcher that has not exhibited the bundle-cache corruption issue seen on Steam. Do not delete and reinstall via Steam unless you have a full save backup; the 90+ GB redownload is not a productive troubleshooting step for this specific bug.