Sword of the Sea (PPSA25344)

[m335671]

• I searched this category for the title ID and no thread exists for this title on any platform.

Title: Sword of the Sea (PPSA25344)

Content version tested: 01.510.000

AnyPS5 commit: main d7a455a90 (the parent of the #6112 merge 376c04029; see the blocker section for why), no PRs on top. Also compared with release v0.1.2.

System tested: Windows 11 Pro (build 26200), WinLibs GCC 15.2.0 (r7), AMD Ryzen 9 7900, 32 GB RAM, RTX 4070 12 GB (NVIDIA 616.92)

Furthest point: In game (the opening cutscene after New Game)

What is shown

Title screen and menu render, audio plays, input works (XInput pad and keyboard). New Game loads into the opening cutscene.

Title screen, warm caches, APS5_PROFILE_DRAW=1 APS5_PROFILE_GPU=1, [present] line per 10 s:

	v0.1.2	main d7a455a90
presents per 10 s	36–37 (~3.7 FPS)	128–129 (~12.9 FPS)
GPU busy per frame	111–115 ms	55.2 ms
idle gaps per frame	159–168 ms	22.7–23.0 ms

v0.1.2 logs Windows direct host imports use comparisons, the #5692 path. d7a455a90 logs direct host imports of private memory are settled and stay watched and [guestmem] compare 151 calls 1 MiB 0.0 s, so #6060 recovers it here as well. On d7a455a90 the title screen is now GPU-bound: about 70 % of the frame is GPU busy time. The guest presents 3840x2160.

The opening cutscene runs at about 4 FPS on the first visit (overlay reading), while shaders are still being compiled: 48, 14 and 11 shader-cache misses per 10 s. I have no warm-cache figure for gameplay yet.

Current blocker

1. Cold cache: UE's render-thread watchdog kills the first New Game load. On the first two New Game attempts with a cold shader cache, the game stopped with:

   LowLevelFatalError [File:.\Runtime/RenderCore/Private/RenderingThread.cpp] [Line: 1415]
   GameThread timed out waiting for RenderThread after 120.00 secs
   

   followed by FATAL: unhandled exception 0xc0000005. During that load the shader cache took about 1,200 misses in 30 s (~1.3 GB written), with two CPU threads at 100 % and the GPU at about 1 %. Each attempt keeps what it compiled, and the third attempt reached the cutscene. That attempt also had -dpcvars=g.TimeoutForBlockOnRenderFence=900000 appended to uecommandline.txt (in a copy, not the original). I can't tell whether the shipping build honoured it or whether the warmer cache was enough. This matches the cold-load numbers in Translate once, cache, use every core: a plan to boot like a PC game and reach 30 FPS on UE5 titles (Sword of the Sea numbers) #5436 (> 700 s, mostly compute pipeline creation).
2. Current main 400baa252 aborts at start-up with Thread destructor callback is already registered on Windows too (Since #6112, every title aborts at start-up on Linux: "Thread destructor callback is already registered" #7636), so I built the parent of the fix(relinker): run guest preinit arrays before the module initializer #6112 merge.

Also seen, without visible effect: [gpu] queue 0x20 WAIT_REG_MEM ... timed out after 1000ms (function 3 ref 0x1 ...) a few times during the New Game load.

How it was run

Converted from the decrypted eboot.bin with its bundled prx/ and sce_module/ modules:

relinker --windows <decrypted>/eboot.bin app.exe

The runtime layout follows the usage guide, and the original eboot.bin, sce_sys/, engine/ and surfer/ sit in app0/. The console output was redirected to a file: with logs going to the Windows console, the title screen was noticeably slower. APS5_THREAD_PRIORITY=high was set for the later runs. No other switches were needed.

---

AI-assisted: yes (Claude Code). I played the game on my machine. Claude Code built AnyPS5, converted the title, read the logs and profiles, and wrote this report.