You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
AGC: Intel Iris Xe (Windows) driver crashes in vkGetPipelineCacheData from the pipeline cache saver thread about 10 s after start #4781
With ANYPS5_GPU=Intel, DOOM Native Port (PPSA99666) and ccleste (PPSA59322) crash 3 of 3 runs at about 10.7 s. Both render at 60 FPS for 120 s on the NVIDIA GPU of the same machine. DOOM's report:
FATAL: unhandled exception 0xc0000005 on thread 27072 ''
rip 0x00007ff93db2d102 igvk64.dll+0x14d102
read of 0x0000000000000170
So the fault is inside the Intel driver during vkGetPipelineCacheData (core/libs/prx/libSceAgcDriver/Graphics/src/PipelineCache.cpp:109-111). The call comes from the saver thread's first periodic save (SaveIntervalSeconds = 10), while the title's threads keep creating pipelines with the same VkPipelineCache.
Checks
Saver disabled: with ANYPS5_NO_SHADER_CACHE=1, so no cache directory and no saver thread, DOOM on the Intel GPU runs past 40 s (27 FPS) instead of crashing at 10.7 s.
Validation layer: Khronos validation (Vulkan SDK 1.4.363.0, errors and warnings, thread safety on by default) reports nothing for DOOM on either GPU before the crash. The crashing call passes through VkLayer_khronos_validation.dll.
Two Intel drivers: reproduced on 32.0.101.6790 (April 2025) and 32.0.101.7092 (September 2026).
The cache is created without VK_PIPELINE_CACHE_CREATE_EXTERNALLY_SYNCHRONIZED_BIT, so concurrent vkGetPipelineCacheData and pipeline creation is valid usage. This looks like an Intel driver bug that AnyPS5 runs into. A workaround could serialize the save against pipeline creation on Intel, or merge into a separate cache before reading it.
Separately
ccleste with ANYPS5_NO_SHADER_CACHE=1 on the Intel GPU gets past 10.7 s but then stops with VkResult is ERROR_DEVICE_LOST from the title's own Vulkan code at about 12.7 s. That is not caused by the saver. I will look at it separately.
Closing: fixed on main by #4795 (pipeline cache snapshots are serialized against pipeline creation). If it still happens on current main, comment here with the title ID and the log line.
Activity
oneandonlydean commentedon Oct 14, 2026
Closing: fixed on main by #4795 (pipeline cache snapshots are serialized against pipeline creation). If it still happens on current main, comment here with the title ID and the log line.