experiencing crashes on AMD

[CitroenGames]

[I][SCRI] Initializing scripting [ScriptManager.cpp:25][ScriptManager][Main]
[I][CORE] Application initialized [App.cpp:347][initInternal][Main]
[I][CORE] Entering main loop [App.cpp:395][mainLoop][Main]
[I][GR  ] Cleaned 6 render targets [RenderGraph.cpp:1327][periodicCleanup][Main]
[F][VK  ] Vulkan function failed (VkResult: VK_ERROR_DEVICE_LOST): vkQueueSubmit(self.m_queues[queueType], 1, &submit, fence->getHandle()) [VkGrManager.cpp:271][submit][Main]

when loading Sponza level it works fine for first 10 seconds but it randomly crashes only on AMD.
Edit: on some AMD cards its causes driver crash too.

[godlikepanos]

Hi. Can you please paste the whole log? Need to see which AMD it is and which driver.

[CitroenGames]

tested on:
RX 6950 XT (Experiencing Crash)
Driver: 24.10.1

RX 580 (Experiencing Crash)
Driver: 24.9.1

RTX 2070 Super (HAS NO ISSUES)

[godlikepanos]

I'm not sure what might be wrong. I normally test on a RX 7800 and there it seems to be working. Don't have something older to test on. 6000 and older are running on a different driver compiler and this might be causing issues.

[CitroenGames]

Interesting... is there a way to debug why its crashing?

[godlikepanos]

One thing you can do now is to use the https://gpuopen.com/radeon-gpu-detective/ to get more info on the device lost.

Or you can wait a bit for me to implement VK_EXT_device_fault. This extension provides information on device losts.

[CitroenGames]

i will try it tomorrow.

[CitroenGames]

hey,

i have issues understanding how to use this thats why i didnt respond yesterday i have it downloaded but its not as easy as i thought.

[CitroenGames]

I managed to figured it out:

===================
CRASH ANALYSIS FILE
===================

Crash analysis file format version: 1.0
RGD CLI version used: Radeon GPU Detective 1.3.0.10
Input crash dump file creation time: Thu Nov 21 15:37:18 2024
Input crash dump file name: C:\Users\USER\Documents\rgd_dumps\Sandbox\Sandbox-20241121-153718.rgd
Crashing executable full path: G:\github\anki\build\Binaries\Sandbox.exe (PID: 30628)
API: Vulkan

===========
SYSTEM INFO
===========

Driver info
===========
Driver packaging version: 24.20.19.01-241011a-408380C-AMD-Software-Adrenalin-Edition
Driver software version: 24.10.1
Dev driver version: dev
Experiments : no driver experiments were enabled.

Operating system info
=====================
Name: Windows 11 Pro
Description: 26100.1.amd64fre.ge_release.240331-1435
Hostname: DESKTOP-AV5AEV9
Memory size (physical bytes): 68618747904 (63.91 GB)
Memory size (swap bytes): 72913715200 (67.91 GB)

CPU info
========
CPU count: 1
CPU #1:
	Name: AMD Ryzen 9 5950X 16-Core Processor
	Architecture: x64
	CPU ID: AMD64 Family 25 Model 33 Stepping 0
	Virtualization: enabled

GPU info
========
GPU count: 1
GPU #1:
	Name: AMD Radeon RX 6950 XT
	Device ID: 0x73a5
	Device revision ID: 0x29
	Device family ID: 0x8f
	Device graphics engine ID: 0xd
	Device PCI revision ID: 0xc0
	Big SW version: 2023.0.33413048
	Memory type: Gddr6
	Memory heap count: 2
		Memory heap #1:
			Heap type: Invisible (GPU memory, invisible to CPU)
			Heap size: 16894656512 (15.73 GB)
		Memory heap #2:
			Heap type: Local (GPU memory, CPU-visible)
			Heap size: 268435456 (256.00 MB)



===================
MARKERS IN PROGRESS
===================

Command Buffer ID: 0x2d70
=========================
Dispatch [2 repeating occurrences]

Command Buffer ID: 0x2d73
=========================
Barrier
Dispatch
Barrier



=====================
EXECUTION MARKER TREE
=====================

Legend
======
[X] finished
[>] in progress
[ ] not started

Command Buffer ID: 0x2d70 (Queue type: Direct)
==============================================
[X] ----------Barrier----------
[X] DrawIndexedIndirectMulti
[X] DrawIndexedIndirectMulti
[X] DrawIndexedIndirectMulti
[X] DrawIndexedIndirectMulti
[X] ----------Barrier----------
[>] Dispatch(ThreadGroupCount=[24,14,14])
[>] Dispatch(ThreadGroupCount=[384,216,1])

Command Buffer ID: 0x2d73 (Queue type: Direct)
==============================================
[>] ----------Barrier----------
[>] Dispatch(ThreadGroupCount=[24,14,1])
[>] ----------Barrier----------
[ ] Draw(VertexCount=3, InstanceCount=1)
[ ] Draw(VertexCount=3, InstanceCount=1)
[ ] Draw(VertexCount=3, InstanceCount=1)
[ ] DrawIndirectMulti
[ ] DrawIndirectMulti



==================
PAGE FAULT SUMMARY
==================

Offending VA: 0x100a9d000

Resource timeline
=================
INFO: no resource timeline information available.

Associated resources
====================
INFO: no associated resources detected for the offending VA.
```

[godlikepanos]

Difficult to spot which commands are those. Which resolution are you running with?

[CitroenGames]

on rx 6950xt im testing it on 4k
on rx 580 1080p

[godlikepanos]

OK I managed to find which shader is failing from the log you pasted before. Thank you. It's the volumetric lighting. Will try to see what might be wrong there.

[CitroenGames]

OK I managed to find which shader is failing from the log you pasted before. Thank you. It's the volumetric lighting. Will try to see what might be wrong there.

good to hear! im more than grateful to help out the community