[BUG] - crash when longtime run PixelStreaming

[mail555]

UE Version:
UE 5.2

Frontend Version:
UE5.2-0.6.5

Problem component
Frontend, Pixel Streaming C++ plugin

Description
after long time run full eidtor PixelStreaming from browser, UE editor crash and show "out of video memory trying to allocate a rendering resource. make sure your video card has the minimum required memory, try lowering the resolution and/or closing other applications that are running. exiting"

Platform (please complete the following information):

• OS: [Windows 11]
• Browser + version [edge 120]

[mail555]

whether ue editor use directx12 or directx11 all crash.

[Belchy06]

Hey @mail555,

We're aware of this issue with the fix having landed in the ue5-main branch, and scheduled to be a part of the 5.4 release.

Alternatively, if you're hard-stuck using 5.2, I'd recommend using a source build of the engine and copying the code change seen here to your engine.

Cheers

[mail555]

after change VideoEncoder.h TSharedPtr& To TSharedPtr, i test stream full editor in ue5.2, browser and editor still crash and show error:
GPU Crashed or D3D Device Removed.

Use -d3ddebug to enable the D3D debug device.
Check log for GPU state information.

The UE-xxx Editor has crashed and will close

Fatal error: [File:C:\Engine\Source\Runtime\D3D12RHI\Private\D3D12Util.cpp] [Line: 873]
Interfaces.CopyCommandList->Close() failed
at C:\Engine\Source\Runtime\D3D12RHI\Private\D3D12CommandList.cpp:268
with error E_INVALIDARG

[Belchy06]

Is this crash after running for a long time, or is it happening instantly?

I'd suggest trying to launch with -d3ddebug so we can try and see what's triggering the crash.

Cheers

[mail555]

I use -d3ddebug testing,find some error log,but not crash until now.
The previous several times crash after running some minutes, not very long time or instantly.

LogWindows: Failed to load 'WinPixGpuCapturer.dll' (GetLastError=126)
LogWindows: Failed to load 'nvcuda.dll' (GetLastError=126)
LogWindows: Failed to load 'nvcuvid.dll' (GetLastError=126)
LogWindows: Failed to load 'nvEncodeAPI64.dll' (GetLastError=126)
LogPixelStreaming: Error: Selected codec was not a supported codec, falling back to negotiating codecs...
LogD3D12RHI: Create ID3D12DescriptorHeap: Addr=0x000001A3B1AD6070, ExtRef=1, IntRef=0
LogD3D12RHI: Error: ID3D12CommandQueue::ExecuteCommandLists: Using ResourceBarrier on Command List (0x000001A4243F1C60:'FD3D12CommandList (GPU 0)'): Before state (0xE0: D3D12_RESOURCE_STATE_DEPTH_READ|D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE|D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE) of resource (0x000001A3D8E51DB0:'SceneDepthZ') (subresource: 1) specified by transition barrier does not match with the state (0x40: D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE) specified in the previous call to ResourceBarrier

[mail555]

seem memory leakage still exists after change TSharedPtr TChildResource, still appear "Out of video memory trying to allocate a rendering resource. Make sure your video card has the minimum required memory, try lowering the resolution and/or closing other applications that are running. Exiting..."

[Belchy06]

Okay, interesting.

From our internal QA, this solution fixed the memory leak and we've been able to have instances running for days without crashing.

To double check, I'd try running a sample project (eg the Third Person template) and seeing if that still crashes with the Out Of Memory exception.

Let me know how that goes and we can see what to do from there.

Cheers

[mail555]

I run pixel streaming not in game such as third person template, i run in full editor pixelstreaming.
my computer memory total 32GB.
my operater in full editor pixelstreaming is click editor custom button, it will createproc to importassetauto,after auto import one asset, proc will finish and loop createproc to importassetauto, in my computer memory will increase until Out of video memory.
but do same operate in editor not full editor pixelstreaming memory is enough do this,so i think is full editor pixstreaming(not game pixstreaming) memory problem.

[Belchy06]

Ah yes, there was a memory leak in the editor PixelStreaming video input. This has been fixed in later versions of the engine. Get Outlook for iOS<https://aka.ms/o0ukef>

…

________________________________ From: mail55 ***@***.***> Sent: Monday, February 5, 2024 11:44:09 PM To: EpicGames/PixelStreamingInfrastructure ***@***.***> Cc: William Belcher ***@***.***>; State change ***@***.***> Subject: Re: [EpicGames/PixelStreamingInfrastructure] [BUG] - crash when longtime run PixelStreaming (Issue #478) I run pixel streaming not in game such as third person template, i run in full editor pixelstreaming. my computer memory total 32GB. my operater in full editor pixelstreaming is click editor custom button, it will createproc to importassetauto,after auto import one asset, proc will finish and loop createproc to importassetauto, in my computer memory will increase until Out of video memory. but do same operate in editor not full editor pixelstreaming memory is enough do this,so i think is full editor pixstreaming(not game pixstreaming) memory problem. — Reply to this email directly, view it on GitHub<#478 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AG7STJNNS6PVDXCJRXCEJXLYSDO2TAVCNFSM6AAAAABCJ6LESCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMRXGA2DIMJQGI>. You are receiving this because you modified the open/close state.Message ID: ***@***.***>

[mail555]

https://github.com/EpicGames/UnrealEngine/commit/ea8a1723f1c1ef25c529232aa7cd3cff76ce3962 change as later version of the engine,effect is not improve markedly in my test.

[mail555]

Hey @Belchy06 , a good update for you,

except synchronization VideoEncoder code, i change GraphicsRHI to dx11, set DynamicGlobalIlluminationMethod to 0 in DefaultEngine.ini, then ue5.2 full editor streaming run normal, make use of memory very stable, memory keep Low utilization rate.

but i don't indeterminacy which change play a decisive role, dx11 or DynamicGlobalIlluminationMethod or VideoEncoder code

[TxMat]

Same issue here. Will this fixes be deployed to 5.3 ?