Experimental v5 crashes, ev4 works

[DavisNT]

Experimental v5 (ALVR server binaries from GitHub release and ALVRClient built from JackD83/ALVRClient@dbaaf13 or JackD83/ALVRClient@7a20a4e + JackD83/ALVRClient#1) is constantly crashing when starting Steam.

Experimental v3 (ALVR server binaries from GitHub release and ALVRClient built from source with the same modification) is working.
All ALVR settings (%LOCALAPPDATA%\ALVR folder) on PC were deleted. FFR was not enabled.

Excerpt from the crashlog:

Exception: Unhandled Exception.
ExceptionCode=E06D7363
Address=000007FEFD2AB87D (000007FEE4DF0000 + 00000000184BB87D)
OSVer: MajorVersion=6 MinorVersion=1 Build=7601
Module: 000007FEE4DF0000
----------------------------------------
[00:20:18.089 698] got tracking info 3582 -0.313637 -0.078612 -0.065789 0.943994
[00:20:18.089 698] CONTROLLER 0 0.000000,0.000000,0.000000 - 0.000734,-0.004234,0.005565
[00:20:18.089 698] Controller0 0 1: 00000000 00000013 0.000000:0.000000
[00:20:18.089 698] CONTROLLER 1 0.000000,0.000000,0.000000 - -0.000308,0.000604,0.002268
[00:20:18.089 698] Controller1 1 2: 00000000 00000011 0.000000:0.000000
[00:20:18.089 698] Rotation Matrix=(0.978984, 0.173519, -0.107151, 0.000000) (-0.074897, 0.794608, 0.602485, 0.000000) (0.189685, -0.581798, 0.790905, 0.000000)
[00:20:18.089 698] GetPose: Rotation=(-0.313637, -0.078612, -0.065789, 0.943994) Position=(0.010037, -0.055899, -0.135191)
[00:20:18.089 698] Try to send.
[00:20:18.089 698] Poller::Do(). Select 10000 us
[00:20:18.089 698] Cuda kernel execution failed. code=18 invalid texture reference
[00:20:18.089 698] Unhandled Exception.
ExceptionCode=E06D7363
Address=000007FEFD2AB87D (000007FEE4DF0000 + 00000000184BB87D)
[00:20:18.089 698] GetNextSwapTextureSetIndex 0000000040001280 0000000040001340 0 0
========== Exception info start ==========
ExceptionCode=E06D7363 Address=000007FEFD2AB87D ThreadId=7388
RaiseException(RaiseException) +61
========== Exception info end ==========

Has anybody else faced similar issue?

[JackD83]

A very strange error.
What hardware setup do you have? You already wrote me that you updated to the latest nvidia driver. I have two working setups with two different quests, but both have a GTX1070 and the latest 436.15 driver.

Have you tried to uninstall the alvr_driver and re-install? Maybe something is stuck.

Maybe @zarik5 has more insight, the changes from ev4 to 5 came from him.

[DavisNT]

I have Nvidia GTX 1660 Ti. Driver version is 436.15.
I have tried to click "Uninstall driver" and "Install driver".
Just tested and ALVR Experimental v4 is working (together with ALVRClient ev3). So the problem is introduced in ev5.
CC: @zarik5

[JackD83]

I just upgraded from Visual Studio 17 to 19 and from CUDA 9.2 to 10.1
Maybe you give it a try. You need to rebuild everything and reinstall the alvr driver

[Kuuko-fokkusugaru]

I can confirm that it crashes for me too. What crashes is SteamVR and then I get an error message which I assume it's from SteamVR too? Or maybe ALVR but the application keeps it open and "working" (not hang). V4 works perfectly fine and it's the one that I am using right now. I have tried several times going back and forth with V5 and there is no way of making it to work. I always uninstall and reinstall drivers. There is something on that build that makes it to not work. I don't even enabled FFR. So, unless there is some charasteristic introduced that requires Win10 (I use Win7 right now) I have no idea what could it be. Either way, if for some reason the addition of FFR would require such feature, it would be better if we could still use the software turning it off otherwise people like me wouldn't be able to use the whole ALVR all together.
Thank you for your hard work ♥

[DavisNT]

Looks like this might have been some kind of build issue - I have built the latest Git version 29416ee (using Visual Studio 2019 and Windows 10 SDK 10.0.17134.12) and it worked.

@JackD83 What version of Windows 10 SDK was used to build Experimental v5?

[zarik5]

@DavisNT the error you reported refers to CUDA; the relative code is executed if "IsWindows8OrGreater() == FALSE". While writing the FFR code I explicitly commented out every CUDA reference because I couldn't compile otherwise (and it wasn't a problem for me because I'm on Win 10). I only touched VideoEncoderNVENC, which in turn creates NvEncoderCuda or NvEncoderD3D11, so I don't see the reason why just the CUDA code would be broken.

[JackD83]

ev5 was build with 17134. With the upgrade I switched to 18362. It should be set to "use latest version" and I hope that works for everyone.

I think the big change was the switch to Cuda 10.1.
The whole architecture of the Geforce RTX and the GTX 1660 was released after the release of the latest Cuda 9.2 update.

Upgrading was not that hard and Visual Studio 19 and Cuda are freely available. They can even be installed side by side. Just make sure to set the CUDA_PATH environment variable to the one you want use.

[Kuuko-fokkusugaru]

For the rest of humans like me, does that means that it's fixed? When will we see the next experimental release? I am not avid enough to compile and such :(

[JackD83]

I made a release of experimental v6 and hope that its fixes this error

[DavisNT]

I can confirm - v6 works well (at least on my PC where v5 always crashed).