"DDE Server Window: Devenv.exe" crash when installing CUDA 10.2

During the install process, I’m getting a devenv.exe crash when installing the CUDA VS integration.
I’ve only selected the “CUDA” components in the installer (don’t want to install GFE or overwrite my current, more recent Game Ready driver).

Anyone else with this problem? Using the latest CUDA 10.2 installer on Windows 10 (1909), VS 2019 CE

The installation continues after the error, and in my limited testing CUDA functionality seems okay otherwise.

(Same error message as this user here: https://devtalk.nvidia.com/default/topic/1064596/cuda-setup-and-installation/cuda-install-nsight-visual-studio-edition-devenv-exe-unknown-software-exception-error/)

Fixed this by doing a “repair install” on my VS2019 installation first.

Just for the hell of it, I then uninstalled CUDA (and assorted tools again), then attempted to reinstall. Same devenv.crash error during install.

So something in the uninstallation of CUDA leaves behind traces in VS that makes re-installation of CUDA crash devenv.exe. I can only get a trouble-free install if I “repair / reset” VS2019 CE first.

Here’s the woefully unhelpful event viewer output:

Faulting application name: devenv.exe, version: 16.3.29509.3, time stamp: 0x5dc69060
Faulting module name: KERNELBASE.dll, version: 10.0.18362.476, time stamp: 0x540698cd
Exception code: 0xc000041d
Fault offset: 0x001135d2
Faulting process id: 0x2d7c
Faulting application start time: 0x01d59ff6e6412675
Faulting application path: C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\IDE\devenv.exe
Faulting module path: C:\WINDOWS\System32\KERNELBASE.dll
Report Id: cae9c241-d6e3-4fcf-ad9c-319c64f45c8b
Faulting package full name: 
Faulting package-relative application ID:

I ran sfc /scannow and DISM repair - my windows system doesn’t seem to be corrupted or anything. Could replicate this bug on a different system (kind of - there I got the same error in uninstalling).

VS Studio was version 16.3.10, the most recent one at the time of writing.

On a fresh install of windows 1909 I even get this error :(
And my “repair VS 2019 first”-trick doesn’t work anymore.

The installation continues though, but I have no idea whether or not the components were installed correctly.

Did some further digging: these errors are logged on all Win10 1909 systems I try this on, just sometimes they get logged silently in the event log, sometimes they’re accompagnied by a popup that gives you “ok” or “cancel to debug”.

The components however seem to be registered within VS 2019 (I have an nsight menu and I just test-stepped debugged through a sample project, which compiled and ran fine …)

Thanks for reply though we didn’t find solution T T

Same happens for me :

Cuda 10.2.89
Visual Studio Community 2019 version 16.3.9

Its a kernel base exception being thrown
76FD35BA push ecx
76FD35BB lea eax,[esp+1Ch]
76FD35BF push eax
76FD35C0 call _memcpy (76FF8634h)
76FD35C5 add esp,0Ch
76FD35C8 lea eax,[esp]
76FD35CB push eax
76FD35CC call dword ptr [__imp__RtlRaiseException@4 (77088404h)]

Same problem here.

When I’m trying to setup up a build with CMake then for a project that is supposed to be compiled with CUDA, I see the following message in the CMake “Configure” output:

Looking for a CUDA compiler - NOTFOUND

I suppose we are not seeing the fix until 2020. What a freaking time it was to switch to VS2019.

@stephan.muehlstrasse: That seems like a different problem.

@ForceflowBE You are probably right. I initially thought that the CMake error message is a result of the CUDA Visual Studio integration not being installed correctly, but after looking deeper into the CMake log files it seems to be a problem in CMake itself.