Visual Studio 2022 - CUDA Toolkit support

Does the current CUDA Toolkit 11.X support Visual Studio 2022 (was released recently) ?

4 Likes

You can always answer this question by looking at the windows install guide associated with the CUDA toolkit version you are interested in. (answer is no for CUDA 11.5.1)

Ok, but the real question is when. When does the CUDA setup support VS2022? Is it on the roadmap yet?

I’m generally not able to answer questions about what will happen in the future, or when. If the past is any indication, VS2022 will eventually be supported.

1 Like

Guys, I mean from Nvidia,

That isn’t very pleasant. I see a lot of questions on the forum related to Visual Studio 2022 support. I know from the past that supporting a new version of Visual Studio is a big thing and takes a lot of time, but it would be great if you share something with the community.

I see that Jensen speaks a lot on the GTC about new shiny products, discussing the future and great things.

But looking into Community moderation
 From the list of questions about Visual Studio 2022, you answered one where Hannes asked about things covered by the User Guide and where you successfully pointed out. But talking about intention: I believe Hannes just want to know “when”, but instead of sharing a roadmap or just telling “no, sadly, it’s not supported”. You tell us “RTFM”.

That’s wrong and sad.
I love Nvidia, but not Nvidia Community support.

I know that’s banality, but I hope soon we feel full-featured CUDA development experience with VS 2022.
Cheers! Have a good time!

9 Likes

Well, I hoped also to get a bit more information when VS 2022 will be supported by CUDA toolkit.
But, actually I am quite sure that it will be supported by the next release of CUDA toolkit (11.6 ? 12.0 ?).
For recent Visual Studio versons, it always has been that way when I remember correctly.
And there is no question that they will support VS 2022 in the future (what would be the alternative on windows ?).

@ ivanov: I have to say the exact opposite than what you suggest.
The support from NVIDIA people (robert etc.) and also from some non-NVIDIA people (njuffa, 
) in this forum is excellent, in my experience. Their support really saved my butt in a couple of occasion with difficult CUDA-related problems I had in the past. That is actually one thing which can really contribute to the success of a framework (like CUDA definitly is) or not. So thanks !

What I also see is that they seem not to be allowed to make ‘forward-looking’ statements regarding the CUDA toolkit. This seems to be an internal policy, but I still would appreciate if that policy could be ‘opened up’ a bit.

1 Like

Hannes,

Thank you for a good example! I have no intention to be rude or ungrateful. I also see a lot of topics here and understand that everyone is overloaded with requests and questions. I am glad that we even have support. Anyway, it’s voluntarily for Nvidia to provide such a feedback model on top of the documentation we have.

I took a chance to raise some noise about VS2022 support because it already has a third version, and I want to know when I can clean up a bit of my storage which was seized by the old Visual Studio. =)
So, maybe I’ve stepped too far, sorry for that.

Have a good day! May the CUDA be with you!
Cheers!

2 Likes

I can’t understand Nvidia’s position on VS2022 either. How about publishing the source code of the code integration for VS2019 on GitHub? Nvidia commendably makes a lot of stuff available on GitHub. Then the integration could be adapted to VS2022 by the community. Less work for Nvidia and faster integration into VS2022.

The lack of integration is a very big problem because CMake’s CUDA support requires Visual Studio integration. I.e. currently libraries like OpenCV cannot be built with CUDA under VS2022.

1 Like

I downloaded cuda 11.2 but I have visual studio 2022 , and now while installing cuda it is not recognizing visual studio , please give me some guidance on how to install cuda correctly, most of the blogs are outdated , they use 2017 nd 2019 visual studio versions, is there any way to install cuda without reinstalling visual studio 2019 or 2017 ,
Pls help me I have my deadline coming faster and yet not completed my models.đŸ˜„đŸ˜„đŸ˜„

5 Likes

Here is some help to get the CUDA SDK to work on VS2022. It’s unofficial (you guessed) and you will bypass version checks, so results will vary but it works for me for a 360 cam editing plugin for Davinci Resolve. It does not install the full toolset, but will allow you to build your project.

Here is how I proceeded. Launch the installer so that all files are exported to a temp directory. Note the path of the directory and open it in File Explorer. From that folder, head to ‘visual_studio_integration\CUDAVisualStudioIntegration\extras\visual_studio_integration\MSBuildExtensions’. Copy all files in that folder to your VS BuildCustomization folder (normally ‘C:\Program Files\Microsoft Visual Studio\2022\Community\Msbuild\Microsoft\VC\v170\BuildCustomizations’). There you go, it’s installed. However, there is an extra check at compile time. To bypass that check, open your project settings, go to ‘CUDA C-C++/Command Line’ and add '-allow-unsupported-compiler ’ to the ‘Additional Options’ textbox.

Enjoy!

3 Likes

Voila 
 CUDA toolkit 11.6 official supports VS 2022 :-)
See https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#cuda-compiler-new-features

Thanks NVIDIA !

7 Likes

I just followed NVidia’s instructions for installing cuda 11.6. My CNTK code in visual studio now works and I didn’t have any kinks in the process. It really was a matter of slowing myself down enough to install the software and set the environment variables. I would post a link but you have to log into Nvidia before you can reach the download and instructions. :(

Nice 
 CNTK framework, I heard about but I think it’s not actively developed anymore.

With respect to pytorch, If one wants to build frameworks like ‘detectron2’ on windows (see answer of ‘east’ at [1]), you cannot use VS 2022 - at the moment. That is, because VS 2022 demands CUDA 11.6, but there is currently no pytorch package on conda channel ‘pytorch’ which is built against CUDA 11.6 
 So at least for now, one has to use VS 2019 and CUDA 11.3, then it works (I just built it). Note VS 2017 is too old (is not able to compile pytorch C++ code).

[1] python - Install Detectron2 on Windows 10 - Stack Overflow

This also allows to use CUDA in wolfram Mathematica without renamaing folder of VS 2022 MSVC compiler! THANK YOU!

So sorry to reply such a post to disturb all of you.Excuse for my broken English.But I really need to know, does CUDA support VS2022 now? If so, how to use correctly? I’ve download and try to use, but Ctrl+F5 always come with MSB368&MSB417,saying “Cannot create XAML”. How to deal with it?

no you had to install vs19

Visual Studio (Community) 2022 Preview 17.10.0 Preview 1.0 no longer works with the latest CUDA (12.3).

According to the following links the version has increased

#if _MSC_VER < 1910 || _MSC_VER >= 1940

#error -- unsupported Microsoft Visual Studio version! Only the versions between 2017 and 2022 (inclusive) are supported! The nvcc flag '-allow-unsupported-compiler' can be used to override this version check; however, using an unsupported host compiler may cause compilation failure or incorrect run time execution. Use at your own risk.