Do you have any plans to support the Windows system？

[zhang0557kui]

[shiremathNV]

@zhang0557kui - thanks for your interest in CV-CUDA. We are considering adding native Windows OS support in a future release. Could you provide more information on the specific OS and compiler of interest to you?

[dlesage-nvidia]

Adding for completeness: CV-CUDA is currently compatible with WSL2 with Ubuntu >= 20.04

[adegerard]

@zhang0557kui - thanks for your interest in CV-CUDA. We are considering adding native Windows OS support in a future release. Could you provide more information on the specific OS and compiler of interest to you?

Hi! I have the same question because all my image processing chain is working on windows and I have some packages compiled with VC++ (windows only) I would like to be able to compile (or directly use it if packaged by nvidia!)

• Windows11 latest version
• mingw64 (msys2) or Microsoft VC compiler (community edition), whatever the version
• RTX4xxx Thanks.

edit: latest cuda/pytorch/driver versions. python >= 11.8

[PaperStrike]

We are considering adding native Windows OS support in a future release. Could you provide more information on the specific OS and compiler of interest to you?

I would like to share my use case for Windows support. I am writing a OBS (Open Broadcaster Software) plugin that requires running OCR in realtime (screen text filtering/highlighting), and everything works (Visual Studio 2022 17.4-17.10) but the throughput. I've tracked down the performance issue and the last bottleneck is the CPU. CPU usage being 100% has limited the concurrency that I can only use 2-3 CUDA streams with GPU usage capped at ~20%. And then I found here, this official CV-CUDA blog has well explained the issue. Mature Chinese OCR pipelines are all two staged that they have two pre- and post- processing steps. (I really remember an official blog post said ByteDance got a 7x speed up on OCR inferences but I can't find the sentence anywhere now.)

The problem in my use case is that most people live stream on Windows, it feels so hard and inefficient to forward every ID3D11Texture2D, Mat, or UMat to WSL processes and forward the results back in realtime. It's also quite difficult to direct our users to setup WSL to use the plugin.

It's not limited to OCR, many fancy things that needs AI computer vision haven't yet come to the live streaming industry. They requires fundamental high-performance architectures.

Hope this helps. Thank you for considering Windows support.

[rememberBr]

Adding for completeness: CV-CUDA is currently compatible with WSL2 with Ubuntu >= 20.04添加完整性：CV-CUDA 目前与 Ubuntu >= 20.04 的 WSL2 兼容

I am even more looking forward to the native Windows support, as there is a performance loss when calling CUDA through WSL2, in my opinion, I expect the tool to bring benefits in real-world scenarios, not just the ability to run, and I look forward to your future work.