CUDA 12

[theo2f]

Hello!
Are there any expectations of being able to run it soon using CUDA 12?
I want to use it on Azure/AWS, but the GPUs offered require the latest version.

[12343954]

On my test, RTX 2080Ti/2060 was very slow on CUDA 12.

I don't have RTX 40 series card,
but I saw someone tested yolov8.net on RTX4060 and CUDA 12, the speed was in milliseconds, within 10 milliseconds, so unbelievable!!!!

you can compile yolo_cpp_dll.dll and test the code if you have CUDA 12 and RTX 40 series card.

I'm very glad to see you report the result.

[theo2f]

Can you help me to compile the dll? I don't know where can I get the code and how to do it.

[12343954]

YOLO strongly depends on your hardware and cuda, cudnn, opencv, operation system. so I can't compile for you, sorry.
here is the official code of YOLO v3/v4 : https://github.com/AlexeyAB/darknet.
follow the compilation steps in README to compile the dll (modern method).
or watch the old tutorial (Windows 10/11 only) : https://youtu.be/zT8eDXpslXw

you may also use other yolo versions from other repos.

[theo2f]

Compiled...

But the yolo_cpp_dll.dll file was not created.
Maybe I need some parameter or something that I don't know.

[12343954]

default compilation is execution file “darknet.exe” (command only).

it cannot use in .net projects directly.

to generate the "yolo_cpp_dll.dll", you must compile the "yolo_cpp_dll.sln", in path "./build/darknet/".

watch the youtube tutorial above.