OpenCV with CUDA on a $480 AI WS PC, is't worth it? How much faster is OpenCV with CUDA? RTX3060

This is my first video to celebrate that my "new" PC built from scond hand parts works with:
Ubuntu 24.04
OpenCV 4.12
CUDA 12.9

The PC:
ASUS Rog Strix Z390-F
Intel i5-9600K
32GB RAM
Fractal Design Effekt 500w PS
Gigabyte RTX3060 oc 12GB
all those parts econd hand, bought via Swedish Tradera auction site

Only new parts:
DeepCool AG300 (cheapest air cooler I could find, INET)
NVMe 512GB (from Amazon)

Total build cost, all above parts: 4600SEK ($480).

The c++ program is very simple, it basically reading a picture, repeats a bilateral filter 100 times with or without GPU.

The difference is aprox 10 times faster with GPU, even with a "old" RTX3060.
It's just a very simple test of the GPU performance.

I start with the GPU version, compile and run.
Then a change the program to run without GPU, compile and run again.

I hope this can inspire others to build their own cheap AI workstation, both for LLMs and OpenCV among other applications.

I bought another motherboard first, but that was faulty, I replaced that.
Then I had some real issues with the CUDA Toolkit, I believe it was the BIOS upgrade which resolved those issues.

I'm not that used with Linux, my main experience is with Raspberry pi.
But I kinda start to like Ubuntu, especially for those AI things with GPUs.

I hope you enjoy the video, I will do more videos, and even put some sound to them so you don't fell asleep.

/Michael, swedish developer in south of Sweden (Blekinge county).

Видео OpenCV with CUDA on a $480 AI WS PC, is't worth it? How much faster is OpenCV with CUDA? RTX3060 канала Michael Steger