Training yolov5 model with 4K images (best GPU)

Hi,

I need to train a yolov5 model with 4K images. I have a RTX 3060 12 GB VRAM but I get a cuda out of memory even with batch size equal to 1. I’m wondering what is a good GPU that can work just fine to handle yolov5x6.pt with batch sizes of 16+. Also I need to convert it to onnx. What is also the best GPU for inferencing a onnx model with 4K images?

Thanks

Does anyone have an idea? This is urgent.

Thanks