Minimum GPU to train yolov5 extra large (version 6) model with 4K images


I’m in a project that requires training a yolov5 model with 4K images. I have a RTX 3060 12 GB and I’m getting a cuda out of memory error. I’m wondering since the model is very heavy and the images are very large what would be the best or minimum GPU to use for training? Also I’ll need to convert the model (pytorch) into onnx. Then It should be deployed d using cudnn. The idea is to train the yolov5 in a machine and make the inference in another machine. What would be the best gpu for inference with 4K images? I mean I’ll need to have 2 GPUs one for training and other for inference. I just want the recommended GPUs for both cases.


Hi @p.carvalho ,

Can you please share the detailed api logs with us?