Скачать или смотреть install cuda 11 cudnn 8 in rtx gpu for deep learning in ubantu 18.04 with correct Nvidia GPU drivers

use the below for environment variables in gedit bahrc
export PATH=/usr/local/cuda-11.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-11.0/lib64\

  / tutorial-cuda-v10-2-cudnn-v7-6-5-installat...  
https://developer.nvidia.com/cuda-11....
https://developer.nvidia.com/

https://stackoverflow.com/questions/6...
https://github.com/pjreddie/darknet/i...
https://docs.nvidia.com/tao/tao-toolk...
https://towardsdatascience.com/yolov4...
https://github.com/aj-ames/YOLOv4-Ope...

nvidia-smi to show how GPU is used dynamically

https://deeplearning.lipingyang.org/2...

if this error comes while you have an old GPU integrated mention the GPU device number in yolov5 training as 0 in my case it's 0 rtx 5000 your case it depends on the GPU where pytorch cuda drivers properly installed only use when it shows the below error as device 0 and install python 3.7 and above hence yolov5 accept GPU pytorch above 3.7

RuntimeError: NCCL Error 1: unhandled cuda error

python train.py --img 640 --batch 16 --epochs 3 --data /home/amit/yolov5/data/Custom.yaml --cfg /home/amit/yolov5/models/yolov5s.yaml --weights '' --name yolov5s_results --cache --device 0

https://github.com/pytorch/pytorch/is...