CUDA out of memory - Githubissues

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.74k stars 7.96k forks source link

CUDA out of memory #2297

Open rrrtype opened 5 years ago

rrrtype commented 5 years ago

Hi , I train with my data set. If the image size used for training is large, will you get "CUDA out of memory" error?

AlexeyAB commented 5 years ago

Hi, No.

rrrtype commented 5 years ago

Thank you, I misunderstood! In my model tiny - yolov 3 and batch = subdivision = 24 it will consume memory immediately but batch = 64 subdivision = 16 will work fine. Is this normal behavior?

bit-scientist commented 5 years ago

@rrrtype batch = 64, subdivision = 16 is fairly good if you have about 12 GB of GPU memory, if not try to increase subdivision to 32

ydixon commented 5 years ago

Increase subdivision till you don't get the CUDA memory error.