torch.cuda.OutOfMemoryError: CUDA out of memory.

[chenzhizhuan]

GPU 1 NVIDIA GeForce RTX 3060 Laptop GPU Driver version: 30.0.15.1181 Driver Date: 2022/2/15 DirectX Version: 12 (FL 12.1) Physical location: PCI bus 1, device 0, function 0

Utilization rate 15% Dedicated GPU memory 2.1/6.0GB Shared GPU memory 0.1/11.9 GB GPU memory 2.1/17.9 GB

Internal memory 24.0 GB Speed: 2667 MHz Used slot: 2/2 Appearance specification: SODIMM Memory reserved for hardware: 177 MB

Available 6.9GB 3.3GB has been cached 25.8/53.8 GB has been committed Paging buffer pool 3.4GB Non-paging buffer pool 546 MB In use (compressed) 16.9GB (0 MB)

[chenzhizhuan]

Do the following alarms cause memory overflow?

Identity-Action\torchreid\metrics\rank.py:11: UserWarning: Cython evaluation (very fast so highly recommended) is unavailable, now use python evaluation.

[DuyNguDao]

Hi, You can Reset the device, Run it again. Because When you run realtime on Pycharm, Quit when stopping pycharm, the part of cuda that was run before it has not been reset, so when you run it next time it will overflow cuda, You should not use stop in pycharm, it will cause cuda or clt+C overflow on terminal

[chenzhizhuan]

32G video memory, I have also tried to reset the device, and it does not run in real time on pycharm. If the device is replaced, once the target is detected, cuda will be removed, and the problem remains the same.