Real-ESRGAN
Real-ESRGAN copied to clipboard
CUDA out of Memory, finetuning RealESRGAN_4X
2022-08-06 19:37:37,657 INFO: Loss [PerceptualLoss] is created. 2022-08-06 19:37:37,684 INFO: Loss [GANLoss] is created. 2022-08-06 19:37:37,718 INFO: Model [RealESRGANModel] is created. 2022-08-06 19:37:38,038 INFO: Start training from epoch: 0, iter: 0 Traceback (most recent call last):
RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 6.00 GiB total capacity; 5.27 GiB already allocated; 0 bytes free; 5.29 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Do I really need minimun of 20GB GPU Ram to get started. ? Thanks
lower batch size in your configuration file to 1 and num worker gpus to 2
hey im having this issue as well where can I find the configuration file
hey im having this issue as well where can I find the configuration file
In the Real-Esrgan folder go to the options directory, and you should see an fine-tune Real-Esrgan.yml file, that's it
Hey so I just changed the settings now am I supposed to restart my device or should it work right off the bat because im still getting the same issue
Yup,Lmk if it works
Restarted my device and im still having the same issue
Says I should lower the —tile but ive zero clue where it is or where to do it
What GPU are you using, what is the resolution of your dataset
Rtx 3080 16gb not sure by what you mean of resolution of dataset however
Rtx 3080 16gb not sure by what you mean of resolution of dataset however
The dimensions of the pictures in your dataset, like 512 x 512
Idk 😭
Wait i can check that nvm lmfao
:p
2048x2048
Also is there a reason why it allocates 13gbs of vram lmfao
It's to do with torch batch memory growth, it isn't well handled as compared to tensorflow, maybe give me a chat over at discord, username: clidegamer254#2034
Let's see if we can resolve that issue :p
Added right now
On Mon, Sep 12, 2022 at 01:35 Cliff Njoroge @.***> wrote:
Let's see if we can resolve that issue :p
— Reply to this email directly, view it on GitHub https://github.com/xinntao/Real-ESRGAN/issues/403#issuecomment-1243281462, or unsubscribe https://github.com/notifications/unsubscribe-auth/APFAGPYMKYNX53OI6LFGFLLV53FLPANCNFSM55ZCQQLA . You are receiving this because you commented.Message ID: @.***>