GPU not really get used just RAM and CPU

[Schakuun]

I have started several training sessions and noticed that the system primarily uses RAM and CPU resources while the GPU is hardly utilized. Even when I adjusted the batch size to 64, which resulted in 90-95% RAM and CPU utilization, the GPU only runs at 15-30% usage. Is this intentional, or can I utilize the GPU more effectively?

Sorry if this question seems too basic.

For my setup, I am using:

OS: Windows 11
Env: Conda
GPU: RTX 3090 Trinity OC 24GB
CPU: i7-10700K
RAM: 32GB 2400MHz

[hypersniper05]

same here

*cuda 11.8

[Lordmau5]

In Task Manager in the GPU tab, make sure you select Cuda for one of the views

The 3D, Copy and Video Encode / Decode aren't being used for CUDA related things

Does that graph show usage when you're training?

(If you can't select Cuda in that or the other dropdowns make sure you have the CUDA toolkit installed and restart your PC afterwards)

[Schakuun]

Thanks for the information, unfortunaly it uses 0% of Cuda.
BatchSize is 32 as seen in config file.

Dataset:

• 770 AudioFiles

[Schakuun]

Found this after restarting the Process. It seems Conda Env cant use the GPU.

[Schakuun]

I have resolved the issue by reinstalling PyTorch-Cuda 11.8 in the conda environment "so-vits-svc-fork". The command I used was: "conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia".

After completing the installation, I was able to run "svc pre-hubert" again. The only missing dependencies were "chardet" and "cchardet", which I installed using the following commands: "pip install chardet" and "pip install cchardet".

After installing these dependencies, everything worked as expected, and the GPU was fully utilized.

[quantumbalance]

yo i am having this same issue:

INFO: GPU available: False, used: False
INFO [17:43:27] GPU available: False, used: False rank_zero.py:48
INFO: TPU available: False, using: 0 TPU cores
INFO [17:43:27] TPU available: False, using: 0 TPU cores rank_zero.py:48
INFO: IPU available: False, using: 0 IPUs
INFO [17:43:27] IPU available: False, using: 0 IPUs rank_zero.py:48
INFO: HPU available: False, using: 0 HPUs
INFO [17:43:27] HPU available: False, using: 0 HPUs

when i try to train. i have evga 3090ti and i9-13900k. miniconda. cuda 11.8. i tried the "conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia". "pip install chardet" and "pip install cchardet". commands and they installed, but still didnt fix the issue. please pleas please how do i fix this??

[Schakuun]

Before i did reintall the pytorch in the Conda ENV i installed "Cuda-toolkit",
Heres the link:
https://developer.nvidia.com/cuda-downloads?target_os=Windows&target_arch=x86_64&target_version=11&target_type=exe_local

Additionally i changed a Setting in the Windows System Setting go to: Settings -> System -> Display -> Graphics settings (down the bottom of page) -> Hardware-accelerated GPU scheduling -> Switch to "Off.

Hopefully that helps u.

[hypersniper05]

I reinstalled everything in the anaconda env and it works now :) thanks

[hypersniper05]

I reinstalled everything in the anaconda env and it works now :) thanks

[toroskuy]

did all of the steps you said.. still nothing

[toroskuy]

is cuda 12 supported?

[jiaxinf]

check out my solution here #779 (comment)

[hypersniper05]

Reinstall everything including the anaconda env. That's what did it for me

…

On Fri, May 12, 2023, 4:57 AM toroskuy ***@***.***> wrote: is cuda 12 supported? — Reply to this email directly, view it on GitHub <#372 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHEZ5V2OAJMFS5FKTNLTOE3XFX3PRANCNFSM6AAAAAAXBDO5OA> . You are receiving this because you commented.Message ID: ***@***.*** com>