Page 1 of 1
[Downgrading Tensorflow] - Training fails w/ 3-GPUs
Posted: Mon Dec 02, 2019 1:22 am
by Chach_Pls
I recently added a 3rd GPU to my system and have been experiencing a general CUDA failure to load all 3 cards. All of my GPUs worked individually and in pairs, just not all 3.
How would I go about downgrading the tensorflow library manually (without having to rebuild from scratch)?
Re: [Downgrading Tensorflow] - Training fails w/ 3-GPUs
Posted: Mon Dec 02, 2019 1:41 am
by torzdf
I assume you are on Windows?
If so, and assuming that you kept your virtual environment name the default faceswap
Start > Anaconda Prompt
Code: Select all
conda activate faceswap
conda remove tensorflow
conda install tensorflow-gpu==1.13.1
cd faceswap
python update_deps.py
The last command is required because uninstalling Tensorflow also removes some libs that Faceswap depends on, so this puts them back.
If you are not on Windows, then these commands should just work from a standard terminal.
Re: [Downgrading Tensorflow] - Training fails w/ 3-GPUs
Posted: Thu Dec 12, 2019 12:29 pm
by BLACKPINK
I m running my gpu at base block and i have tried tensorflow 1.12 and 1.13.1. All doesnt work Hope theres a solution
Re: [Downgrading Tensorflow] - Training fails w/ 3-GPUs
Posted: Thu Dec 12, 2019 2:01 pm
by Khilos
What Keras, Tensorflow, and CUDA versions are you on? There was a similiar issue on the GitHub and user fixed some issues by downgrading to
CUDA 10.0
tensorflow==1.13.1
tensorflow-gpu==1.13.1
keras==2.3 .1
keras-gpu==2.3.1
As just downgrading to TF alone didn't solve the issue, but needed the combination above.