Hello there, how are you guys doing today?
As the title mentions I have quite the problem…
I have a linux laptop: i7 CPU / 2060 rtx GPU / 16gb RAM, running my own keras sequential model around 100 iterations per sec, and it takes around 30 minutes.
Have another windows desktop: i5 CPU / 3090 rtx / 16gb RAM, running same exact code at 40 iterations per sec, and it takes around 1 hour.
They are both running on GPU (rest assured I’m using device context), yet 3090 is so much slower.
I of course tested the CPUs to compare separately, and the i5 is slower than the i7 as expected.
Installation of tensorflow was done in both sides with the pip install guide:
Windows Machine:
1.- Download Microsoft Visual c++ dlls (directly without visual studio)
2.- Install miniconda
3.- Run
conda create --name tf python=3.9
4.- Run
conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
5.-
pip install --upgrade pip
6.-
pip install tensorflow
Linux Machine: (conda install linux version, works perfectly fine)
Windows runs just fine, and it detects the GPU on tf.config.physical_devices(‘GPU’), but it’s just so slow in comparison to 2060 rtx laptop somehow?
If you guys have any idea why this is happening, or how to fix it, I thank you in advance.
Have a good day!