Mar-19-2020, 01:44 PM
Hello,
I had a lot of troubles while trying to make work TensorFlow with GPU support.
I plan to reinstall all the necessary tools.
If my vision is correct, the tools necessary to run TensorFlow on a computer with NVIDIA card are the following:
TensorFlow must be last tool in this list to install.
And all of these tools (or at least some of them) will not work properly (or simply will not work at all) if during installation you are not careful about the versions of the corresponding tools.
So, before starting I want to clarify two questions:
Thanks
I had a lot of troubles while trying to make work TensorFlow with GPU support.
I plan to reinstall all the necessary tools.
If my vision is correct, the tools necessary to run TensorFlow on a computer with NVIDIA card are the following:
- TensorFlow itself
- CUDA
- CUDA dev. kit
- cuDNN
- NVIDIA drivers for graphical card
TensorFlow must be last tool in this list to install.
And all of these tools (or at least some of them) will not work properly (or simply will not work at all) if during installation you are not careful about the versions of the corresponding tools.
So, before starting I want to clarify two questions:
- Actually the last version of TF is 2.1.0. Is it compatible with last version of CUDA which is 10.2 ?
- In all manuals I've seen until now TF is installed using pip. Why not pip3 ?
Thanks