how to deal with `GCC` in tensorflow installation with gpu?

Question

I am installing tensorflow with Gpu support. There is an installation guide in the tensorflow web site which explain more about CUDA Toolkit . In this guide system requirement is described. A row in table 1 dedicate to GCC version and other ones is Kernel and etc. I know my ubuntu 16.4 LTS is based on kernel 4.4 but the GCC version dose not match with the table. My gcc is 5.4.0 and the table said that gcc-5.3.1 needed. My OS is Ubuntu 16.4 LTS xenial xerus and i have a Geforce GTX 1080 in my system. There is two question.

1- What should i do? Do i need to downgrade it or gcc-5.4.0 work as well?

2- If it is necessary to downgrade , could anyone explain how ?

3- Why gcc-5.4.0 compiler does not exist in /usr/bin ? how should refer to it ?

Answer 1

gcc-5.4 may work, but as it's not a supported configuration it's hard to say for sure. It may be worth giving it a try to see if you hit problems if you're reluctant to switch.

If you do want to switch, I like to use the update-alternatives approach to manage different versions of gcc on Ubuntu, as described here:

https://codeyarns.com/2015/02/26/how-to-switch-gcc-version-using-update-alternatives/