If you have an iMac it’s super-extra likely that you have an AMD graphic card which makes your data science hobby/life harder than it should be when trying to train Deep Learning models.
There are three possible way out:
- Buy an eGPU Box and an NVidia card, this road is going to be very bumpy with occasional tornados. Like compiling every new release of Tensorflow in GPU mode.
- Boot on Linux, with an external drive, then using AMD ROCm
- Using Keras with PlaidML, an OpenCL compatible backend
As the third solution is by far the simplest and less intrusive option I am relying on this option as much as possible for most of my daily hobby/work.
For the cases that cannot be handled by this solution and for long training, I continue using Google Cloud as my main TPU/GPU Cloud.
On occasions, I don’t want to deal with all the mess of a platform that is regularly neglected by Apple when it comes to data science and in those cases, I fire up my corner dust collecting machine that hosts Ubuntu. If you ever need to install Ubuntu and set it up for Deep Learning work, start from this post. It will save you hours of search and hope.