I am using a Tesla K80 for deep learning purposes with tensorflow and it works just as expected. However, I was wondering if it would be possible to use the Tesla K80 as a single GPU with 24 GB VRAM unified memory instead of two 'separate' GPUs with 12 GB VRAM discrete memory as it is per default?
I think that it should be possible because, for example, in Blender there exists the option to do exactly that. My use case would be to do that in tensorflow for the learning process of deep neural nets. I already did some research and couldn't find any suggestions.
Is that possible? And if so, how would one achieve that?