Using CUDA unified memory (shared memory)

Hello, i have a question regards unified memory (the shared memory part).

I’m trying to train BERT and SQuAD 2 but i get on one of my computers a OOM error. But on the other computer not because the system is using cuda’s unified memory (so it has 24GB’s of memory instead of 4GB)…

How can i enabled cuda’s unified memory on the other computer? The only difference between the 2 systeems are the graphic cards (i’m experimenting for my school). The computer that produces the OOM error has a GTX 1060 graphics card and the other one has a Quadro M2000M.

Any possible idea’s?