That's normally due to the cost of copying the data to GPU space. Here, they've reformulated the training algorithm to be a less-computationally-complex problem

That's normally due to the cost of copying the data to GPU space. Here, they've reformulated the training algorithm to be a less-computationally-complex problem. It's a huge difference.