In deep learning, the algorithm that is used, back propagation, is basically a chained matrix multiplication, which is good for the GPU structure. However, for

In deep learning, the algorithm that is used, back propagation, is basically a chained matrix multiplication, which is good for the GPU structure. However, for boosted trees the optimization problem is a little bit different, there is a sorting and then a computation of a gradient and hessian. In this case, it is not as efficient for GPU than deep learning.