Some machine learning or deep learning data sets can take a lot of CPU time, their doing a lot of matrix arithmetic operations. There is a lot of number crunching here.
If you want to write code for GPUs then I suggest you to try the following Udacity course. Introduction to Parallel Programming With CUDA
So the abbreviations are crazy, CPU stands for central processing unit, GPU means graphics processing unit. What is more crazy is that almost all computers have a CPU and a GPU. The CPU does almost all computing to run the various applications (programs) on the computer. The GPU runs the graphics card, but not all GPU’s are easy to use for math calculations beside its graphics.
Then Google is working on a variety of custom machine learning chips that are even faster.