Turi Create resource utilization

I'm looking into utilizing a cloud Linux machine to create and train my machine learning drawing classifier model with Turi Create. If I use my MacBook Pro it'll take at least 55 hours, so I want to offload that work and speed it up dramatically allowing me to iterate on the model to get the best result.

Now there's a lot of options for RAM, CPU, and GPU capabilities. What will Turi Create take advantage of? For example, if I run it on a machine with 96 CPU cores will it utilize them all and possibly speed it up to minutes rather than days, or would a better GPU be preferred? How much RAM would be good? I'll be training it to recognize 6500 classes with 20 images of each. Thanks!