I just signed up for the Paperspace $8/month plan. Wondering if anyone has any information regarding which machines are preferable over others for the purposes of this course. Is there some sort of comparison chart or something. I’m currently experimenting with the Tesla K80, and it seems to be slower than the free one I was using yesterday. As an example, is the GPU+ machine preferable to the K80? The GPU+ has more CPU’s and RAM, but perhaps the GPU is underpowered, and that is of more importance for what we are doing?
Here’s a fairly good ranking of both training and inference speed over a wide range of GPUs and a wide range of problem types.
Basically, scroll down whatever problem type is of interest to you (e.g. ResNet-V2-50 column, train subcolumn), and do a search on the GPU model on the left hand side.
You’ll find that the free P5000 machines are actually really not bad - half the training time of the K80 machines. …if they have capacity for free machines available when you want it, that is 