I am considering using one of the big-3 cloud platform services - AWS/GCP/AZURE - for machine-learning projects of my team. In viewing the compute services for each of them, I faced the concept of vCPUs. For example, GCP provides the following solution:
- GPU : NVIDIA T4
- GPU memory : 16GB GDDR6
- vCPUS : 1~24
I understand the advantage of having more GPUs and GPU memories. But does it significantly matter to have more vCPUs for running deep learning algorithms on cloud platforms?
My interest is in convolutional neural network, clustering, and recommendations using deep learning.
CPUs can be useful for tasks like data preparation, but usually don't matter that much if your training runs on GPU.