Cloud GPUs for Training

EE 641: A Computational Introduction to Deep Learning

Homework runs on a laptop CPU or a small GPU. Serious training — the final project in particular — generally does not. This page lists the options.

Google Colab (recommended)

Notebooks on Google’s machines with an attached GPU. Colab Pro ($9.99/mo) covers everything in this course. See the Colab guide for tiers, GPU selection, and checkpointing to Drive.

Hourly GPU Rental

Providers such as Lambda, RunPod, and Vast.ai rent GPU machines by the hour. You get a full Linux machine over SSH: more control and often cheaper per GPU-hour than Colab at scale, but you manage the environment, data transfer, and shutdown yourself. Worth it only if your project outgrows Colab.

Your Own Hardware

A recent NVIDIA gaming GPU handles most of the course. Apple Silicon works through PyTorch’s mps backend for small models; expect missing ops and slower training on anything large.

Whatever You Use: Checkpoint

Sessions disconnect, spot instances vanish, and meters run while you sleep. Save best and last checkpoints every epoch to storage that outlives the machine, and stop the machine when the run ends.