Google Colab

EE 641: A Computational Introduction to Deep Learning

Colab runs Jupyter notebooks on Google’s machines with an attached GPU. It is a good way to train models in this course.

Tiers

Tier Price Compute units / month Notes
Free $0 T4 when available, short sessions, no guarantees
Pro $9.99/mo 100 longer sessions, better GPU availability
Pro+ $49.99/mo 500 background execution up to 24 h
Pay As You Go $9.99 per 100 CU top-up, no subscription

Get Pro. 100 compute units per month covers everything in EE 641, including the final project. Unused units carry over for 90 days. Prices as of August 2026.

What a Compute Unit Buys

GPUs burn compute units at different rates: a T4 costs roughly 1.2–1.8 CU/hour, an A100 (40 GB) about 5.4 CU/hour. Pro’s 100 CU is therefore ~60+ T4-hours but only ~18 A100-hours. Use the T4 unless your model does not fit in its 16 GB.

Enabling the GPU

Runtime → Change runtime type → select a GPU. Verify:

import torch
print(torch.cuda.is_available(), torch.cuda.get_device_name(0))
!nvidia-smi

Sessions Die — Checkpoint to Drive

Colab disconnects idle sessions and enforces runtime limits. Anything on the session’s local disk is lost. Mount Drive and point checkpoints there:

from google.colab import drive
drive.mount('/content/drive')

CHECKPOINT_DIR = Path('/content/drive/MyDrive/ee641/checkpoints')
CHECKPOINT_DIR.mkdir(parents=True, exist_ok=True)

Stop the Meter

Compute units drain while a GPU runtime is connected, working or not. When done: Runtime → Disconnect and delete runtime. Check your balance under the runtime resources panel.