Phase 4 · Deep learning · Tooling
GPUs and compute
Why deep learning runs on graphics cards.
Open this lesson Create a free account
Why it matters
Every conversation about AI eventually becomes a conversation about compute: why a model needs a certain GPU, why training runs cost millions, why quantized models run on laptops, why inference prices fall. A few back-of-the-envelope formulas let you check vendor claims, size a fine-tuning job, or understand why a feature is expensive, without being an infrastructure engineer.
What you'll be able to do
- Explain why matrix multiplication dominates deep learning and why GPUs suit it
- Count the floating-point operations of a layer and of a whole training run
- Estimate the memory a model needs to run and to train, by numeric precision
- Turn compute estimates into GPU time and cost, the way AI budgets are planned
Before this lesson
Next in this phase: Capstone: train a classifier