Deep Learning (DL) · Intermediate level

Intermediate level · Module 4 of 6 · 13 guided hours

GPUs and Efficient Training

Use GPUs effectively, speed up training and manage compute time, cost and experiments.

Learning objectives

  • Use GPUs and monitor their utilisation
  • Speed up training with mixed precision and sensible batch sizes
  • Plan experiments within compute and cost limits

Topics

  • GPUs vs CPUs
  • free and paid cloud GPU options
  • mixed precision
  • batch size and memory
  • data loading bottlenecks
  • experiment planning

Lessons

  1. Lesson 1 Using GPUs Effectively

    Understand why GPUs speed up deep learning and check that your training actually uses them well.

    parallel computationGPU memoryutilisation monitoringdata loading bottleneck

  2. Lesson 2 Mixed Precision and Batch Size

    Train faster and fit larger models with mixed precision, and choose batch sizes sensibly.

    16-bit vs 32-bit precisionautomatic mixed precisionbatch size and memorygradient accumulation

  3. Lesson 3 Planning Experiments within a Budget

    Plan and track experiments so limited compute time produces the most learning.

    experiment plancompute budgetearly stopping of runsexperiment tracking

Practical activity

Profile a training run, remove its bottlenecks, add mixed precision and plan a set of experiments that fits a fixed GPU budget.

Expected outcomes

  • Faster training
  • Realistic plans for limited compute

Assessment

Before-and-after training time report and an experiment plan within a budget.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.