Skip to content

Module 07: Distributed Systems

Prerequisites

Outcomes

  • Derive collective semantics and first-order communication costs.
  • Match data, tensor, pipeline, and sharded parallelism to constraints.
  • Design observability and recovery for distributed correctness and performance.

Ordered free resources

  1. PyTorch Distributed Documentation (pytorch-distributed) — learn process groups, collectives, and training APIs.
  2. NCCL Documentation (nccl-docs) — inspect GPU collective semantics and topology concerns.
  3. Deep Learning Systems Course (dlsyscourse) — connect communication with computation graphs.
  4. Full Stack Deep Learning (fsdl) — place distributed training inside operational practice.

Checkpoints

  • Complete Distributed All-Reduce Reasoning.
  • Produce a communication-volume estimate for two parallelism strategies.
  • Write a triage tree for hang, out-of-memory, and divergence failures.

Self-tests

  1. Why is all-reduce not equivalent to gathering every full tensor on one rank?
  2. Which topology details can invalidate a simple bandwidth estimate?
  3. Where can communication overlap with useful computation?

Capstone

Design a multi-GPU plan for a model that does not fit on one device. Include tensor placement, collectives, topology assumptions, memory and communication budgets, failure recovery, and a small-scale validation experiment.

Next: Evaluation

Explore connectionsGraph and backlinks