Skip to content
nodeza.learn
RoCE course / 3. Measurement / Checkpoint

Module 3 checkpoint

Produce and defend a number: run perftest and nccl-tests correctly, prove GPUDirect is actually on, and translate busbw into a statement about which piece of hardware is the bottleneck.

Module 3 of 5 in RoCE and RDMA for AI fabrics14 questionspass mark 80%about 12 minbelow the mark, retry after 24 h

Loading your checkpoint state…

What it covers

  1. 3.1What ib_write_bw actually measures
  2. 3.2NUMA, data validation, and doca_perftest
  3. 3.3GPUDirect RDMA: proving it is actually on
  4. 3.4NCCL environment variables on a RoCE fabric
  5. 3.5algbw, busbw, and reading a nccl-tests result