Practice free →
HomeGCP ML Engineercloudcomputingmletraining › Which Vertex AI distributed-training feature sho…

Which Vertex AI distributed-training feature should an ML Engineer use to AGGREGATE gradients efficiently across many GPU workers — reducing all-reduce communication overhead in large-scale training?

AReduction Server on Vertex AI (or Horovod for similar all-reduce)
BCloud Memorystore
CSingle-worker training only
DCloud DNS
Answer & Solution
Correct answer: A. Reduction Server on Vertex AI (or Horovod for similar all-reduce)
Reduction Server + Horovod accelerate distributed all-reduce (per PMLE §3.3). Single-worker doesn't scale; the others aren't distributed-training tools.
Solve this in the app — GCP ML Engineer practice & 24k+ MCQs →
Related questions