Home › GCP ML Engineer › cloudcomputing › mletraining › 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.
Related questions
Which compute option should an ML Engineer choose for an INFERENCE workload on a low-powerWhich GCP service should an ML Engineer use to STORE training data (images, audio, video) Which Vertex AI capability should an ML Engineer use to train a model on data PARTITIONED Which Vertex AI feature should an ML Engineer use to deploy a Kubeflow-Pipelines DAG runniWhich interpretability requirement should an ML Engineer balance against accuracy when CHOWhich Vertex AI workflow should an ML Engineer use to build a TABULAR ML model with featurWhich file format should an ML Engineer prefer for HIGH-VOLUME training data stored on CloWhich Vertex AI / Model Garden capability should an ML Engineer use to FINE-TUNE a foundat