Home › GCP ML Engineer › cloudcomputing › mletraining › Which Google Cloud accelerator should an ML Engi…
Which Google Cloud accelerator should an ML Engineer choose for LARGE-SCALE training of foundation models requiring high inter-chip bandwidth — Google-designed silicon optimised for tensor operations?
ACloud Memorystore
BCPU-only training
CCloud DNS
DTensor Processing Units (TPUs)
Answer & Solution
Correct answer: D. Tensor Processing Units (TPUs)
TPUs are Google's custom ML accelerators (per PMLE §3.3). CPU-only is too slow; the others aren't accelerators.
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