Which description of Google's Tensor Processing Units, available within Google Cloud's AI infrastructure, is accurate?
Show answer and explanation
Answer: A
Tensor Processing Units are application-specific accelerators that Google designed for the matrix operations at the heart of machine learning. They are used for training and serving large models on Google Cloud. The other descriptions mislabel TPUs as general-purpose CPUs, a retired product, or a database, none of which is accurate.
Why the other options are wrong
- B. TPUs are not general-purpose processors and do not replace virtual machines. They are accelerators attached to workloads that need heavy matrix computation.
- C. TPUs remain part of Google Cloud's AI infrastructure and are offered to customers. They have not been retired or replaced by serverless compute.
- D. A distributed relational database is a different category of product. TPUs are hardware accelerators for machine learning math, not a database engine.