QuestionQ117

VerifiedDesigning data processing systems

A TensorFlow machine learning model running on Compute Engine virtual machines (n2-standard-32) takes two days to finish training. The model uses custom TensorFlow operations that must execute partially on a CPU. You want to reduce the training time cost-effectively. What should you do?

Explanation

GPU-attached Compute Engine VMs can accelerate TensorFlow training while retaining CPU execution for custom or unsupported operations. Cloud TPUs are optimized for compatible, XLA-compiled matrix workloads and are not suited to neural-network workloads with custom operations in the main training loop; simply changing CPU VM families does not add accelerator compute.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!