Announcing general availability of GPUs for Cloud Machine Learning Engine

Today, we’re excited to announce that Cloud GPUs are generally available for use with Cloud Machine Learning Engine. Back in February, we announced the availability of NVIDIA K80 GPUs in beta for both Google Compute Engine and Cloud Machine Learning Engine. Our customers have seen significant performance improvements using these GPUs for machine learning.

• Airbus saw a 40X speedup using GPUs over CPUs to train a deep learning model to detect clouds in satellite imagery. It was also able to simultaneously reduce training time from 50 hours to 30 minutes using distributed training, and decrease the error rate from 11% to 3%.

• Global Fishing Watch uses deep learning to detect overfishing, monitoring 140 million square miles of ocean. They similarly saw a 10X speed increase in training by using GPUs over CPUs.

We look forward to hearing how Cloud GPUs will help you scale and speed up your machine learning jobs on Cloud Machine Learning Engine. Learn more here about how to use Cloud GPUs on Google’s Cloud Machine Learning Engine.