Google Cloud Platform Adds GPUs And Expands Machine Learning

Google said it will make high-end graphics processing units (GPUs) available to users of its cloud platform from “early in 2017”, adding a feature already offered by competitors such as Amazon and Microsoft.

The search giant said it will also centralise its cloud machine learning operations and offer new and updated APIs for the service.

High-end processing power

GPUs are used to power highly complex tasks, such as machine learning training, financial analysis and scientific simulations, and the addition means users will be able to carry out such processes on remote Google systems rented by the minute.

Google said it will offer direct or “passthrough” access to AMD FirePro S9300 x2 and Nvidia Tesla P100 and K80 GPUs, giving bare-metal performance.

Users will be able to mix and match infrastructure, attaching up to 8 GPU dies to any non-shared-core machine, Google said.

The company didn’t disclose planned pricing, but said the feature will be priced per minute, as is the case with virtual machines.

Customers using an early access programme have been using GPUs on Google’s cloud for machine learning, seismic analysis, simulations and visualisation, the company said.

Machine learning

Google said it plans to centralise its cloud machine learning operations with a new group under researchers Fei-Fei Li and Jia Li.

“Building a centralised team within Google Cloud will accelerate our ability to deliver machine learning products and services to enterprise customers in every industry,” Google stated.

A new API for Google Cloud Machine Learning, called Clloud Jobs API, aims to apply the technique to help match job applicants with roles, Google said.

Currently available for limited alpha testing, the API is intended for use by job boards and career sites, and is being tested by customers Jibe, Dice and CareerBuilder.

Google said it has added improved logo and label detection to its Cloud Vision API and has reduced its price by about 80 percent, due to the use of customised chips to improve performance.

Neural Machine Translation

The company said its Google Neural Machine Translation (GNMT) system, a high-end automated translation tool that also runs on customised processors, is now available to those with premium Cloud Translation API accounts.

The new version of the tool, which is intended for long-form and highly detailed or technical translations, significantly reduces errors and offers support for English to Chinese, French, German, Japanese, Korean, Portuguese, Spanish and Turkish, with 16 language pairs.

Google said it has now made its Cloud Natural Language API text-analysis tool available to all businesses.

Following testing the service now features improved entity recognition, sentiment analysis and syntax analysis, Google said.

How much do you know about the cloud? Try our quiz!