Artificial IntelligenceCryptocurrencyMetaverseCybersecurityTech Forward

Google Has a More Flexible Way to Buy Its Cloud

November 18, 2015, 5:00 PM UTC
Web Summit 2014
DUBLIN, IRELAND - NOVEMBER 05: Urs Holzle Senior Vice President of Technical Infrastructure at Google, speaks on centre stage at the 4th annual Web Summit on November 5, 2014 in Dublin, Ireland. (Photo by Clodagh Kilcoyne/Getty Images)
Photograph by Clodagh Kilcoyne—Getty Images

Google on Wednesday announced a set of new custom cloud capabilities that should make it easier for customers to select the cloud resources that best meet their needs, while cutting down on a common problem—buying more capability than needed.

The new Custom Machine Types let developers select up to 32 virtual central processing units (called vCPUs), and allocate memory from 6.5 gigabytes per vCpU on up. The new model means that customers can move a job to another, more appropriate, vCPU configuration if the job changes, according to Google.

Customers can pay for these machine types by the minute or run them for a month at a time, adjusting the number of vCPUs they use as needed. In the latter case, if their job runs at 100% capacity, Google’s “sustained use discount” automatically kicks in, which can bring the price down significantly.

Amazon Web Services is the pioneer in public cloud—a model in which massive compute services are divvied up for use by many customers who typically pay by the hour. Amazon [fortune stock symbol=”AMZN”] offers a huge array of instance types and several pricing models. Google (GOOG) and Microsoft [fortune stock symbol=”MSFT”] are playing catchup in that field, investing billions in infrastructure.

In its blog post announcing the new machine types, Google quoted David Siuta, a Ph.D. candidate at the University of British Columbia who lauded the flexibility they afforded.

Siuta, who runs compute intensive weather prediction models, said it’s important for him to use the right type of virtual machines on Google Cloud Platform.

Per Siuta:

The number of cores per VM, and total number of VMs used, influence overall simulation time and total cost to run the model. The customized VMs have allowed us to zero in on the optimum number of cores per VM that yields the shortest time required to run the model. This has allowed us to capitalize on cost savings by knowing the best setup for our use case.”

Customers can assess cost in advance using the company’s pricing calculator.

Later on Wednesday, Urs Hölzle, Google’s senior vice president of technical infrastructure, will talk about how the cloud will look in 2020 at the Structure Conference.

For more coverage from Barb follow her on Twitter at @gigabarb, read her coverage at or subscribe via her RSS feed.

And please subscribe to Data Sheet, Fortune’s daily newsletter on the business of technology.