The company calls its new feature Custom Machine Types, and it allows users to essentially build their own compute instance to fit an application's needs. Instances can have between 1 and 32 vCPUs, incremented by even numbers of processors. To that, developers can add up to 6.5GiB of memory per vCPU in an instance (1GiB, or gibabyte, is equivalent to approximately 1.074GB, or gigabytes.)
Using that feature means that users can set up a custom machine with 12 vCPUs and 45GiB of memory that sits in between two existing instance offerings that Google already has available.
Being able to create custom instance types lets developers create a configuration that works best for their applications, without paying for extra capabilities. For example, Google estimates that the custom instance laid out above would cost $321.75 a month, compared to $408.80 per month for its standard counterpart with more capacity.
Right now, custom instances can only run Debian, CentOS, CoreOS, OpenSUSE and Ubuntu, though Google Product Manager Sami Iqram said in a blog post that the company will support additional operating systems in the future.
Letting users set up their own instance types is a powerful pricing move from Google, since its major competitors in the cloud market only let users pick from a limited menu of different instance types. Rather than building their ideal configuration, users have to pick one of the preset ones that works right for them.
Making a move like this makes sense for Google, since the company lags behind its key competition from Amazon and Microsoft. It could help attract price-conscious users who have to worry about keeping their cloud budgets fully under control.
Wednesday's announcement isn't the first time that Google has pushed user-friendly pricing for its cloud products in an attempt to woo users away from its competitors. The company also offers sustained usage pricing that gives users discounts for running a virtual machine for more than 25 percent of a month. Google also charges for compute resources by the minute, which means that as soon as users stop running a workload, they stop getting charged.