Web giant Google has introduced new pricing tiers for its machine learning services that allow people to rent out partial nodes, saving money on jobs that don't require quite as much horsepower. These tiers start out with Standard, which grants your operation just over half of the power of one of Google's specialized nodes. This much power is sufficient for a large number of different types of basic machine learning operations, and can wind up being dirt cheap on this new plan. These nodes normally rent for $0.49 per hour, prorated by the minute with a 10-minute minimum. Using this new basic plan, 10 minutes of compute time will only cost you about a nickel, whereas you previously would have had to spend at least 9 cents to fulfill the minimum time requirement.
While this change may seem small at that scale, it can really add up; running a 24-hour compute job, say, for a cloud app that uses machine learning, would have cost $11.76 before, but now costs less than $7, nearly cutting the cost in half. When you bring the scale up to a daily cost of running an app or service, those savings become a very significant number over time. The list of incremental tiers includes the aforementioned standard tier, along with a large model tier that rents out 1.4111 nodes, a small complex model tier that rents out .845 nodes, a medium complex model that rents out 1.69 nodes, and a large complex model that rents out 3.38 nodes. The list goes up to include more specialized and expensive models. Batch prediction is available on top of your node rental for $0.09262 per node hour, while online prediction is available at $.30 per node hour.
This move comes just after the introduction of a cloud machine learning service called SageMaker by Amazon. The competing service offers many of the same things that Google's machine learning service offers, and while it's aimed primarily at existing Amazon Web Services customers, it could still pose a threat to Google. The new pricing tiers will help to make Google's option a bit more appealing and accessible, hopefully keeping current Google Cloud Platform customers happily in place, and perhaps even winning over some Amazon Web Services customers.