Savings Plans & Spot Instances

Recently, AWS has begun recommending that people transition away from Reserved Instances and instead use Savings Plans. However, in true AWS fashion, Savings Plans are not quite as straightforward as we might think. The good news is that they share many attributes with Reserved Instances.

For example, they can be purchased for a one-year or a three-year term. We can also choose to pay for them all upfront, partially upfront, or on a monthly basis. Presently, AWS offers EC2 Instance Savings Plans, Compute Savings Plans, and SageMaker Savings Plans. In essence, Savings Plans are a commitment to spend a certain dollar amount per hour for the agreed term. In comparison, Reserved Instances are more about a commitment to use a specific type of instance or instance family at a discounted price. This distinction will become more clear when we get into the details.

EC2 Savings Plan

EC2 Savings Plans provide a similar discount as Standard Reserved Instances and represent a commitment to spend a certain dollar amount on EC2 instances. Like Reserved Instances, EC2 Savings Plans are specific to a region. However, unlike Standard Reserved Instances, we do not have to specify an operating system or tenancy when we purchase a Savings Plan.

We can use any supported OS in any Availability Zone and any instance size within a family. For example, we can use an m6.xlarge one day and upgrade overnight to an m6.8xlarge. Because our discount is at the dollar level rather than the compute power level, we don’t need to worry about scaling factors. The savings are just a flat discount against the On-Demand prices.

Compute savings plan

Compute Savings Plans are more flexible than EC2 Savings Plans, but offer less of a discount. With Compute Savings Plans, we are not restricted to just EC2 instances. Our discount rate can be applied to AWS Fargate and AWS Lambda as well. Additionally, Compute Savings Plans apply to all regions, all EC2 instance types, all OS, and all tenancy models.

SageMaker Savings Plans

SageMaker Savings Plans are specifically for SageMaker workloads using the ML instance types. The plan can be used across all instance sizes, regions, and components, such as for training or inference.

Savings plans limitations

Savings plans do, however, have some limitations. It is up to us to fully use the amount of compute capacity that we contracted for, as we do not have the option to buy or sell unneeded capacity on a market like we can with Reserved Instances. Additionally, Savings Plans do not provide a capacity reservation. If we must be certain we will have the needed specific resources in a particular Availability Zone, we need to use On-Demand Capacity Reservations.

Normally, On-Demand Capacity Reservations are charged at the same rate as if we had running instances, but with a Savings Plan, the discount is automatically applied. Savings Plans do not apply to Spot Instances or other Reserved Instances, and these contracts cannot be canceled during their term.

The table below shows comparisons between the various Savings Plans and Reserved Instance plans.