RightScale Blog

Cloud Management Blog
RightScale 2018 State of the Cloud Report
Cloud Management Blog

Comparing Cloud Instance Pricing: AWS vs Azure vs Google vs IBM

As cloud computing matures, instance types from the top public cloud providers are becoming more similar, making it easier to compare cloud instance pricing. At the same time, cloud providers continue to reduce prices and add discounting options, creating opportunities to save money for the 85 percent of organizations that have a multi-cloud strategy.

In this article we dive down into the details to show you which providers have the lowest-cost options for cloud compute instances and under which circumstances. We also cover what’s been changing in cloud pricing over the last six to nine months.

What’s Changed in Cloud Instance Prices

The RightScale Cost Management Team continually monitors changes to cloud prices. There have been a large number of changes in pricing for cloud instances over the last six to nine months since our last comparison. Here is a quick rundown:

Many prices go down: 70 percent of the 104 price points we include in our comparison have gone down since our last comparison in April 2017. Although these comprise a fraction of the total price points, they represent some of the most commonly used instances. Of the 26 price points we analyzed for each cloud provider:

  • AWS dropped 19 of 26 prices
  • Azure dropped 24 of 26 prices
  • Google dropped 4 of 26 prices
  • IBM dropped 26 of 26 prices

New instances for our comparison:

  • AWS added the c5 instance family, which is the next generation for c4 instances.
  • Azure updated instances with Dv3, Ev3 and Fv2 families. Dv3 and Ev3 instances are hyper-threaded and although the price is lower, they offer less CPU power (as measured by Azure Compute Units) than the v2 families.
  • IBM moved from fully custom options to instance families.

Discounting options increase: 

  • Azure introduced Reserved Instances (RIs) with up to 72 percent savings.
  • AWS added 1-year Convertible Reserved Instances.
  • Google introduced 1-year and 3-year Committed Use Discounts (CUDs).

Per-second billing:

  • AWS moved from per-hour to per-second billing for EC2 and several other services.
  • Google has always provided per-second billing, but reduced the minimum time billed for an instance from 10 minutes down to 1 minute.
  • Azure offers per-second billing only on container instances.

Local disk pricing evolves: Cloud providers seem to be steering users away from a reliance on local disks for instances and instead pushing them toward attached storage:

  • AWS offers instance families with and without local storage. However AWS is shifting toward more instances with EBS only (no local storage). It is also reducing prices more quickly for instance families with no local storage. EBS optimization makes it possible to use attached storage for more/most use cases.
  • Azure continues to offer local storage on all instance families but has decreased the amount of local storage on the most recent generations.
  • Google continues to offer no local storage out of the box, making it an “optional add-on” for any VM type. Because the minimum is 375GB for local storage, the cost has historically been high but was reduced recently.

Understanding Cloud Discounting Options

To accurately compare cloud instance prices, you need to understand the discount methods for each of the cloud providers since that will be a major driver in the price you pay.

With the recent availability of Azure Reserved Instances, the big three cloud providers  AWS, Azure, and Google  all offer publicly available discounts (reaching as high as 75 percent) in exchange for committing to usage on the cloud provider for a one-year or three-year period. In all cases, you can decide how much usage to commit and how much to leave as on-demand. IBM only offers public discounts for monthly usage, which saves about 10 percent over on-demand usage.

Google also offers a Sustained Use Discount (SUD), which requires no commitment but offers an automatic discount on each instance type that is running in a region for more than 25 percent of a month. For instances running 24x7 for an entire month, the discount maxes out at 30 percent.

 

AWS

Azure

Google

IBM

Type of discount

Reserved Instances (RIs)

Reserved Instances (RIs)

Sustained Use Discount (SUD)

Committed Use Discount (CUD)

Monthly pricing

Length of commitment

1 or 3 years

1 or 3 years

 

SUD: No commitment

CUD: 1 or 3 years

Commit by month

Range of discount levels

Up to 75%

Up to 72%

SUD: Up to 30%

CUD: 37% (1Y) or 55% (3Y)

About 10%

Other documented programs

RI Volume discounts

Spot instances

Hybrid Use Discount

Enterprise Agreement

Preemptible VMs

Negotiated

 

In addition to the publicly documented discounts listed above, there are opportunities for privately negotiated discounts on a case-by-case basis. The more you are spending with a particular cloud provider, the more likely you are to negotiate further discounts.

Comparing Commitment-Based Discounts

The three programs for commitment-based discounts vary in flexibility as outlined below.

Ease of purchase: Google offers the simplest approach for purchasing, where your CUD can apply to any instance family and size in a region and there is no need to request any changes.

Ease of changes: With Google CUDs there is no need to request any changes — it automatically applies to any instance type in the region. AWS Convertible RIs and Azure RIs allow you to change your purchase (including the region), but require you to go through a manual process to do so.

Ability to cancel: Azure is unique in allowing you to cancel your RIs, but it is important to note that there is a 12 percent fee for doing so.

Most flexible payment options: AWS offers you the most payment options and enables you to save more the more you pay upfront.

 

AWS RIs

Azure RIs

Google CUD

Buy based on:

Term

Region

Instance family

OS

Network type

Term

Region

Instance type

 

 

Region

# of vCPUs

# of GBs RAM

Able to change?

Convertible: Can exchange for combo of RIs of equal or greater value

Standard: Can change Availability Zone, Size (for vanilla Linux only), Network type

Can exchange and apply remaining amount to the new purchase price

 

SUD: Automatically applies to all instances in the region

Able to cancel?

Can sell on Marketplace (often difficult to find buyers)

Yes, for a 12% cancellation fee

No

Payment options

3 options: No Upfront, Partial Upfront, All Upfront (providing increasing levels of discount)

All Upfront (can use EA commitment to pay)

No Upfront

Combine with other options?

Combine with RI volume discounts

Not combinable with EA instance pricing

Continue to get SUD only on instances not covered by CUD

The details on how each type of commitment-based discount works will be covered in a future article, but the most important thing to know is that these discounts don’t require you to change anything about your physical instances or VMs. Rather, you can think of them as “discount coupons” that can be applied to any instances that match the parameters of that “reservation” or “commitment.”

Deciding How Many RIs to Buy

It can be complex to decide your committed coverage level — how much of your footprint in a particular cloud provider should be covered by commitment-based discounts. In general, the answer is it's never a 100 percent coverage level (despite what your cloud provider sales rep might tell you), but it is also rarely 0 percent.

You need to consider a variety of factors when purchasing, including:

  • Historical usage (by region, instance family, etc.)
  • Steady-state usage vs. part-time usage
  • Future plans for:
    • Growth or decline in use
    • Changing cloud providers
    • Changing instance families
    • Moving regions
    • Shifting to other compute models (containers, serverless, etc.)
  • Balance between savings over time and cash payments upfront
  • Level of flexibility required

RightScale Optima is a cloud cost management solution that leverages a process that combines analytical reporting on historical usage with human intelligence about future plans for a tailored plan for commitment-based discounts. We help you decide on the appropriate coverage level and then turn that into a comprehensive Optima Discount Plan for purchasing RIs or CUDs as shown below. This plan can then be used to provide options to your IT managers and finance teams.

Comparing Instance Pricing

Now that you understand the different discount options, let’s compare both on-demand and discounted prices.

Important note: Keep in mind that our analysis of annual on-demand prices includes the Google SUD because it will automatically apply. Our analysis of discounted prices includes AWS and Azure RIs as well as Google CUDs. For IBM, we use monthly pricing and assume a 30 percent negotiated discount. As a result, your IBM prices could vary depending on your negotiated rate.

When comparing compute prices, it’s important to consider where the compared instances are similar and where they are not. In our analysis, we chose six scenarios to compare, as shown in the first column of the table below. We looked at standard, high memory, and high-CPU instance types with 2 vCPUs. For each of these three instance types, we looked at scenarios where a local SSD is required and where a local SSD is not required, resulting in a total of six scenarios. All scenarios are based on the lowest-priced region for each provider and, except for the Linux vs. Windows comparison, assume using one of the standard, free Linux distros that are available at no extra charge (not RHEL or SLES).

For each cloud provider, we then mapped the appropriate instance type for each scenario. You can quickly see that everything cannot be compared exactly “apples-to-apples”:

  • Notice that the amount of memory varies for the similar instances across the cloud providers. In most cases, the difference in memory is relatively small (10 percent or less), but for highcpu scenarios, AWS and Azure offer more than double the memory of Google and IBM.
  • AWS has instances families with and without the local SSD for standard (m3 and m4) and highcpu (c3 and c5) instance types. For highmem types, there is only the r3 family, which includes an SSD drive.
  • Azure always includes local SSD with all of its instance types, so as a result you are “paying for it” as part of the instance price whether you need it or not.
  • Google never includes local SSD with the instance type, so you need to pay for it as an add-on cost. The minimum size for an add-on local SSD is 375 GB, which is quite large. You can use Google Persistent Disk as an alternative, but it does not offer the same access times as the local SSD.
  • IBM usually includes a SAN drive, but there is an option for a local drive on the standard type of instance.

Comparing On-Demand Cloud Instance Pricing: AWS vs. Azure vs. Google vs. IBM

For each of the six scenarios below, you can see the hourly on-demand (OD) price for each cloud and then the hourly price per GB of RAM for each. We calculate both so that you can normalize the prices if the amount of memory is important to you.

In the chart below, red indicates the highest price of the cloud providers within a scenario while green represents the lowest price. If there are ties, then both cloud providers are highlighted green or red.


In pure on-demand prices for one hour:

  • Azure has the lowest price for 8 scenarios; highest price for 1.
    • Azure has the lowest price for all of the scenarios that include a local SSD and all comparisons based on per GB RAM.
    • Azure matches or is lower than AWS for all scenarios.
  • Google Cloud has the lowest price for 4 scenarios; highest price for 5.
    • Google Cloud tends to be the lowest price when no SSD is needed.
    • Google Cloud is higher priced on the “per GB RAM” cost for highcpu due to the fact that it includes less than half the memory of AWS and Azure.
  • AWS has the lowest price for 2 scenarios; highest price for 2.
    • AWS is most often a middle-priced option.
  • IBM has the lowest price for 1 scenario; highest price for 5.

Comparing Discounted Cloud Instance Pricing: AWS vs. Azure vs. Google vs. IBM

In comparing discounted prices, we looked at annual (instead of hourly) costs to give a better comparison. This is because the commitment-based discounts are for a minimum of one year.

Below for each of the six scenarios, you can see the discounted annual price for each cloud and then the discounted annual price per GB of RAM for each. We calculate both so that you can normalize the prices if the amount of memory is important to you.

In the chart below, red indicates the highest price of the cloud providers within a scenario while green represents the lowest price. If there are ties, then both cloud providers are highlighted green or red.


In annual prices for a one-year commitment:

  • Azure has the lowest price for 9 scenarios; highest price for 1.
    • Azure has the lowest price for all of the scenarios based on per GB RAM.
    • Azure matches or is lower than AWS for all scenarios.
  • Google Cloud has the lowest price for 2 scenarios; highest price for 7.
    • Google Cloud tends to be the lowest price when no SSD is needed.
    • Google Cloud is higher priced on the “per GB RAM” cost for highcpu due to the fact that it includes less than half the memory of AWS and Azure.
  • AWS has the lowest price for 0 scenarios; highest price for 2.
    • AWS is most often a middle-priced option.
  • IBM has the lowest price for 1 scenario; highest price for 4.
    • IBM pricing will depend on the negotiated price.

The Impact of Regions on Cloud Pricing

Different cloud regions have different pricing for each cloud provider. The most expensive regions in Asia Pacific can be 50 percent more than the lowest-priced regions in the U.S. For our analysis, we used the lowest-priced region that offered the instance types that we were comparing within each geographic area (U.S., Europe, and Asia).

On an hourly basis, the relative prices among cloud providers may change. In Asia, the price differences are especially pronounced.


On an annual basis, the built-in SUD discount makes Google the lowest-priced option for on-demand costs when no local disk is required. When commitment-based discounts are applied, Azure becomes the lowest-priced option in two of the three regions.


When a local disk is included on an annual on-demand basis, the built-in SUD discount still makes Google the lowest-priced option except in Europe, where Azure wins. When commitment-based discounts are applied, Azure becomes the lowest-priced option in all of the three regions.


We looked at options (U.S. only) that combined instances without a local disk with attached storage. Google has the lowest annual on-demand cost due to the SUD discount. IBM is the lowest priced of the annual discounted options.

Cloud Pricing for Linux vs. Windows

Each cloud provider charges a premium for instances running Windows in order to cover the licensing costs for Windows. The options below are all in the U.S. with no local disk. Google is still the lowest priced in the on-demand comparison due to the SUD. IBM is the lowest priced in the discounted comparison because it has a significantly lower premium for Windows than the other three providers.

Keep in mind that the Azure Hybrid Use Discount, which allows you to apply existing Windows licenses from on-prem to Azure, could bring that Azure price for Windows down. However, you will be forgoing the use of that license on-prem unless you have Datacenter Editions.

In looking at the premium paid for Windows, you can see that:

  • AWS and Azure have the same Windows premium of $806 annually.
  • Google has a slightly higher premium for Windows of $700 annually.
  • Interestingly, IBM has a far lower premium for Windows of only $303 for on-demand.

The Bottom Line on AWS vs Azure vs Google vs IBM Cloud Pricing

To get the best idea of overall pricing when comparing the top public cloud providers, we calculated the annual costs per GB/RAM across 22 different Linux scenarios. Azure came out the lowest for both on-demand and discounted costs. Keep in mind that your mileage will vary. For example, if you don’t use local disk, you may see Google as a lower-cost option. Also, if you use Windows, you may find that the Azure Hybrid Use Discount or the lower Windows premium on IBM may be a better option for you.

Total Cost per GB RAM: All Linux Scenarios

You Are Likely Wasting 35 Percent of Your Cloud Spend

Your costs on cloud instances are going to depend on your particular use case: the instance types you choose, the regions you run in, whether you need local disk for your instances, and your use of discounts.

Keep in mind that, you are also likely wasting 35 percent of your cloud spend. Focusing on optimizing your use of cloud can result in significant savings.

Lastly, a multi-cloud strategy is another critical component of your cloud cost management efforts. The ability to use more than one public cloud provider provides you with negotiating leverage and allows you to use cost as one of the factors you consider when you deploy to public cloud.

Free Cloud Comparison Tool from RightScale