What TrustRadius Research Says
Amazon Sagemaker Pricing 2022
Seeing into the future is probably the best superpower. In the corporate world, we already can do this by looking into past analytics to predict future outcomes. Thanks to machine learning (ML), we can do this a thousand times better than with a human brain.
Amazon Web Services offers Amazon SageMaker as a powerful, hotels and managed machine learning application that can build models to make those valuable predictions.
What Is Amazon SageMaker?
Amazon SageMaker is an AWS application for building and deploying machine learning models so experts can make predictions for their companies and clients. Machine learning is a field under the umbrella of Artificial Intelligence.
You create predictive models in machine learning by coding algorithms. SageMaker is designed for working with deep learning, a form of machine learning that is centered around predictive metrics.
Sagemaker is a service used by data scientists, business analysts, and ML engineers. It’s a full service from model training, deploying, scaling, and then interpreting by non-developers.
AWS SageMaker is most commonly used by tech, financial firms, and healthcare industries. Computer software and IT companies are the most avid user of the application which makes sense since software engineers are needed to code the algorithms.
In terms of the company size of their general customers, the answer is more spread out. The majority of users are from large companies with over 1000 employees. There’s still a sizable amount of teams that are under 200 employees that use SageMaker. Smaller companies also use the service.
The software is used by small to large companies across the board. This is likely because it’s another pay-per-use pricing model like the majority of AWS services. If you don’t want to hear about what SageMaker offers skip to the pricing section.
What Does SageMaker Offer?
AWS SageMaker has a lot to offer. They comply with 22 different programs to ensure data security. Their compliance includes HIPAA, ISO, and PCI. They have 10ms inference latency to provide a high-quality high connection and as much as 50% faster training with efficient GPUs. AWS SageMaker can offer as much as a 10x increase to your team's productivity of their daily workloads.
They make up 2.78% of the market compared to over 800 other service providers. This makes them one of the top Machine Learning model builders to choose from.
What Is The Price of SageMaker?
Amazon Sagemaker has two choices for prices: Free Tier and On-Demand pricing. Their lists are very detailed because they account for what you use and your usage. They account for things like instance types like compute instances and training instances.
We briefly talk about the pricing for each version of Sagemaker, but if you want to see an actual estimate for the total bill you should go to the pricing calculator section and the cost examples section.
Is Amazon SageMaker Free?
Amazon Sagemaker does have a Free Tier but it’s more like a free trial. It expires after 2 months instead of 12 like other AWS free service options. Note that there are a couple of AWS free service tiers that don’t expire after the first 12 months like Amazon CodeCommit.
The table below shows the usage amounts for each Free Tier capability that you are allowed each month, expiring after 2 months. Most of the usage is measured in hours or seconds. For request-based tasks like Feature Store, it's measured by every million requests, reported in units.
Studio notebooks, and on-demand notebook instances
250 hours of ml.t3.medium instance on Studio notebooks
250 hours of ml.t2 medium instance or ml.t3.medium instance on on-demand notebook instances
RStudio on SageMaker
250 hours of ml.t3.medium instance on RSession app
free ml.t3.medium instance for RStudioServerPro app
25 hours of ml.m5.4xlarge instance
10 million write units, 10 million read units, 25 GB storage
50 hours of m4.xlarge or m5.xlarge instances
125 hours of m4.xlarge or m5.xlarge instances
150,000 seconds of inference duration
750 hours/month for session time, and up to 10 model creation requests/month, each with up to 1 million cells/model creation request
You will still have access to all of these features after the Free Tier is over with one exception. Canvas will be available but its pricing is completely separate from the On-Demand Pricing services we go over below. The application has its own separate AWS info page you can find here.
Canvas costs are based on the number of hours used. Training data is priced for each number of cells per million cells.
Number of cells
First 10M cells
$30 per million cells
$1.9 per hour
Next 90M cells
$15 per million cells
Over 100M cells
$7 per million cells
Canvas is special in the sense that its a codeless, intuitive interface for data analysts. Business analysts will be able to build machine learning models with full datasets for accurate predictions.
Amazon Sagemaker On-Demand Pricing
The free choice above only has 8 different services. Sagemaker offers pricing lists for 11 different options. The new features in On-Demand include Asynchronous Inference, Batch Transform, and JumpStart.
We recommend going to Amazon SageMaker’s pricing page to look at their associated costs for different instances, memory, and vCPU. The cost is determined by the type of instance, memory, and vCPU you use per hour, billed monthly. Below is an image example of the pricing list for Sagemaker Studio Notebooks. You can go to the pricing page with this link here.
You should know that the costs will change for the AWS region you chose so it’s best to make sure it’s the right setting.
Only Feature Store and Serverless Inference are priced differently. Feature Store is based on per million read/write requests, data storage per GB a month, and data transfer per GB.
Serverless Inference is priced based on memory (MB) used per second, and data going in and out per GB.
Amazon SageMaker Studio Notebooks
RStudio on SageMaker
On-Demand Notebook Instances
Amazon SageMaker Processing
Amazon SageMaker Data Wrangler
Amazon SageMaker Feature Store
Amazon SageMaker Training
Amazon SageMaker Hosting: Real-Time Inference
Amazon SageMaker Asynchronous Inference:
Amazon SageMaker Batch Transform
Amazon SageMaker Serverless Inference
Amazon SageMaker JumpStart
If we’re all being honest, looking at intricate, mind-numbing pricing lists barely helps. You need examples with a total, not the damn cost per hour. We have a section where we go into Amazon’s pricing calculator to find that number.
AWS actually does provide examples for the cost of each feature and we talk about those next. They, unfortunately, excluded examples for Canvas, Edge, and Data Labeling. We talk about Data Labeling and Edge in a later section.
Amazon SageMaker’s Pricing Examples
SageMaker has pricing examples for most of its services. The examples are detailed and intuitive to read through but they do not include the monthly calculated total. Instead, it’s essentially a pricing scenario with all the math and costs mapped out, but no monthly calculations for the actual cost are made.
If you are someone that wants help on where to start your own estimate then these scenarios can be helpful to you. The example is very specific and shows the rate for a TensorFlow kernel, the notebook instances and hours cost.
For those that just want the estimated bill, these examples are going to be annoying and you should go straight to using the pricing calculator. In the next section is an image example of Studio Notebooks.
Other Noteworthy Info About Amazon SageMaker
One of the big benefits AWS offers is access to SageMaker Studio for free, mostly. It’s not completely free, you can use it for free but you do need to pay for the compute and storage. The cost for storage isn’t disclosed until your bill.
SageMaker Studio (don’t confuse this with Sagemaker Notebook Studio or RStudio) is an integrated development environment (IDE). It allows you full insight and control with model deployment.
You can also use SageMaker Studio in combination with other AWS applications like AWS SDK for Python (Boto3), and AWS command-line interface (CLI). Examples AWS highlights include:
SageMaker Pipelines to automate and manage ML workflows
SageMaker JumpStart to easily deploy ML solutions for many use cases.
SageMaker Experiments to organize and track your training jobs and versions
SageMaker Inference Recommender to get recommendations for the right endpoint configuration
SageMaker Debugger for debugging anomalies while training
Write and execute code in Jupyter notebooks
For more examples that can help you optimize the capabilities of your projects see the SageMaker pricing page here. As a reminder, using other AWS applications in combination will increase your usage and overall bill.
It's common to achieve real-time costs much higher than your estimates because of interactions with other services. For more information about Sagemaker Studio see their info page here.
SageMaker also offers the services Data Labeling and Edge. Their Data Labeling feature is designed as a managed tool to make the process of labeling training datasets easier. It can make your training jobs and overall workflow much easier. They have a separate info page you can find here. Edge is for improving your team's machine learning models across devices. You can find more info on their page here.
The hardest part of using services like Amazon SageMaker is controlling and budgeting for the cost. You can easily run into unexpected costs without noticing them. SageMaker has a Savings Plan service to help teams bring down their bills by as much as 64%. If you are interested you can learn more and find out if you’re eligible here.
Amazon Sagemaker in the AWS Pricing Calculator
The number is going to be big, let’s rip that bandaid off real quick.
We chose Sagemaker Studio Notebooks to do the estimate. The bright side is the input for the estimate is super easy and takes less than a minute.
We went with the preselected filed for AWS region, but it’s very import you check to make sure which region is yours because the costs will definitely be different.
We imputed 5 data scientists, all of which have 1 notebook instance they work on for 8hrs per day. We went with 29 days per month for how often the team would use the service. For your selected instance, the calculator will remind you the vCPU, memory, GPU, etc.
It will show you the end estimate at the bottom but won’t have an itemized breakdown. The total is $2,839.68 USD a month.
You will not get the full estimate until you click on the orange button labeled “Add to my estimate.” You can share this estimate and keep working on it if you generate a public URL. Once you have played with the estimate enough then it would be a good time to contact them and learn more about the service.
The estimated breakdown shows your yearly cost at over $34k a year. This estimate can easily be more dependent on your region, instance type, and team size. It could also be less if you are an especially small startup.
Ideally, it would be great to look at other options, especially ones that can be less expensive. There is another top competitor on the market.
What Is The Difference Between SageMaker and Azure Machine Learning?
Azure Machine Learning is a series of Microsoft products designed specifically for machine learning and deep learning. There is plenty to choose from like Azure Open Datasets and Machine Learning Studio.
Azure does not provide pricing lists like AWS does. We did a quick estimate with Machine Learning Studio (classic). We again did a team of 5 like with SageMaker. The process was fast, easy and even more intuitive than with AWS calculator. This does mean that there are fewer details provided in comparison.
In the estimate depicted above, we did 5 seats for a 5-person team. We estimated that each person worked on the experiment for 160hrs a month (40hrs each week).
The estimate shows you the formula. It's 5 seats x 9.99 for each seat which is 49.95. Then you need to calculate the hours of usage, we did 160hrs x 5 x $1.00 per month coming to 800.00. All together $849.95, excluding any taxes or fees. Its also important to remember that with cloud computing applications you can run into unexpected costs, especially when working with ML models and data processing.
Click here to explore all of our reviews for Machine Learning tools
For those that have used any of the platforms discussed here please leave a review to help other buyers make informed decisions.