With the cloud computing market projected to reach a value of $1,240.9 billion by the end of 2027, you can be sure that there are some powerhouses in the market gunning to dominate. AWS is one of these, and over the last few years have developed AWS Graviton to help them do just that.
But what exactly is AWS Graviton, and how will it help Amazon to get its share of the more than $1 trillion pie?
Well, that’s what we’re here to elaborate on today. We’ll take you through everything from the basics of what AWS Graviton is to how it’s enhancing AWS and, in particular, the EC2 instance offerings.
We’ll cover:
- What is AWS Graviton?
- EC2 instances that use AWS Graviton
- Take the pain out of your AWS financials
Let’s get started.
What is AWS Graviton?
“AWS Graviton” is the name of a series of server processors that AWS designed and released back in 2018, based on Arm architecture. The aim was to create processors which could offer minimal latency while also having high price performance and scalability, which they largely succeeded in achieving.
Graviton processors typically offer up to seven times higher performance than other EC2 instance options, partially due to their having quadruple as many computing cores, caches of roughly twice the size, and five times as much memory to boot.
AWS Graviton processors were successful and profitable enough to warrant further development, leading to the release of Graviton2 (the second generation) in late 2019 and Graviton3 (the third generation) in mid-2022.
The Graviton2 family is a general improvement on their predecessors, offering vastly improved performance and capabilities, and are still utilized to offer the best price performance for workloads in EC2. The instances using these are wide-ranging in purpose, meaning that almost anyone will be able to utilize the power of Graviton2 processors no matter their niche. Their universality also means that they’re some of the more popular choices, especially where price performance is a major concern.
Gravitron3 processors, being the most recent iteration, are a little more specialized than Graviton2. While offering better performance in some aspects (such as compute, higher floating-point, and cryptographic workload performance), these are mostly utilized to power machine learning setups.
Before we get too bogged down in the specifics, let’s take a look at which EC2 instances are powered by AWS Graviton processors.
EC2 instances that use AWS Graviton
To make things simple to follow, we’ve split this section according to what type of EC2 instance is being discussed. As in, we’ll cover the general purpose instances running on AWS Graviton first, then move our way through the categories to eventually finish with the accelerated computing ones.
General purpose
There are two general purpose EC2 instance types powered by AWS Graviton processors, being:
- M7g
- T4g
M7g instances are designed with price performance as a priority, balancing compute, memory, and networking to prove useful in a wide range of situations. M7g uses AWS Graviton3 processors.
As stated in our discussion of the EC2 instance types as a whole, general purpose instances don’t have a set use case for which they shine - they’re the jack-of-all-trades instances that are a great choice for when you aren’t sure what specifications your application will have. The M series of instances as a whole is a little more expensive than the A series, but also provide more computing power.
M7g instance specs fall into a range between the following extremes:
- M7g.medium = 1 vCPU, 4 GB memory, EBS-only instance storage, up to 12.5 Gb/s network bandwidth, and up to 10 Gb/s EBS bandwidth
- M7g.metal = 64 vCPU, 256 GB memory, EBS-only instance storage, 30 Gb/s network bandwidth, and 20 Gb/s EBS bandwidth
T4g instances are similar to other T instances, in that they’re great for workloads where CPU usage is generally low (less than 20%) other than the occasional spike. This is because T instances build up “credits” while you’re below 20% utilization, then those credits are used up when you go beyond that limit. If your utilization doesn’t drop before your credits run out, your performance will be throttled. T4g instances run on AWS Graviton2 processors.
T4g instance specs range between the following:
- T4g.nano = 2 vCPU, 0.5 GB memory, 5% baseline performance per vCPU, 6 CPU credits earned per hour, up to 5 Gb/s network burst bandwidth, and up to 2,085 Mb/s EBS burst bandwidth
- T4g.2xlarge = 8 vCPU, 32 GB memory, 40% baseline performance per vCPU, 192 CPU credits earned per hour, up to 5 Gb/s network burst bandwidth, and up to 2,780 Mb/s EBS burst bandwidth
Compute optimized
The following compute optimized instances run on AWS Graviton processors:
- C6g, C6gd, C6gn
- C7g, C7gn
Compute optimized instances differ from general purpose ones not only in their intended usage, but also in terms of what the different instances offer from each other. These instances are designed for high-performance use cases such as batch processing workloads, media and video encoding, and machine learning inference.
Also, while certain general purpose families are best utilized for vastly different use cases, all compute optimized instances differ mainly in terms of their pricing.
Standard non-Graviton C4 instances are less powerful than C5s, or even their Graviton-powered C6 and C7 cousins, but they’re also markedly cheaper. This means that the main limiting factor on which instance to use from this type is that of how much you’re willing to spend, and whether or not you need the better performance of the more expensive instance families.
Having said that, beat for beat the C6 instances offer up to 40% better price performance than C5 instances, and C7 offers up to 25% better performance than C6. C6g, C6gd, and C6gn all run on AWS Graviton2 processors, while C7g and C7gn run on AWS Graviton3 and AWS Graviton3E, respectively.
To round out, here are the minimum and maximum specs of each compute optimized instance powered by AWS Graviton.
C6g:
- C6g.medium = 1 vCPU, 2 GB memory, EBS-only instance storage, up to 10 Gb/s network bandwidth, and up to 4,750 Mb/s EBS bandwidth
- C6g.metal = 64 vCPU, 128 GB memory, EBS-only instance storage, up to 25 Gb/s network bandwidth, and up to 19,000 Mb/s EBS bandwidth
C6gd:
- C6gd.medium = 1 vCPU, 2 GB memory, 1 x 59 GB NVMe SSD instance storage, up to 10 Gb/s network bandwidth, and up to 4,750 Mb/s EBS bandwidth
- C6gd.metal = 64 vCPU, 128 GB memory, 2 x 1900 GB NVMe SSD instance storage, up to 25 Gb/s network bandwidth, and up to 19,000 Mb/s EBS bandwidth
C6gn:
- C6gn.medium = 1 vCPU, 2 GB memory, EBS-only instance storage, up to 16 Gb/s network bandwidth, and up to 9.5 Gb/s EBS bandwidth
- C6gn.16xlarge = 64 vCPU, 128 GB memory, EBS-only instance storage, up to 100 Gb/s network bandwidth, and up to 38 Gb/s EBS bandwidth
C7g:
- C7g.medium = 1 vCPU, 2 GB memory, EBS-only instance storage, up to 12.5 Gb/s network bandwidth, and up to 10 Gb/s EBS bandwidth
- C7g.metal = 64 vCPU, 128 GB memory, EBS-only instance storage, up to 30 Gb/s network bandwidth, and up to 20 Gb/s EBS bandwidth
C7gn instances do not currently have their stats publicly available.
Memory optimized
Here are the memory optimized EC2 instances that are powered by AWS Graviton:
- R7g
- X2gd
Memory optimized instances offer (surprise, surprise) a larger memory than most others, and are ideal for processing large data sets in memory while maintaining fast performance. This could be a great fit if you’re running a HPC application, running in-memory applications or databases, and so on.
R instances are all great for running services such as Redis, RabbitMQ or ElasticCache, as long as you aren’t overly concerned with storage persistence and aren’t a fan of the AWS serverless alternatives. The specific instances (R4 vs R5, etc) are similar to the C instance family, as each subsequent number will be more expensive but offers a more powerful setup - if you need the power the larger numbers such as R7g are better performing for their price, but otherwise you should stick to the cheaper, less powerful options. R7g instances run on AWS Graviton3 processors.
Here are the minimum and maximum specs of R7g instances:
- R7g.medium = 1 vCPU, 8 GB memory, EBS-only instance storage, up to 12.5 Gb/s network bandwidth, and up to 10 Gb/s EBS bandwidth
- R7g.metal = 64 vCPU, 512 GB memory, EBS-only instance storage, up to 30 Gb/s network bandwidth, and up to 20 Gb/s EBS bandwidth
X2gd instances, along with the rest of the X family, are suited to fully in-memory applications because they offer a higher memory ratio than the other instances. Basically, they’re great for open-source databases, real-time caching servers, real-time analytics, and so on. X2gd instances run on AWS Graviton2 processors.
The minimum and maximum specs of X2gd instances are:
- X2gd.medium = 1 vCPU, 16 GB memory, 1 x 59 GB NVMe SSD instance storage, up to 10 Gb/s network bandwidth, and up to 4.75 Gb/s EBS bandwidth
- X2gd.metal = 64 vCPU, 1,024 GB memory, 2 x 1,900 GB NVMe SSD instance storage, 25 Gb/s network bandwidth, and 19 Gb/s EBS bandwidth
Storage optimized
The storage optimized instances that utilize AWS Graviton are:
- Im4gn
- Is4gen
Storage optimized instances specialize in dealing with workloads that require access to large data sets on local storage, such as MPP data warehousing or NoSQL databases.
Both Im4gn and Is4gen instances utilize Non-Volatile Memory Express (NVMe) SSD-based storage to optimize for I/O heavy workloads in particular, making them an especially good choice for data warehousing or analytics. The difference between the two is mostly that of price versus performance, with Is4gen offering the lowest cost per TB of SSD storage across the whole of EC2, and Im4gn providing excellent price performance for its higher specs. Both Im4gn and Is4gen instances run on AWS Graviton2 processors.
The minimum and maximum specifications of both instance types are as follows.
Im4gn:
- Im4gn.large = 2 vCPU, 8 GB memory, 1 x 937 GB NVMe SSD instance storage, up to 25 Gb/s network bandwidth, and up to 9.5 Gb/s EBS bandwidth
- Im4gn.16xlarge = 64 vCPU, 256 GB memory, 4 x 7,500 GB NVMe SSD instance storage, 100 Gb/s network bandwidth, and 38 Gb/s EBS bandwidth
Is4gen:
- Is4gen.medium = 1 vCPU, 6 GB memory, 1 x 937 GB NVMe SSD instance storage, up to 25 Gb/s network bandwidth, and up to 9.5 Gb/s EBS bandwidth
- Is4gen.8xlarge = 32 vCPU, 192 GB memory, 4 x 7,500 GB NVMe SSD instance storage, 50 Gb/s network bandwidth, and 19 Gb/s EBS bandwidth
Accelerated computing
Finally, we have the accelerated computing instances that use AWS Graviton processors. These are:
- G5g
Accelerated computing instances are the powerhouses of EC2, utilizing hardware accelerators to provide you with sustained high performance. The G family of instances in particular are suited to graphics-intensive workloads, with G5g specializing in graphics applications such as Android game streaming and ML inference. G5g runs on AWS Graviton2 processors
The minimum and maximum specs of G5g instances are:
- G5g.xlarge = 1 NVIDIA T4G Tensor Core GPU, 16 GB GPU memory, 4 vCPUs, 8 GB memory, up to 3.5 Gb/s EBS bandwidth, up to 10 Gb/s network bandwidth, 0.42 On Demand price per hour, 0.252 1-yr ISP Effective Hourly (Linux), and 0.168 3-yr ISP Effective Hourly (Linux)
- G5g.metal = 2 NVIDIA T4G Tensor Core GPU, 32 GB GPU memory, 64 vCPUs, 128 GB memory, 19 Gb/s EBS bandwidth, 25 Gb/s network bandwidth, 2.744 On Demand price per hour, 1.646 1-yr ISP Effective Hourly (Linux), and 1.098 3-yr ISP Effective Hourly (Linux)
Remove the pain of tracking your EC2 bills
Once you’ve decided which AWS Graviton-powered instance is best for your needs, you’re left dealing with the most frustrating aspect of AWS; the financial side.
AWS bills are a nightmare to consolidate, track, manage, and especially optimize, making it easy to pay for things that you’re not using or missing out on savings that you could make according to your needs and usage.
That’s why Aimably is here.
Our AWS Invoice Management Software lets you effortlessly consolidate, track, and manage your AWS bills in one easy-to-use location. Whether you’re attempting to find discrepancies between your invoices and actual usage, accurately report your AWS spending, or simply handle invoice processing with your financial team, we’ve got you covered.
Once you’re certain of your current bills, we also have an AWS Cost Reduction Assessment which will take all of the effort out of optimizing your setup. Whether you’re running G5g.metal instances or the most humble M7g.medium, we take a look at your entire AWS account to make expert suggestions on how to reduce your bill without affecting your business’ performance.
No matter what, if you’re using AWS, Aimably can get rid of your nightmares.
