Amazon Web Services Pushes The Price Performance Envelope Again With Graviton3

By Patrick Moorhead - July 25, 2022

Amazon Elastic Compute Cloud (Amazon EC2) C7g instances supported by AWS Graviton3 processors have been available in preview since Amazon's annual re:Invent last year. Now generally available, it is an excellent time to dig into the details. 

The Six Five Summit (June 7-9, 2022) is a virtual conference on technology innovation led by myself Pat Moorhead (Moor Insights & Strategy), and DanIel Newman (Futurum Research). Last year, we featured a session with Dave Brown, VP of Amazon EC2, focusing on Amazon Web Services (AWS) silicon innovation, where we also announced the Graviton Challenge. We welcome Dave Brown to discuss AWS silicon innovation and the recent Graviton3/C7g GA announcement again this year.


The AWS Decoder Ring

If you are familiar with the AWS vernacular, skip this section. An Amazon instance is a virtual server in Amazon's Elastic Compute Cloud (EC2). There is a dizzying array of instances with different CPU, memory, storage, and networking resources available in various sizes to address specific workload requirements.

We can demonstrate the naming convention by breaking down the latest instance, "C7g". The "C" denotes an instance for compute-intensive workloads. The "7" indicates that this is the seventh generation of this family. The "g" refers to AWS Graviton.

AWS has over 500 instances with a broad choice of compute, memory, networking, and storage capabilities. These include instances powered by the latest generation Intel Ice Lake and AMD Milan processors and Habana Gaudi accelerators, and NVIDIA A10G Tensor Core GPUs

AWS has also launched new storage-optimized instances that feature the new AWS Nitro SSDs, custom-designed for storage performance for I/O intensive workloads running in Amazon EC2. 

And now, recently, the AWS Graviton3 processors and the seventh-generation of compute-optimized instances, the C7g instances powered by Graviton3. 

Graviton3 a giant leap forward

The first-generation Graviton processors previewed in 2018 contained 16 cores and 5 billion transistors. Graviton2 appeared in 2019 with 64 cores and 30 billion transistors. The latest Gravition3 processor has 64 cores and an incredible 55 billion transistors. Each new generation has been an enormous leap forward in performance, price performance, and the supported workloads. 

AWS claims the Graviton3 processors provide up to 25% better performance than Graviton2 processors with up to 2x higher floating-point performance, up to 2x faster cryptographic workload performance, and up to 3x better machine learning (ML) workload performance. 

Graviton3 processors also support the latest DDR5 memory, providing up to 50% more bandwidth than DDR4. Graviton3 processors are also highly energy-efficient, using up to 60% less energy for the same performance than comparable EC2 instances. 

Workloads that will benefit from C7g instances

C7g instances feature a 1:2 vCPU to memory ratio ideal for compute-intensive applications. vCPU is the abbreviation for virtual CPU, which shares the underlying physical CPU assigned to a virtual machine (VM).

C7g instances are well-suited for any application that requires more CPU power, higher floating-point performance, and better cryptographic performance. Applications that can take advantage of the faster memory bandwidth with DDR5 are also a good fit, including compute-intensive application servers and microservices, distributed analytics, ad serving, high-performance computing, machine learning, media encoding, and gaming. 

C7g instances come in eight sizes with 1, 2, 4, 8, 16, 32, 48, and 64 vCPUs. C7g instances support up to 128 GiB (gibibytes) of memory, 30 Gbps of network performance, and 20 Gbps of Amazon Elastic Block Store (EBS). C7g instances utilize the AWS Nitro System, dedicated hardware, and a lightweight hypervisor.

Customer feedback from the preview period

Hundreds of customers have tried out the C7g instances; here are some examples:

Twitter ran several benchmarks representative of workloads and found that C7g delivered 20%-80% better performance than Graviton2-based C6g instances. In addition, there was a reduction in tail latency by as much as 35%. Reducing tail latencies (or high-percentile latencies) makes users happy because if you guard against the worst-case response times, you improve the average response time.

Formula 1 ran Computational Fluid Dynamics (CFD) workloads on C7g and saw 40% better performance than C6g. CFD uses advanced mathematics and computer simulation to model and predict how the laws of physics and racing conditions will affect a race car's performance on race day. That is pretty much the essence of Formula 1 success.

Sprinklr observed 27% better workload performance. experienced a 35% performance improvement and a 30% reduction in latency compared to C6g for a telemetry ingestion workload. 

Developers have options to get started with Graviton-based instances

The Graviton3-based C7g instances are currently available in two of the most popular US AWS Regions and will be available in more regions in the coming months. 

Given that Graviton is Arm architecture, one must migrate applications from x86. Graviton3 instances are supported by choice of operating systems, ISVs, container services, agents, and developer tools, enabling migration with minimal effort.

Applications and scripts written in high-level programming languages such as Python, Node.js, Ruby, Java, or PHP will typically require redeployment. Applications written in lower-level programming languages such as C/C++, Rust, or Go will require a re-compilation.

In EC2, any developer can spin up a Graviton-based instance within minutes, including the latest C7g instance. There is a free trial on the Graviton2-based t4g.small instances for up to 750 hours per month.

Graviton-based instances in managed services such as AWS Lambda, AWS Fargate, and Amazon Aurora require little or no code change. 

Wrapping Up

AWS is committed to providing a choice of compute that best meets workload needs. AWS works with partners including Intel, AMD, and NVIDIA while also building custom silicon in-house.

AWS is innovating in silicon through the compute stack, starting from the Nitro System hypervisor to the Nitro offload cards and the newly introduced Nitro SSDs, all the way down to the Graviton processors and Inferentia and Trainium accelerators for deep learning.

As businesses bring more workloads to the cloud, AWS anticipates the need for cost-effective and high-performance infrastructure to rise. No doubt that AWS will continue to innovate to meet this need. 

Let me close with a shameless plug for the Six Five Summit, a three-day, 100% virtual, on-demand event designed to share new and relevant strategy, innovation, and thought leadership from the world's leading technology companies, including AWS. There, you can see Dave Brown’s full talk.

+ posts
Patrick founded the firm based on his real-world world technology experiences with the understanding of what he wasn’t getting from analysts and consultants. Ten years later, Patrick is ranked #1 among technology industry analysts in terms of “power” (ARInsights)  in “press citations” (Apollo Research). Moorhead is a contributor at Forbes and frequently appears on CNBC. He is a broad-based analyst covering a wide variety of topics including the cloud, enterprise SaaS, collaboration, client computing, and semiconductors. He has 30 years of experience including 15 years of executive experience at high tech companies (NCR, AT&T, Compaq, now HP, and AMD) leading strategy, product management, product marketing, and corporate marketing, including three industry board appointments.