Driven by cloud economics, the AWS-designed Arm-based Graviton2 processor is leading an HPC performance revolution
Reminder: You can learn a lot from AWS HPC engineers by subscribing to the HPC Tech Short YouTube channel, and following the AWS HPC Blog channel.
Since the debut of the first ‘Beowulf’ cluster in 1994, HPC has been a race between technologists squeezing as much performance as possible from hardware, and scale economics driving mass-production prices to affordable levels. That first HPC cluster was an artful combination of commodity-grade computers networked together with ethernet and novel software techniques to allow distributed processing, and set the pattern for a generation. Today, the AWS-designed Arm-based Graviton2 processor is at the heart of similar revolution in HPC, but this time driven by cloud economics at an internet scale.
With the global scale of Amazon Web Services (AWS) and our Arm-based Graviton2 processors, AWS is helping democratize access to HPC, but leaving nothing on the table when it comes to performance. AWS-designed Graviton2-based instances offer up to 40% better price performance over comparable 5th gen x86-based Amazon Elastic Compute Cloud (EC2) instances. Beyond access to economical compute infrastructure, AWS offers a high-performance file system, and low-latency, high-bandwidth networking. Whether you are sequencing the human genome, running a 500M cell computational fluid dynamics (CFD) workload or analyzing billions of molecular compounds to create a vaccine, AWS offers the HPC tools and infrastructure you need to solve your hardest problems and get results sooner.
The Graviton2 C6gn instance has networking speeds up to 100 Gbps using Elastic Fabric Adapter (EFA). EFA is a network interface for Amazon EC2 instances that enables applications requiring high levels of inter-node communications. With EFA, HPC applications using the MPI and machine learning applications using NCCL can scale to thousands of CPUs or GPUs. EFA’s OS bypass mechanism provides a low-latency, low-jitter channel for inter-instance communications. By using the libfabric interface and libfabric APIs for communications, EFA makes it easy to migrate existing HPC applications to the cloud with few modifications. To learn more about what drove AWS to reinvent the network stack underlying HPC, check out this discussion from our HPC Tech Shorts channel.
Some of the most demanding HPC applications can benefit from the price performance of Graviton2 and the networking capabilities of EFA. For example, AWS shared recent benchmarking results of the C6gn with GROMACS, a popular open-source molecular dynamics application. We deep dive into GROMACS in our HPC blog, with a detailed overview of the AWS architecture used to run GROMACS and explain how we were able to achieve up to 40% better price performance by using C6gn. Weather forecasting simulations, which can be one of the most challenging HPC applications to run, are another example of where the C6gn exceled. We ran UCAR’s own model on thousands of cores with identical results, but achieved 37% lower costs on the C6gn (more in this HPC blog post and this HPC Tech Short video).
Whether you are running embarrassingly parallel workloads, or MPI-based, tightly-coupled applications that require high-levels of inter-instance communications, AWS has the HPC products and services to securely deliver the performance you need. To learn how you can use AWS infrastructure to run your largest, most complex simulations, subscribe to the HPC Tech Short YouTube channel, and follow the new AWS HPC Blog channel for the latest technical updates and discussion from the HPC engineers at AWS.