Amazon Web Services (AWS) today announced the general availability of Hpc8a instances, a new computing option optimized for high performance computing (HPC) workloads. The instances are powered by 5th Generation AMD EPYC processors and are designed to accelerate compute-intensive simulations and engineering tasks.
The Hpc8a instances deliver up to 40% higher performance and 42% greater memory bandwidth compared to previous generation Hpc7a instances, according to AWS. They also offer up to 25% better price-performance. The new instances feature 192 cores, 768 GiB of memory, and 300 Gbps Elastic Fabric Adapter (EFA) networking, enabling high levels of inter-node communication for large-scale applications.
AWS stated that Hpc8a instances are well-suited for applications such as computational fluid dynamics, simulations for faster design iterations, high-resolution weather modeling, and complex crash simulations. The instances utilize sixth-generation AWS Nitro cards, which offload CPU virtualization, storage, and networking functions to dedicated hardware, enhancing both performance and security.
The Hpc8a instances are currently available in a single 96xlarge size, offering a 1:4 core-to-memory ratio. AWS noted that customers can customize the number of cores needed at launch to align with specific workload requirements. The instances are available with both On-Demand Instances and Savings Plan purchasing options.
AWS recommends using Hpc8a instances with AWS ParallelCluster and AWS Parallel Computing Service (AWS PCS) to simplify workload submission and cluster creation. Amazon FSx for Lustre is also supported, providing sub-millisecond latencies and high throughput for storage. To optimize performance, Simultaneous Multithreading (SMT) is disabled on these instances.
The Hpc8a instances are initially available in the US East (Ohio) and Europe (Stockholm) AWS Regions. AWS directs users to the CloudFormation resources tab of AWS Capabilities by Region for information on regional availability and future rollout plans.
