The flexibility and heterogeneity of HPC cloud services provide a welcome contrast to the constraints of on-premises HPC. Every HPC configuration is potentially accessible to any given workload in a well-resourced cloud HPC deployment, with vast scalability to spin up as much compute as that workload demands in any given moment.
At the core of any HPC solution is flexible compute. With AWS, secure, resizable compute capacity in the cloud comes courtesy of the Amazon Elastic Compute Cloud (Amazon EC2). EC2 instances are optimized for a range of workloads. You can choose the CPU, GPU (or, if relevant, FPGA), memory, storage, and networking capacity that you need to run your applications.
For most AWS HPC customers, the Amazon EC2 C instances represents the best option for their workloads. “C” here represents CPU or compute-intensive workloads. C5 instances (the current generation) are powered by 3.0 GHz Intel® Xeon® Scalable (a.k.a. Skylake) processors and allow a single core to run up to 3.5 GHz with the Intel® Turbo Boost feature. Other powerful members of the C5 family include the C5.12xlarge and C5.24xlarge instances run on Intel’s Second Generation Xeon Scalable processors (code-named Cascade Lake) with sustained all-core turbo frequency of 3.6GHz and maximum single core turbo frequency of 3.9GHz. C5n instances use up to 100 Gbps of network bandwidth with 33% higher memory footprint compared to C5. C5n would be the preferred instance when network throughput and packet rate performance are a premium
The M family of AWS instances provide a balance of compute, memory and network resources — and are the option to select for workloads requiring more memory than available on C instance types. M5 has up to 48 physical Xeon cores (96 vCPUs) and 384 GiB of memory (a ratio of 8 GiB memory per physical core). Shared memory applications, for instance, or other tightly coupled workloads may find particular use for M5’s large number of cores. This enables, for instance, manufacturing organizations to run their Finite Element Analysis (FEA) and simulation jobs to completion more quickly. AWS instances in the R family represent memory intensive jobs. R5 instances have 1:8 vCPU to memory ratio, with the largest offering as much as 768 GiB per instance. These instances feature Intel’s Xeon Platinum 8000 series (Skylake-SP) with a sustained all-core Turbo CPU clock speed of up to 3.1 GHz.
Z instances provide high single-thread performance courtesy of a custom Intel Xeon Scalable processor (exclusive to AWS) that achieves up to 4 GHz sustained, all-turbo performance, enhanced networking and up to 25 GB throughput. Z1d instances work well with memory and computeintensive applications that require high-frequency processing. For instance, Z1d users leverage the sustained core performance for applications with license restrictions that require single-threaded applications and those whose software is licensed by the core. For all these instances, AWS and Intel also enable faster HPC app development and code modernization with a broad range of optimized software tools, frameworks, and libraries on AWS.
Here is a quick guide to choosing the right instance:
Read the full whitepaper: Achieving optimal price/performance for your HPC workload