Ansys Mechanical* price-performance on Amazon EC2 Hpc6id* instances
To achieve ambitious product design objectives and reduce the time to market, modern engineering projects require immediate access to innovative computing infrastructure, and advanced computer-aided engineering (CAE) software.
In this blog post, we demonstrate price and performance analysis of running Ansys Mechanical* workloads by leveraging Amazon Elastic Compute Cloud (EC2)* HPC-optimized instance powered by 3rd Gen Intel® Xeon® Scalable Processors. Ansys Mechanical is a finite element analysis (FEA) solver used for solving complex structural, thermal, acoustics, and nonlinear problems and make better, faster design decisions. It helps engineers optimize product design, reduce the need for physical prototypes, and enhance product performance and reliability.
Solution Overview
For the purpose of this blog post we perform FEA benchmarks using Ansys Mechanical on six models that are part of the standard Ansys Mechanical benchmarking suite. The analysis is performed by comparing the single-node simulation price and performance outputs between Amazon EC2 memory-and HPC-optimized instances, R5dn and Hpc6id, respectively.
Amazon EC2 Hpc6id instances are powered by 3rd Gen Intel® Xeon® Scalable Processors purpose-built for memory-bound and data-intensive tightly coupled HPC workloads that involve handling extensive data sets in the memory. Hpc6id instances offer up to 5 GB/s memory bandwidth per vCPU, 16 GB of memory per vCPU, and up to 15.2 TB local storage making it suitable for accelerating the FEA calculations.
Following are the instance types used for benchmarking Ansys Mechanical:
Benchmarking configuration:
Leveraging Intel MPI for Ansys Mechanical
The Intel® MPI Library is a software package that facilitates the development of parallel applications for high-performance computing (HPC) clusters. It provides a standard interface for inter-process communication (IPC) and message passing, which enables multiple processors to work together to solve complex computational problems. The library is designed to run on multi-node and multi-instances cloud cluster and provides a high-performance and flexible solution for building scalable and efficient applications that can take advantage of the latest advances from Intel and compatible processors. The Intel® MPI library is part of the Intel® oneAPI HPC Toolkit, which provides a comprehensive set of tools for high-performance computing, such as compilers, performance libraries, and tools for debugging and optimization.
Intel MPI is available as a module for clusters built using the AWS ParallelCluster and supports Amazon Machine Images (AMIs) for operating systems such as Amazon Linux 2*, Centos 7*, Ubuntu 1804*, and Ubuntu 2004*. To enable Intel MPI, the user needs to load the Intel MPI module using the following command:
$ module load intelmpiTo verify that Intel MPI is enabled, run the command:
$ mpirun --version
Intel(R) MPI Library for Linux* OS, Version 2019 Update 7 Build 20200312 (id: 5dc2dd3e9)
Copyright 2003-2020, Intel Corporation.More information can be found through link below:
https://docs.aws.amazon.com/parallelcluster/latest/ug/intelmpi.html
For performing benchmarking analysis, we collect the following metrics from each simulation:
1. Total Time: the total execution time of the workload
2. Core Solver Time: the time to run the solver
3. Cost per Simulation: total on-demand cost for the simulation
Testing Ansys Mechanical Performance
FEA simulations, especially for non-linear analysis, can be performed using either the iterative solver or the direct solver available in Ansys Mechanical. We run the standard benchmarks, based on iterative and direct solvers and we ensure that only one job at time is running on the instance under investigation. The results are presented as comparison on single-node for (1) total time in seconds (2) core solver time in seconds, and (3) cost per simulation.
1. Total simulation time (seconds)
Total simulation time is the time required in seconds to complete the entire simulation. The following table shows the total time for each benchmark on both Amazon EC2 instances. The performance comparison between the two instances is plotted in Figure 1.
As shown in Figure 1, we observe a performance improvement of 43% (geomean across the 6 benchmarks) of running Ansys Mechanical on Amazon EC2 Hpc6id instance type compared to R5dn instance.
1. Core Solver Time
Simulation speedup is the measure of simulation performance on an instance type compared to older instance type. In this case, speedup is calculated by comparing core solver time for simulations on Hpc6id and R5dn instances.
As shown in Figure 2, we observe a 46% speedup (geomean across the 6 benchmarks) of running Ansys Mechanical on Amazon EC2 Hpc6id instance type compared to R5dn.
1. Cost per Simulation ($)
While performance is a key metric, the cost to run a simulation is also a major consideration. Using recent pricing from the us-east-2 (Ohio) AWS Region, we have calculated the cost per simulation across the 6 benchmarks. As shown in Figure 3, we observe a 50% cost benefit (geomean across the 6 benchmarks) of running Ansys Mechanical on Amazon EC2 Hpc6id instance type compared to R5dn.
Note that the costs represented in Figure 3 correspond to Amazon EC2 cost only and does not include other common costs attributed to storage, network, and Ansys licensing costs.
1. Per core comparison
The Amazon EC2 instances used for this benchmarking consist of different number of physical cores. The R5dn instance has 48 physical cores and the Hpc6id instance has 64 physical cores. In order to understand the improvement from Hpc6id instance over R5dn instance, a normalized per core comparison is carried out by comparing the performance and cost characteristics on 48 physical cores for both instances.
As shown in Figure 4, on 48-core comparison, we observe 31% core solver speedup and 50% cost benefit (geomean across the 6 benchmarks), of running Ansys Mechanical on Amazon EC2 Hpc6id instance type compared to R5dn.
Conclusion
This blog highlights the impressive performance of Ansys Mechanical on Amazon EC2 Hpc6id, the HPC-optimized instance powered by 3rd Generation Intel Xeon Scalable processors and the Intel oneAPI MPI library. By using 6 different Ansys Mechanical benchmarks, we compare the results for two different instance types, Amazon EC2 R5dn and Hpc6id. We demonstrate up to 43% improvement in time to solution, up to 46% performance improvement in speedup on Hpc6id, and leading to a cost reduction of 45% by utilizing the latest HPC instance with Intel-based architecture. You can leverage AWS and Intel technologies to achieve faster, more efficient, and cost-effective solutions for HPC needs.
Authors
· Diego Bailon Humpert — AWS Global HPC and AI GTM Lead, Intel
· Dr. Fabio Baruffa — Technical Consulting Lead for Cloud and Quantum Computing, Intel
· Karthik Raman — Principal HPC Application Engineer, AWS
· Dnyanesh Digraskar — Principal HPC Partner Solutions Architect, AWS
Notices & Disclaimers
Performance varies by use, configuration, and other factors. Learn more on the Performance Index site.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software or service activation.
Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
