OpenRadioss is a finite element solver for dynamic event analysis. It simulates how materials interact and respond to outside influences. This can help engineers model events like car crashes, a bridge straining under a heavy load, or even a phone dropping on the floor. As a high-performance computing (HPC) workload, OpenRadioss requires powerful CPUs, large amounts of memory, high-speed interconnects, and fast storage.
In a study sponsored by Intel, Prowess Consulting ran OpenRadioss on Amazon Web Services® (AWS®) instances powered by both Intel® Xeon® processors and AWS Graviton® processors—both compute-optimized C6 and C7 generations. We chose OpenRadioss for this analysis because it has similar infrastructure requirements to data analytics, AI, and other HPC workloads, and our findings can be extrapolated to these workloads. We tested two OpenRadioss models—the Chrysler® Neon 1M model and the Ford® Taurus 10M model—both of which are popular among engineers.
Our approach was to identify which AWS instances offer the best value. We know development teams must balance the project-management variables of schedule, cost, and quality. We focused on the tradeoffs between schedule and cost by measuring performance and performance per dollar.
Oftentimes, instances are chosen based on core count and hourly instance pricing. We took a different approach, including the cost of engineering time. We also explored instances with different vCPU values, in addition to configuring instances with 64 physical cores.
We measured performance as the number of simulations that an engineer could run in one eight-hour day, and we called this variable “simulation throughput.” Prowess Consulting found that engineers can achieve up to 92% higher simulation throughput by choosing the AWS C6i instances [C6i.32xlarge] with Intel shown in Table 1, as compared to the AWS C6g instances [C6g.16xlarge] with Graviton2. For this specific comparison, we set the 64-vCPU C6i Intel instance OpenMP® thread to one to match the Graviton2 processor, which is single-threaded. With this configuration, the 64-vCPU C6i instance operates with 64 physical cores, which matches the physical core count of Graviton2. We used our research findings to create a checklist for maximizing project value.