Accelerating Ansys LS-DYNA on Azure using HB-series VMs

This post has been republished via RSS; it originally appeared at: Microsoft Tech Community - Latest Blogs - .

Authored by: Shuichi Gojuki, Cloud Solution Architect, Aimee Garcia, PM AI Benchmarking, Jon Shelley, Principal TPM Manager

 

Ansys LS-DYNA enables users to simulate complex real-world problems. LS-DYNA is an industry leader in explicit simulation software that is used in industries such as automotive, aerospace, bioengineering, military, and construction. It has the capability to accurately predict automotive collision behavior which includes the impacts collisions have on occupants. These real-world simulations allow automotive companies to significantly reduce the number of experimental design car prototypes, which in turn reduces time and expenses.

 

Azure HB-series offerings are enabling our Ansys LS-DYNA customers to accelerate their innovation by using the latest AMD CPU and NVIDIA InfiniBand offering. The Azure HB-series offerings are enabling our Ansys LS-DYNA customers to accelerate their innovation by using the latest AMD CPU and NVIDIA InfiniBand offering. The HB-series virtual machines (VMs) are designed to deliver leadership-class performance, scalability, and cost efficiency for various real-world HPC workloads. These VMs feature NVIDIA EDR and HDR InfiniBand backend network (100 to 200Gb/s depending on VM type).

 

Screenshot 2023-01-31 153749.jpg

Table 1: Configuration of each HB-series VM benchmarked.

 

Performance Benchmarking and Simulation Models

 

We used the standard , a common, a set of automotive crash simulation models that have been used for years to measure LS-DYNA  performance, to compare the various H-series VM offerings. The software configuration for LS-DYNA is 2020 R1 (ls-dyna_mpp_s_R13_0_0_x64_centos610_ifort190_avx2_openmpi4.0.0) These benchmarks enabled us to compare performance across the HB-series VM offerings using various VM offerings sizes of benchmarks in an objective way. The 3 TopCrunch benchmark models we tested can be seen in the table below.

 

Screenshot 2023-01-31 160812.jpg

Table 2: Description of the 3 TopCrunch benchmark models used.

 

Below are images of the collision simulations generated by each TopCrunch model. The 3cars model simulates a van rear-ending a compact car which causes the compact car to then rear-end a midsize car. In the Caravan2m-ver10 model, simulates an angled two vehicle collision, based on the National Crash Analysis., is simulated. The ODB-10M benchmark is a simulation of a vehicle colliding into an offset deformable barrier.

 

Screenshot 2023-01-31 163327.jpg

Figure 1: Visual representation of the 3 TopCrunch benchmark models used.

 

Performance comparison for real-world HPC workloads

 

Below are benchmark results for the HB, HBv2 and HBv3-series VM offerings across different LS-DYNA simulations. Each model is representative of an automotive collision, with different element sizes and simulation lengths. For each benchmark model, we scaled up to a maximum of 4 VMs and measured the elapsed time to run each benchmark. Using the elapsed time, we calculated the cost[1] per solution. We then calculated the relative cost compared to what was achieved on HB-series VMs for a given number of VMs to understand the relative cost per solution when compared to HB VMs (3–4-year-old hardware).

 

Graph1.jpg

Figure 2: Performance and Relative Cost comparison of HB, HBv2 and HBv3 series VMs on the 3cars model.

[1] Learn more about the pricing of each HB-series VM.

 

In the 3cars benchmark model (794K elements), the HBv3-series VMs reduced the time-to-solution by an average of ~69% while reducing the cost per solution by 42-56% compared to HB-series VMs.

 

graph2.jpg

Figure 3: Performance and Relative Cost comparison of HB, HBv2 and HBv3-series VMs on the Caravan2m-ver10 model.

 

In the Caravan2m-ver10 benchmark model (2.4 million elements), the HBv3-series VMs reduced the time-to-solution by an average of ~64% while reducing the cost per solution by 40-45% compared to HB-series VMs.

 

graph3.jpg

Figure 4: Performance and Relative Cost comparison of HB, HBv2 and HBv3 series VMs on the ODB-10M model.

 

In the ODB-10M benchmark model (~10.6 million elements), the HBv3-series VMs reduced the time-to-solution by an average of ~58% while reducing the cost per solution by 30-37% compared to HB-series VMs.

 

Continuous innovation with cutting edge capabilities

 

With Azure’s HB-series VMs, customers can reduce the time to solution and cost per each LS-DYNA simulation. When compared to 3–4-year-old technology (HB-series), the various benchmarks ranging from 1 to 4 VMs (60 – 480 cores), HBv3-series VMs provide the fastest time-to-solution at the lowest relative cost per LS-DYNA simulation. These performance gains on the HBv3 are due to additional memory bandwidth, 3x increase of L3 cache, and the 200Gb/s HDR InfiniBand. The benchmarks show HBv3-series VMs deliver up to 75% reduction in run times while reducing the over cost per solution as much as 40% compared to HB-series VMs.

 

To learn how Azure can accelerate your LS-DYNA workloads we invite you to check out the links below and contact your Microsoft account manager.

 

Additional Information

 

 

 

 

 

 

 

Leave a Reply

Your email address will not be published. Required fields are marked *

*

This site uses Akismet to reduce spam. Learn how your comment data is processed.