Computers for LS-DYNA

Here you can find information on computers for LS-DYNA and how to install LS-DYNA. We cover everything from single user Windows Laptops to massive Linux Supercomputer systems.

Introduction

Here you will find information on computers for LS-DYNA and how to install LS-DYNA. We cover everything from single user Windows Laptops to massive Linux Supercomputer systems.

 

LS-DYNA runs efficiently on hardware ranging from laptops to super computers with more than 100 000 cores.

Windows Laptops and Workstations for LS-DYNA are very easy to set up, but make sure to pick the right hardware to get the best performance.

 

Setting up a larger computer cluster with several compute nodes can be more challenging and some customers in that situation prefer to outsource the set up or to buy are ready to use system from e.g. DYNAmore Nordic or other computer cluster providers.

Reduce the runtime with more cores

LS-DYNA runs much faster the more CPU-cores that are used. For explicit simulations often the speedup of the simulation is proportional to the number of used cores, so that doubling the cores used for the simulation leads to reducing the run time by a factor near 2.0. The image below shows the reduction in run time as more cores are used. The reduction of run time is close to the theoretical optimum - when increasing the number of cores from 64 to 192 the run time is reduced by a factor 2.92 which is very near the theoretical optimum 3.00.

 

Please note that there is a limit at which adding more cores will not reduce the run time any more. As a rule of thumb for explicit simulations this limit occurs about when the number of elements in the model divided by the number of used cores is less than about 10 000.  Very complex models (many contact definition, many parts et c) will have a larger deviation from the theoretically optimum speedup. The speedup is also affected by the used domain decomposition and other LS-DYNA solver parameters.

 

In summary it is well worth to evaluate how much time you can save by adding more cores to your LS-DYNA simulation. You are welcome to contact us at  support@dynamore.se for tips on how to speed up your simulations.

 

Scaling

Figure: Run time for the car2car crash benchmark (www.topcrunch.org) as a function of used CPU-cores for a Computer Cluster running MPP/LS-DYNA R7.1.3 with Platform MPI 9.1 on Dell PowerEdge computers each with 2  Intel Xeon E5-2697A v4 CPUs. The computer are connected by Mellanox EDR Infiniband.

Windows or Linux

For Workstations  bot Windows and Linux are popular choices to run LS-DYNA, though Windows is by far the most common. Today's workstations can have 92 cores or more, e.g. 2 x Intel Xeon AP 9282 CPUs, with more than 1 terabyte of memory and can thus run very demanding simulations.

 

For Computer Clusters, Linux is by far the most popular alternative though Windows Compute Clusters are also used. For really large Computer Clusters nearly all systems use Linux, see top500.org.  According to customer feedback, Linux Computer Clusters work very well with Windows Laptops and Workstations and provide an stable and easy to use work environment. Managing Linux Computer Clusters can be easier than managing other IT-systems as there usually are few changes or updates during the life-length of the Computer Cluster.

License and Software setup

A general information on set up of LS-DYNA and the needed license server can be found here:Overview of set up and installation of LS-DYNA and the License server.

Windows Workstation or Laptop

For smaller groups or single engineers using LS-DYNA it is common to use standalone Windows 10 laptops or workstations with local pre- and postprocessing. Set up is easy using the LSTC WinSuite installer that provides "everything": SMP/LS-DYNA, MPP/LS-DYNA, tutorials, manuals, LS-PrePost, and more. Read here about WinSuite.

 

To get the best performance out of LS-DYNA and the pre- and postprocessor make sure to use suitable hardware and graphics card. Popular choices of graphics cards and CPU and memory are given in the technical guide on Windows Workstations at the bottom of this page.

 

Share a Windows Workstation among several LS-DYNA users

For smaller groups of engineers using LS-DYNA, one option is to use personal laptops or basic workstations for pre- and postprocessing but share a more powerful workstation to run LS-DYNA. Read about sharing a workstation here .

Linux Computer Cluster - mobility

Linux Computer Clusters are a very popular option to run LS-DYNA. The clusters can have from 40 to 100 000 cores or more. Many today use hardware accelerated virtual desktop server(s), because then there is no need for workstations. Instead standard business laptops with integrated graphics and a separate display using a docking station are sufficient.  A bonus is increased security against data loss and the freedom to work and access LS-DYNA results from almost anywhere, both in- and outside* the office.

 

A technical guide Linux Computer Clusters for LS-DYNA is available at the bottom of this page.

 

Access from anywhere - Virtual desktop servers

Virtual desktop servers are  popular and with this solution the pre- and postprocessing is done on high performance servers. According to customer feedback this gives better performance, reduced cost, better security, and mobile access. A movie is available here showing the use of one of our own Virtual desktop servers which is based on the Thinlinc sofware from Cendio: Virtual desktop server demo.

 

Setting up and tuning MPP/LS-DYNA performance

Non-optimal configuration of hardware, operating system settings, network, queuing system, LS-DYNA executable, and MPI-parameters can easily lead to a performance reduction of 50% and in extreme situations to  90% or more. DYNAmore has guides that can help you detect performance issues and tune the performance. If problems persist please contact our support.

 

Guides on benmarks and tuning:

 

Hybrid/LS-DYNA for large models

There are three variants of LS-DYNA, SMP, MPP and the more recent Hybrid. Hybrid/LS-DYNA is a hybrid of the SMP and MPP methods for parallelization and thus uses both OpenMP and MPI technology. The Hybrid version scales better than the MPP-version for CPUs with many cores when running on many cores, typically more than 200 cores. It has been shown, see the image and reference below, that the Hybrid version can reduce the run time by 20% or more for car-crash simulations running on 200+ cores. Installation of Hybrid/LS-DYNA is straightforward but requires some modifications of the queuing system. Please contact DYNAmore Nordic support by e-mail or phone (support@dynamore.se)  for more information.

 

Hybrid LS-DYNA Scaling

Figure: MPP/LS-DYNA vs Hybrid/LS-DYNA for the car2car crash benchmark - Image from the reference below.

 

Reference: T-T Zhu, J. Wang, "LS-DYNA scalability analysis on Cray supercomputers", 13th International LS-DYNA Conference, Detroit, 2014

Buy a turnkey Linux Computer Cluster supporting mobility

DYNAmore Nordic offers turnkey Linux Computer Clusters including pre- and postprocessing on Thinlinc Virtual Desktop Servers with server side hardware OpenGL acceleration. Contact us for more information.

Set up a Windows Computer Cluster

Microsoft has a special HPC server package that can be used to build Windows Computer Clusters. Setting up a Windows Computer Cluster is in our experience straight forward for an IT-engineer. A technical guide is available at the bottom of this page.

Which CPU?

The CPU is the single most expensive piece of hardware in a computer with prices up to 3000€ or more per CPU. Selecting the right CPU is important. An overview is given below:

 

Intel Xeon is by far the most popular CPU for high performance computing including for LS-DYNA, see e.g. the statistics from Top500  (www.top500.org/statistics/list/). Starting 2017 there are also powerful AMD EPYC CPUs available which are also used for running mpp/LS-DYNA.

 

Popular recent Intel CPUs used in high performance workstations and computer clusters include:

  • Xeon 6144 (8 cores), 6146 (12 cores), and 6148 (20 cores). They were released in 2017 Q3. They each have 6 memory channels and can be used in single or dual CPU-computers.
  • Xeon 6244 (8 cores), 6246 (12 cores), and 6248 (20 cores). They were released in 2019 Q2. They each have 6 memory channels and can be used in single or dual CPU-computers.

 

Performance results from SPEC.org indicate that the new Xeon 6248 could be about 10% faster than its predecessor the Xeon 6148. Please note that for the above CPUs the AVX2 compiled version of LS-DYNA, which uses the special AVX2 vector instructions, can be significantly faster (i.e. > 10%) than the legacy SSE2 compiled version. From experience the “SPECrate 2017 Floating Point” benchmark performance rating is a good indicator of the relative performance of CPUs also when running mpp/LS-DYNA.

 

Recent AMD EPYC CPUs for high performance workstations and computer clusters include:

  • First gen. EPYC: AMD 7551 (32 cores) and 7451 (24 cores). They were released in the middle of 2017. They each have 8 memory channels and can be used in single or dual CPU-computers.
  • Second gen. EPYC: AMD 7542 (32 cores), 7502 (32 cores), 7402 (24 cores). They were released in the middle of 2019. They each have 8 memory channels and can be used in single or dual CPU-computers.

 

Regardless of which CPU that is used it is essential to populate all memory channels with high performance memory else performance will be reduced significantly.

Benchmarks and test/example models

LS-DYNA benchmark models to verify performance and example models useful for testing you computer or cluster are available here

  • LS-DYNA benchmarking site for computer clusters. www.topcrunch.org
  • LS-DYNA examples ranging from crash analysis to CFD: LS-DYNA Examples for R9.3 or later
  • LS-DYNA benchmark models with timing results from workstations and small clusters: LS-DYNA Benchmarks

Need more information?

You are most welcome to contact our computing experts listed to the right.

Referenced Technical Information and Guides

1 Overview of set up and installation of LS-DYNA and the License server

2 Information about LSTC WinSuite for Windows

3 Windows Computer Clusters for LS-DYNA

4 Linux Computer Clusters for LS-DYNA

5 Choosing LS-DYNA variant and CPU

6 Sharing a Windows Workstation for LS-DYNA

7 Movie: Virtual desktop server demo - Thinlinc based server from DYNAmore Nordic.

8 Set up of Intel MPI for LS-DYNA (To be published in 2019 Q3)

9 Suite of LS-DYNA benchmarks:LS-DYNA Benchmarks

10 Selection of LS-DYNA examples:  LS-DYNA Examples

Sales & technical contact

For more information please contact

contact Daniel Hilding
  • Director DYNAmore Nordic AB
contact Anders Jernberg
  • Licensing
  • LS-PrePost Development
  • LS-DYNA