Hardware Requirements   

Since there are no meaningful minimum requirements for ANSYS, it is best to determine your needs and then select the most appropriate solution. Choosing the wrong hardware, however, can create an invisible drain on your productivity for the lifespan of the machine. This page is meant to be a resource for engineers looking to make informed hardware decisions.

Form Factor:

There are a broad range of possible hardware solutions and the form factors below are the general categories. Take a look at them and decide the best fit for your organization before diving into the more detailed information.

Workstation

Gorilla-server.svgA workstation is primarily intended for a single engineer and will generally offer the best user experience in terms of pre and postprocessing as long as the hardware is suitable to solve the models. This is also the only form factor where consumer grade CPUs are an option, making the price-point potentially extremely reasonable.

Typical Specs:

  • 1-2 CPUs, 4-12 physical cores each
  • 16GB RAM (minimum), 64GB+ recommended
  • 2TB+ hard drive space
  • Workstation graphics card
  • OS: Windows 7/10 or RedHat/SUSE Linux

Add-ons:

Server (Graphics)

170px-Server-graphicsA server form factor is a good way to provide simulation resources to multiple engineers. Engineers will generally remotely log into the server and perform pre and postprocessing, as well as the solve. This workflow is similar to the Workstation and concentrating hardware investment in a common resource makes financial sense. The downsides is that computational capacity and resources are compromised, to a degree, to serve graphics to users.

This form factor also lends itself well to remote (VPN) access because large results files can be postprocessed directly on the server without the need to send them through the internet.

Typical Specs:

  • 2-4 CPUs (>2 requires server OS), 4-16 physical cores each
  • 64-256GB+ RAM
  • 4TB+ of hard drive space
  • High end workstation graphics card
  • Server OS required if multiple simultaneous users or 3-4 CPUs

Add-ons:

  • OzenCloud for overflow compute capacity
  • Nvidia Tesla GPU
  • RAID0 SSDs + backup
  • Prep-Post license (to pre/postprocess during solution)
  • RSM (Users can submit jobs to machine without logging in)
  • NICE Desktop Cloud Visualization enhancement
Server (Headless)

170px-Server.svgA headless server is a specialized machine meant for the sole purpose of computation. The server form factor, as well as the removed need for graphics capability, allows for maximization of computational ability. This form factor generally requires workstations capable of pre and postprocessing models. While users can manually remote into the machine, copy files over and press solve, the setup and usage of Remote Solve Manager is highly recommended to automate this process. Network speed is an important consideration, especially for transferring large result files.

Typical Specs:

  • 2-4 CPUs, 4-16 physical cores each (>2 CPUs requires server OS)
  • 64-512GB+ RAM
  • 6TB+ hard drive space
  • Server OS if 3-4 CPUS

Add-ons:

  • OzenCloud for overflow compute capacity
  • Nvidia Tesla GPU
  • RAID0 SSDs + backup
  • Prep-Post license (to pre/postprocess during solution)
Cluster

170px-Server-multiple.svgFor organizations with computational needs that outstrip what a single machine can provide, clusters will be the only solution. In some cases clusters can also make financial sense when comparing two good machines to one machine with top-of-the-line everything. More so than any other form factor, this will require involvement of company IT from the beginning. Remote Solve Manager with a commercial queuing system is generally a must. Also in clusters, communication speed between the nodes are highly important and Infiniband is recommended for any case except two node CFD clusters.


Typical Specs (per node):

  • 1-2 CPUs (4-16 physical cores each)
  • 64-256GB+ RAM
  • 2TB+ Hard drive space
  • Server OS

Add-ons:

  • OzenCloud for overflow compute capacity
  • Nvidia Tesla GPU
  • RAID0 SSDs
  • Prep-Post license (to pre/postprocess during solution)

Want one of the machine types above? We have experience configuring every one of them for our customers as part of our world class support services. Click the button to contact us with your hardware needs.

Operating System

OS Platform Support | OS Platform Support – By Application

ANSYS products are supported on 64-bit operating systems. ANSYS Mechanical ANSYS Fluent Most of our customers successfully run ANSYS software on Windows 7 on Workstations. ANSYS R17.0 added support for Windows 10. See the links above for a complete overview of OS platforms supported.

A Server OS (Windows Server or Red Hat Linux/SUSE Enterprise Linux) will be required in the following circumstances:

  • More than 2 Physical CPU Sockets in a System
  • Multiple Machines Running in a Cluster
  • Simultaneous Users on a Machine (Remote or Local)

Processor

The latest 64-bit multi-core Intel Xeon and AMD processors with the highest clockspeed and core counts available are recommended. Hyper threading will not improve the speed of simulations, always evaluate the number of physical cores for ANSYS simulation. Always try to get the most recent architecture version of the CPU, even if the clockspeed or number of cores don’t seem to be improved. CPUs today are almost twice as fast as CPUs from 3 years ago listed at the same clockspeed.

Keep in mind that Windows 7/8/10 only support a maximum of two physical CPUs. For more than two physical CPUs, a Windows Server or Linux OS is needed.

Memory

A minimum of 16GB of memory is recommended. It is best to have as much memory as financially feasible. The actual memory required for a particular problem will depend on the mesh, physical models that are enabled, and domain complexity. As of 2016, 64GB of memory has been sufficient for 90% of the FEA and CFD projects completed by OEI engineers. EMAG products more often require more memory and 100+GB is recommended.

In terms of the effect of memory on performance, you either have enough or you don’t. If your operating system runs out of memory it will fall back to using the hard drive as ‘virtual’ memory, which will have a catastrophic effect on system performance.

To get an idea why this is, it is useful to consider how the CPU works. CPUs have an extremely small amount of memory that they can access immediately. We’ll call this the register. To access something not already in the register, the CPU will have to wait for the process to bring it into the register to complete before it can continue. There are several levels of memory in ascending size and descending performance that the CPU has access to. The cache levels are directly on the CPU itself and have various levels, termed L1, L2, etc… The system memory, or RAM, is modularly added to the motherboard, as is the hard drive. To compare the proportional speed of these memory levels, we can use the metaphor from this excellent article on the subject:

  • L1 Cache: Grabbing a piece of paper from your desk (3 seconds)
  • L2 Cache: Picking up a book from a nearby shelf (12 seconds)
  • System Memory: Taking a walk down the hall to buy a Twix Bar (4 minutes)
  • Accessing the Hard Drive: Leaving the building and roaming the earth for 1 year and 3 months

It doesn’t pay to pinch pennies on system memory!

Storage

A minimum of 1TB is recommended for the installation and use of your ANSYS software. The precise effect of storage on performance will depend on how I/O bound a particular analysis type is but it is uniformly better as model sizes get larger. Strongly consider one of the advanced storage recommendations below if your expected analysis type is one of the following where I/O is typically a bottleneck on performance:

I/O Bound Analysis Types:

  • Out of core Sparse Solver in Mechanical
  • Block Lanczos Eigensolver
  • Distributed Memory Parallel (DMP) solves (in SMP, there is one set of files, in DMP each core has its own set of files and IO becomes a bottleneck)
  • Transient FEA or CFD runs where many results are being written to disk

There are two main ways to increase storage performance:

RAID0: While there are many different RAID configurations that have trade-offs between speed, redundancy and efficient usage of space, RAID0 is the only configuration that should be considered for performance. Redundancy should only be considered for separate ‘storage’ drives or arrays, especially since RAID0 sacrifices redundancy the most for performance (if any of the drives in a RAID0 array fails, all the data is lost).

SSDs: Significantly more expensive on a per GB basis than mechanical hard drives but can have 2 orders of magnitude faster read performance and an order of magnitude faster write performance. Make sure to have a modern operating system with TRIM support or the write performance of the SSD will degrade over time.

The ideal storage setup would have several TB of general storage with a smaller, specialized drive used for solves. In ANSYS Mechanical R18, you can specify a Solver Scratch Directory to ensure that solutions are automatically performed on high performance drives but stored on general purpose storage.

SSDs + RAID0: This is the maximum possible performance configuration for storage but make sure TRIM is supported specifically for RAID0 with your chosen brand and operating system. The SSD alone supporting TRIM does not mean that it is supported in RAID0 arrays, which is a very recent development that often requires the latest operating system (eg Windows 10) and drivers.

Graphics Card

Supported Graphics Cards

For large assemblies it is recommended to make use of a graphics card to avoid display latency issues. A list of validated graphics cards can be found in the links above.

On board graphics will degrade the pre and postprocessing experience on the machine. Additionally, newer and graphically intensive applications such as AIM and SpaceClaim will not work without a discrete graphics card.

GPU

GPU Computing Resources | Supported GPU Cards

In an effort to provide faster performance during solution, various ANSYS products (ANSYS Mechanical & ANSYS Fluent) support offloading key solver computations onto graphics cards to accelerate those computations. All HPC license products (HPC, HPC Packs, and HPC Workgroups) enable GPU-accelerated computing and one GPU will count as one core. Note that not all CUDA enabled graphics cards are supported, this feature is intended for the high end NVIDIA Tesla and Intel Phi cards. It is recommended to get the card with the highest amount of memory. The supported cards for GPU computing can be found in the document linked above:

Notes about GPU Computing:

  • The NVIDIA Tesla solutions are more feature complete as of 2017. Not all analysis types are supported for GPU computing.
  • GPU computing is currently well suited for particular types of problems:
    • Mechanical: In-Core Sparse Solver runs with solid elements (vs shells) of 500k+ degrees of freedom but still able to be contained within the GPU memory (typically <8M DOF)
    • Mechanical: PCG/ICG Solver runs with the Level of Difficulty setting at a lower value and MSAVE off
    • Fluent (& Icepak): Single Phase, flow dominated, coupled solvers, model size > 3-4M elements

HPC

HPC Features Overview

To take full advantage of your computer hardware, make sure that you have the appropriate HPC licenses. There are flexible HPC, HPC Pack and HPC Workgroup options. HPC licenses are on a per-core basis. A single GPU is licensed as a single core. The HPC Packs add non-linearly to quickly get to a large amount of cores. HPC Parametric Packs allow parallelization of design point evaluations in ANSYS Workbench, essentially duplicating the solver and associated HPC licenses for a new kind of High Performance Computing. For crunching through all the different designs in a parametric space, the speedup is incredible. See the graph below for details. For the best results, submit to a Remote Solver Manager queue with multiple computers to automatically distribute the simulations.

HPC_Packs_Graphic

Example Desktop Workstation

The following specs are an example of a desktop workstation used by OEI’s engineers. Please keep in mind that these systems were configured in 2013 and still perform well using ANSYS R17.0. Other server hardware is available for more demanding tasks.

Chassis: Full size desktop workstation tower

Processor: Intel Xeon E5

Motherboard: Single processor workstation/server motherboard

Memory: 64GB RAM

Storage: 256GB SSD + 2TB HDD

Media: DVD+-RW

Operating System: Windows 7 Pro 64-bit 

For non-workstation systems feel free to contact us for guidance.

Additional Resources:

Understanding Hardware Selection to Speedup Your Mechanical Simulations

CAD Support

Remote Display Support

Interconnects Support

Job Schedulers Queuing Systems Support

To be informed about updates to this and other pages, sign up for our newsletter below:

Subscribe to Newsletter

  • This field is for validation purposes and should be left unchanged.