What hardware do you recommend for COMSOL Multiphysics?

Solution Number: 866
Title: What hardware do you recommend for COMSOL Multiphysics?
Platform: All Platforms
Applies to: All Products
Versions: All versions
Categories: Installation
Keywords: hardware recommendation recommends recommended recommendations

Problem Description

I am going to buy a new dedicated computer for running COMSOL Multiphysics. What hardware do you recommend? What about CPU speed and amount of RAM?


With regards to good COMSOL performance, the single most important factor is that you have enough physical RAM memory. If the amount of memory needed to solve the problem exceeds the available physical RAM you will notice a marked decrease in solution speed, so it is important that you have enough available physical RAM memory in your computer. Predicting RAM requirements can be done by solving smaller models that contain the same physics that you want to solve. Monitor the memory used and the degrees of freedom from these smaller models to extrapolate the memory requirements for your larger models. Be aware that memory usage versus degrees of freedom can be very different between combinations of physics, so you will want to repeat this procedure for every type of model you wish to solve.

After you have decided how much RAM you need, your will want to select optimal processors. COMSOL solvers use all available processor cores in parallel by default in shared-memory processing. See details in knowledgebase solution 1096. Therefore, CPU properties that affect the parallelization efficiency, like memory bandwidth and number of memory channels, are important for good COMSOL performance. In general those numbers are more important than the CPU clock rate. For example, for a four-socket system with 16 memory channels (four channels per CPU socket), 16 is the maximum possible speedup for the solver algorithms, regardless of the number of CPU cores available. In practice, depending on physics and solver settings, the maximum measured speedup will be lower than 16, typically somewhere between 8 and 12. Adding even more CPU sockets is not expected to lead to further speedup, as maximum attainable speedup is limited by the non-parallel fraction of the algorithms (for further details see Amdahl's law).

Please note that the recommendations below assume that you plan to solve COMSOL models of significant size - for smaller models, performance (mainly solution time) will not be significantly improved by adding more processors and/or more RAM. For small models the CPU clock rate is still important.

In this solution we will first discuss which architecture parameters to look at. Then we will translate that into specifications you typically will find in vendor price lists and quotes and give some general recipes how to select hardware.

Parameters that Affect Performance

  • A 64-bit operating system is highly recommended.
  • For a given hardware, the choice between Windows, Linux, or Mac OS X will not significantly affect performance.
  • At least 4-8 GB physical RAM per core of the computer is recommended.
  • Dual-socket nodes are recommended.
  • The number of CPU memory channels should be as high as possible.
  • A CPU with as fast a memory bus as possible is beneficial. The numbers are often measured in GigaTransfers/second (GT/s) and can be found on manufacturers' webpages (see example below).
  • As high memory bandwidth as possible is beneficial.

Selecting Hardware

Since the hardware and hardware driver market is a moving target, it is a challenge to specify a COMSOL-optimal CPU type and amount of RAM. This will rapidly change over time. Also, there is a trade-off between amount of physical RAM memory and cost. Past a certain amount of RAM, cost starts to grow exponentially. For different budgets, you will find different optimal choices.

A rule of thumb is to pick the highest CPU model number in a series. Below are specifications taken from couple of vendor quotes for two different Intel-based platforms.

A. "Intel Xeon E5-2440 v2 1.90 GHz, 48 GB RAM, 2 Processors, 1600 MHz Max mem"
B. "Intel Xeon E5-2640 v2 2.00 GHz, 64 GB RAM, 2 Processors, 1600 MHz Max mem"

Our intuition tells us that the higher model number (E5-2640 v2) should be better, and more RAM must be better too. Now is a good time to go to a detailed spec chart to do some comparisons. This processor comparison page gives some interesting information: http://ark.intel.com/compare/75263,75267

  1. Both quotes are 2-processor systems, which is the same thing as dual-socket, so we are fine here on both quotes.
  2. The link above tells us that the E5-2440 v2 in quote A has 8 cores. Dual sockets makes it 16 cores in total and thus 48GB/16 = 3 GB/core. The E5-2640 v2 in quote B also has 8 cores, and the memory per core counts to 64GB/16 = 4 GB/core. Quote A does not match the minimum recommendation of 4GB/core.
    QUOTE B wins
  3. Quote B has 8 memory channels (4 per processor) whereas A has 6 memory channels (3 per processor). This together with point 2 above gives us reason to believe that the spec in quote B is at least 33% faster only based on these numbers. Note that the clock rate (1.90 and 2.00 GHz) is probably unimportant here. The number of memory channels is the bottleneck.
    QUOTE B wins.
  4. The Max mem statement is a measure of how fast the memory is (memory bandwidth). Both Quote A and B Quote B offers support for DDR3-1600 memory (and also lower speeds).
  5. Regarding the system bus speed, we can see that both Quote A and B has a QPI of 7.2 GT/s.

Quote B will be more expensive due to the extra RAM, but wins over quote A on two important metrics: it has at least 4GB RAM/core, and more memory channels. Even if we upgrade Quote A to 64GB RAM (4GB/core), Quote B would still win due to its memory channel advantage. For Quote B, if we go higher up in model series than E5-2640 v2 we can get a QPI of 8 GT/s, but no higher than that. In this particular case it can be worth considering upgrading Quote B with E5-2650 v2 CPUs. This type of comparison can be done for AMD processors as well.

In summary, below are three generic examples of computers systems in increasing order of performance

  • Budget alternative, less recommended for high fidelity COMSOL modeling:
    Quad-Core Processor, 16 GB RAM, 64-bit OS, 512 MB Graphics card.
  • Mid-range alternative, good for many COMSOL applications:
    Six-core processor, 48 GB RAM, 64-bit OS, 1GB Graphics card.
  • High-end alternative, for large-scale COMSOL models:
    Dual 8-core processors, 192 GB RAM.


We recommend modern AMD FirePro or NVIDIA based dedicated graphics cards. A list of tested graphics cards can be found on the system requirements page.


General-purpose computing on graphics processing units (GPGPU, also referred to as GPGP and less often GP²) is currently not supported by COMSOL.

Hard Drives

Hard drive speed does not affect COMSOL performance as long as the problem fits into main memory. If main RAM memory is insufficient, the hard drive swap will affect performance severly, regardless of which hard disk you have.

See Also

Selecting hardware for a compute cluster, solution 1116.
COMSOL and Multithreading, solution 1096.



COMSOL makes every reasonable effort to verify the information you view on this page. Resources and documents are provided for your information only, and COMSOL makes no explicit or implied claims to their validity. COMSOL does not assume any legal liability for the accuracy of the data disclosed. Any trademarks referenced in this document are the property of their respective owners. Consult your product manuals for complete trademark details.