The solution time exhibits two distinct regions. Next, let’s look at the solution time for this problem versus, again, the degrees of freedom. Reported virtual memory (blue) and physical memory (orange) needed versus problem size, in terms of millions of DOFs, for a model involving heat transfer in solids. Clearly, adding more memory to our computer would let us solve a larger model, and we can linearly extrapolate to predict the memory requirements.įigure 1. There is also a point beyond which the software will report that there is not enough memory, so the model cannot be solved at all. There is a point where the available space in the physical RAM is used up, and a significant fraction of the data resides on the SSD.

Physical memory will always be less than both virtual memory and the amount of installed RAM, since the OS, as well as other programs running on the computer, will also need to occupy some fraction of the RAM. (These data are from a Windows® machine, and memory management on all supported OS’s is very similar.) The physical memory is the amount of RAM memory that is being occupied. The virtual memory is the amount of memory that the software is requesting from the operating system (OS). We will first look at the memory requirements as the model size is increased in terms of two reported quantities: the virtual memory and the physical memory. That is, the model is solved on successively finer and finer meshes while using the default solver on a typical desktop workstation machine - in this case, an Intel® Xeon® W-2145 CPU, with 32 GB RAM, and a solid-state drive (SSD). We will start by setting up a three-dimensional model for the steady-state temperature distribution within a domain and study the memory requirements and solution time with an increasing number of degrees of freedom (DOFs). It has since been updated to reflect new features and functionality.

Let’s take a deeper dive into this topic.Įditor’s note: The original version of this post was published on October 24, 2014. One of the more common questions we get is: How large of a model can you solve in the COMSOL Multiphysics® software? Although at some level this is a very easy question to answer, it inevitably leads to a longer conversation about numerical methods, modeling strategies and solution algorithms, and computer hardware performance, as well as how to best approach computationally demanding problems.

Comsol Xfem