Posted Fri, 11 May 2018 11:43:06 GMT by Johannes Merkl Student
Dear Feflow-Community,
we are setting up a new workstation for transient groundwater models. Since we are still in the planning stage, I would like to get your recommendations on hardware components, especially processors.
Right now I would prefer the  Intel i5-8400, but I was thinking of switching to the AMD Ryzen 7.

What processors/CPU do you use and would recommend? 
Are there any other "I wish I would've known that earlier" suggestions on the hardware-setup?
Is there a way to quantify the relationship of nodes/timesteps vs. computing time for each CPU to have definite arguments, like a benchmark?
Thanks for all comments,

Posted Fri, 11 May 2018 13:49:44 GMT by Björn Kaiser
Information about hardware resources for FEFLOW 7.1 are provided by the release notes:

If you want to make some performance tests, I suggest to use Python. The following FEFLOW API functions provide what you are looking for:

[font=courier][color=blue]doc.getNumberOfNodes()[/color][/font]: Retrieve the number of nodes
[font=courier][color=blue]doc.getAbsoluteSimulationTime()[/color][/font]: Retrieve the absolute simulation time (t >=0.). You may call the function in the callback [font=courier][color=blue]postTimeStep()[/color][/font]

Apart from the FEFLOW API, the Python library [b]psutil[/b] ( is a great library for hardware diagnostics. [b]time[/b] ( and [b]datetime[/b] ( are a nice module for time-related functions.

Posted Sun, 13 May 2018 05:24:30 GMT by Johannes Merkl Student
Great! Thank you Björn for the quick reply!
Posted Mon, 14 May 2018 21:04:23 GMT by adacovsk
Hi Bjorn,

My understanding is that depending on the solver (serial or parallel) there could be a bottleneck... [code]doc.getNumberOfNodes()[/code] would only test one of these computations?

PCG (serial) solver being good for flow transient models, and SAMG (parallel) being good for transient contaminant transport from my experience.

Currently, I plan on using i9-9700X to maximize serial and parallel computation.
Posted Wed, 16 May 2018 19:52:41 GMT by Björn Kaiser
Hi Adam,

I have a similar observation. As you pointed out, the PCG solver is fast for transient flow models, because the matrix assembly is highly parallelized. According to my observations, the SAMG is quite fast for steady-state or quasi steady flow models.

Yes, the number of nodes are very interesting with regard to serial VS. parallel. Moreover, the number of cores are also interesting. More cores do not necessarily imply faster computations. Most likely a drastic / exaggerated example: if the number of cores are "high" and if the number of nodes are "low", a reduction of cores could even accelerate the computations.

You must be signed in to post in this forum.