[hatecrew]

I have a dell workstation running on dual xeon e5-2665, 64Gb of rams
It is 16 threads and 8 cores per cpu. That is a total of 32 threads and 16 cores.

I am currently using it to run simulations in AVL FIRE.
The license that i have can only use 8 nodes.
So i'm currently using only 8 cores or 8 threads on 4 cores (hyper threading)

My question is that, is there a way to run simulations at full potential of my machine?

These are some questions that i came up with

1. Can I group several cores as 1 node, like 4 cores per node.
If this is not possible then #2

2. Declare each physical cpu as single node, then run simulation at 2 nodes.

3. Do hyper-threading improve the performance? In the case that I could use
all cpu cores, what would be better between 32 threads on 16 cores or
16 threads on 16 cores?

If there is other way please tell me.

Thank you.

[flotus1]

Originally Posted by hatecrew

My question is that, is there a way to run simulations at full potential of my machine?

No.

Originally Posted by hatecrew

1. Can I group several cores as 1 node, like 4 cores per node.
If this is not possible then #2

2. Declare each physical cpu as single node, then run simulation at 2 nodes.

3. Do hyper-threading improve the performance? In the case that I could use
all cpu cores, what would be better between 32 threads on 16 cores or
16 threads on 16 cores?
Thank you.

The usual licensing model is for threads, no matter if they are called "nodes" or whatever. I guess this is what causes your confusion. Your license is valid for 8 threads, the best you can do is to use 8 cores for a simulation.
Hyperthreading usually does not help for CFD simulations. This is especially true in your case since you have more physical cores than parallel licenses available. It is usually recommended to disable hyperthreading in the bios.

[evcelica]

Agreed, definitely disable hyper-threading in the bios.

[RobertB]

If you can insert mpirun commands into the start up script you will almost certainly go fastest if you core lock the code and use four cores from each cpu.

My guess is that you should probably use cores 0,2,4,6 but you would have to dig into the cache design to really understand the optimum.

[larsenmm]

As "RobertB" mentioned, if you can lock the code to use specific CPU cores you potentially should have each core running at 3.0 Ghz turboboost, instead of the 2.4 Ghz standard (if CPU world data can be trusted, http://www.cpu-world.com/CPUs/Xeon/I...20E5-2665.html).

Have worked with AVL before (Aftertreatment simulations, spray) and as far as i remember it does not act exactly the same as Fluent CFD when it comes to utilizing CPU frequencies. Had 16 licenses to use and experienced more or less "linear" scaling from a 4core I7-3770 CPU to a 16 core Xeon server, same frequency and all. I think actually the optimum was 15 cores, due to the "overhead" of AVL Fire simulations, which made it slightly more effecient to use 15 cores and leaving 1 core free for the "other" stuff. OPtimum = less simulation time. At least that was my experience, in a medium to low scientific test i did