Hardware method to reduce the computation time

Hello mike community,

I model a 23 km river. I use the Ecolab module. I did custom water quality equations myself.
I model 3 month data by run. The computational time for a run is approximately 20 hours.

I know I can reduce modeling time by:
- Increasing the time step ( but it reduce stability )
- Increasing the "space step" (but it reduce accuracy )
- Simplifying the flow hypothesis and water quality equations. ( but, I will have to add even more equation quality )
- ...

But, what is the hardware element limiting the calculation speed?
I run 4 (apparently the maximum) simultaneous version of the model at the same time.
Each take 8% of a 12 core CPU.
So I use only 33% of the total CPU powerness.
How to found the bottleneck of the computational speed ? Any idea ?
Thank you in advance if you have some ideas.

A SSD harddrive could speed up calculation?


Configuration :
CPU: i7 3930k
GPU: Geforce 210 ( not use in calculation, right ? )
Motherboard: Asustek p9x79
HD:  WD5000AAKX
OS: win7 64bits
RAM : 12 GB

An SSD option could speed up the computation. Though note that this only is if your simulation is IO intensive i.e. do you write results very frequently e.g. at every time step?

GPU is not used with a MIKE 11 computation. You may utilize GPU if you switch to MIKE 21 FM.

The way you describe your set-up you have one streatch of river with no confluences. Such a set-up is numerically solved through a double sweep algorithm which is very well suited for single thread calculations.

You could consider the various time steps i.e. the WQ time step, the HD time step and the AD time step. Time step may be limited by the WQ equations due to the ODEs being solved

It may be an excellent idea to do a various time steps for the WQ and the HD.
I will try and asses the performance on WQ result.

My time step is low :
time step : 0.5 s
maximum dx : 100 m
(a part of my river widens promptly, it seems to decrease stability)

I write results every 5 minutes.
And yes, I have only one branch. Confluences are integrated as point source inflows.