GPU Enabled MIKE 21 FMHD

The GPU enabled MIKE 21 FMHD in Version 2014 has been out for some time now and I was wondering whether any user would like to share their use experience, especially with regard to the use of non-DHI tested/benchmarked GPU chips. Personally, I have used a Quadro K5000, which is supposed to be of a relatively high-end professional pedigree but the speed enhancement is only 1-2 times, way off the 100X enhancement stated in the DHI verification report, though that is for the GeFORCE GTX Titan chip.

Hi Say Lee,
The enhancements are dependent on a number of factors. The 100x number is compared to a one core processor with/without GPU e.g. Tests have also showed that the potential is greatest when you approach ~500 000 cells, do not use double precision computation scheme etc. Also additional power supply and cooling unit could likely be a good idea. (I assume your workstation specs are very good as you use such an expensive GPU)
The above are a few factors as I remember them (it was some time ago I read about it) :-)
We've got a work station on the way with double GPU slots on the mother board, so we can utilize one high end GPU for number crunching and one for display management. I think that maybe you can't use the full GPU potential if you don't dedicate it just to calculations...

You should probably contact a company that builds custom work stations, they should know more specifically how an optimal setup would look. Perhaps they can gice some suggestions.

All the best!
Jonas

Thanks, Jonas.

I read the run log file for a completed GPU-enabled HD sim and it has the following at the beginning:

======================= Computing Environment ========================
  Computer name      : [BLANKed out]
  Number of processors: 32
======================================================================


==================== CUDA GPU Device Information =====================
  Number of GPU devices  : 1

  GPU device number      : 1
  GPU device name        : Quadro K5000
  CUDA Compute Capability : 3.0

    selected_device_number : 1 (default)
    number_of_threadsPerBlock : 128 (default)
======================================================================

and the end:


============================ Performance =============================
  Number of threads on CPU: 8
 
  Double precision GPU calculations
======================================================================

Clearly it the double precision computation has been invoked. However, I was not able to find where I can disable the double precision computation during the run setup. Can you advise? Thanks.

Say Chong

I usually enable/disable by opening the m21 file in a text editor, and under the section HYDRODYNAMIC PARAMETERS you should see this under the OPTIONS. Try that.
If you write "Include_DB_computations = true" DB is enabled. Try to erase or set to "false" and run again and see if DB is disabled.
Otherwise you should be able to do it in pfs editor but I'm not sure how to do so as I never go that way myself.

Also; Are you using a remote desktop connection, or actually sitting by the computer with the GPU?
I'm asking because Remote access and GPU does not work with any type of software. I think RealVNC works though.

Best regards,
Jonas

Thanks, Jonas, for the help and apologies on the late response. I have run using an amended version of the run setup file (for changing the model start time in an add-on module such as Transport coupled to the HD) but need to run it in DOS using a batch file.

Anyway I will try your recommended approach and see whether that will help keep the run time manageable.

And yes, we have to launch the GPU-enabled runs locally at the workstation as the Remote Access facility that we use does not work.

Say Chong

Hi,

The single calculation can also be selected within Mike Zero by going to File/Option/User settings, then go to the Simulation Setting tab.
That’s where the Single Precision calculation is ticked (single precision) or unticked (double precision).

Regards,

That is a great advice, Nilor. Thanks a million.

Say Chong

The software is the biggest difference between Tesla, Titan, and Quadros (and the prices).

As of right now, Titan Blacks are likely to have the best cost/benefit for MIKE 21, but their drivers do not support remote desktoping (unlike the Tesla TCC drivers, which allow you to dedicate to specific applications). However, Titans do not have ECC memory, unlike Quadros and Teslas, which are good for scientific computing, but I think iterative solvers should account for any large convergence issues due to bitflips??? An added benefit to the Titan is that it's possible to support an additional GPU (see Titan Z), but for the most part, Teslas are far more scalable for scientific computing purposes.

Bottom line is to get a Titan over a Quadro or Tesla for a workstation. If you're going the server route and have a lot of cash lying around, you can probably get more performance out of Teslas due to their scalability assuming you can use multiple GPUs. I don't think MIKE 21 supports multiple GPUs currently, so this probably isn't the most ideal route.

Hello everyone,

I'm currently looking into purchasing a Titan Z, but since it's functionally two GPUs, I am wondering if anyone knows if it's supported by the software or if the hardware acts like one GPU?

Oh yeah, one more thing: we've tried and tested a Titan Black its definitely worth the cost, which is why we're now considering a Titan Z.

Thanks,

Adam

Hi Adam,

Now that the 2015 version comes with multiple GPU capability, is your Titan Z problem resolved and how is its performance if you would like to share your experience?

Any views from others are welcome too!

Thanks and regards,

Say Chong