k-Wave

1. saturnnan
   

   Member
   Joined: Jul '16
   Posts: 3

   Hello, everyone:
   I am trying to simulate acoustic shear wave propagation using k-wave pstdElastic3D function.It took a long time for one simualtion.
   Is there any method or related function that makes this simulation work on the GPU?

   Thanks a lot!

   Hank

   Posted 5 years ago #

2. Patrick
   

   Member
   Joined: Jun '19
   Posts: 14

   Hi,

   You can use the DataCast option set to 'gpuArray-single' to run pstElastic3D on the GPU. This requires the MATLAB parallel computing toolbox.

   Hope this helps,

   PC

   Posted 5 years ago #

3. saturnnan
   

   Member
   Joined: Jul '16
   Posts: 3

   Hi,Patrick:
   Thanks a lot!
   That is really helpful to me.The time consumption of simulation reduced by 35%.
   The computational efficiency of this function does not seem to be as efficient as the kspaceFirstOrder3DG.
   Can not wait for the cuda version of pstElastic3D :)

   Thanks again!

   Hank

   Posted 5 years ago #

4. elliotP
   

   Member
   Joined: Mar '19
   Posts: 10

   Hi Patrick,

   I am not familiar with the MATLAB parallel computing toolbox. How can I use the DataCast option setting 'gpuArray-single' to running pstElastic3D on the GPU? Can you recommend any document for me?

   Thank you,
   Michael

   Posted 5 years ago #

5. Patrick
   

   Member
   Joined: Jun '19
   Posts: 14

   Elliot,

   Usually pstdElastic3D is called like this:

   sensor_data = pstdElastic3D(kgrid, medium, source, sensor)

   To run it on the GPU, define an input_args cell array with DataCast set to gpuArray-single and give it to pstdElastic3D, like this:

   input_args = {'DataCast','gpuArray-single'};
   sensor_data = pstdElastic3D(kgrid, medium, source, sensor, input_args{:});

   I recommend reading the k-Wave documentation/examples, which has this information and more (http://www.k-wave.org/documentation/pstdElastic3D.php)

   ---

   Additionally, to change the GPU you are running the simulation on, MATLAB's parallel computing toolbox has a function called gpuDevice, which indexes devices differently than usual: it starts at 1 instead of 0. So to set your device to the zeroth Nvidia device, use:

   gpuDevice(1);

   Posted 5 years ago #

RSS feed for this topic

Reply

You must log in to post.