k-Wave

1. Tomluk
   

   Member
   Joined: Oct '10
   Posts: 1

   Hi.
   Did you do some tests on how much faster the simulations should work?
   I used one of yours example (example_us_snells_law.m) for tests. When using GPUmat, calculations were much slower. It took 19 seconds for normal calculations, and 33 sec. using GPUmat. So my question is: is it my computer (specification below) or maybe something wrong with the script or is it a GPUmat problem?
   I am using notebook with Windows 7 32bit, Intel I5-M430 2.27Ghz, nvidia 330M 1GB, 3GB RAM, Matlab 2010a.
   Regards
   Tomasz

   Posted 14 years ago #

2. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,643

   Hi Tomasz,

   The speed-up you observe when running k-Wave on the GPU will depend on two things:

   (1) The specifications of the GPU. Notebook GPUs in particular tend to have only a small number of cores (48 in your case) with a relatively low clock speed and often only modest amounts of memory. This will reduce the achievable performance.

   (2) The size of the simulation grid. Because of the additional computational overhead associated with setting up the computation on the GPU, you will likely only see performance increases when using larger grid sizes.

   For some published benchmarking comparisons, see Fig. 7 in k-Wave: MATLAB toolbox for the simulation and reconstruction of photoacoustic wave-fields. In this example, using the GPU only becomes faster when the grid size reaches 2^18 elements. You could try increasing the size of the grid in the example you are running to see if you observe any speed-up as the grid size increases.

   We are currently working on collating performance comparisons using a much bigger range of PCs and GPUs. I haven't had much experience with notebook GPUs so I would definitely be interested in hearing how you get on.

   Brad.

   Posted 14 years ago #

RSS feed for this topic

Reply

You must log in to post.