From FarmShare

Revision as of 09:52, 9 September 2013

Matlab GPU demos

GPU devices in Matlab are supported by the parallel computing toolbox. No special setup is required. Matlab will discover and use Cuda devices automatically.

Resources:

 Information can be found here: http://www.mathworks.com/products/parallel-computing/index.html
 For a list of examples, see: http://www.mathworks.com/products/parallel-computing/examples.html?s_tid=brdcrb
 These matlab functions have GPU support: http://www.mathworks.com/help/distcomp/using-gpuarray.html#bsloua3-1
 Example scritpts: http://www.mathworks.com/help/distcomp/examples/index.html#gpu
 matlab file exchange: http://www.mathworks.com/matlabcentral/fileexchange/34080-gpubench

In this example we will run the Benchmarking A\b on the GPU one found here: [Benchmarking A\b on the GPU]

matlab commands used below:

paralleldemo_gpu_devices
paralleldemo_gpu_backslash(.75);

example output

Here we launch Matlab, run paralleldemo_gpu_devices to print out the Cuda device discovered by Matlab. Then we run the A\b demo.

$ module load matlab
$ matlab -nodesktop
Warning: No display specified.  You will not be able to display graphics on the screen.
Warning: No window system found.  Java option 'MWT' ignored.

                            < M A T L A B (R) >
                  Copyright 1984-2013 The MathWorks, Inc.
                    R2013a (8.1.0.604) 64-bit (glnxa64)
                             February 15, 2013

No window system found.  Java option 'MWT' ignored.
 
To get started, type one of these: helpwin, helpdesk, or demo.
For product information, visit www.mathworks.com.
 
>> paralleldemo_gpu_devices

numDevices =

     1


origDevice = 

  CUDADevice with properties:

                      Name: 'Tesla C2070'
                     Index: 1
         ComputeCapability: '2.0'
            SupportsDouble: 1
             DriverVersion: 5.5000
            ToolkitVersion: 5
        MaxThreadsPerBlock: 1024
          MaxShmemPerBlock: 49152
        MaxThreadBlockSize: [1024 1024 64]
               MaxGridSize: [65535 65535 65535]
                 SIMDWidth: 32
               TotalMemory: 5.6366e+09
                FreeMemory: 5.5344e+09
       MultiprocessorCount: 14
              ClockRateKHz: 1147000
               ComputeMode: 'Default'
      GPUOverlapsTransfers: 1
    KernelExecutionTimeout: 0
          CanMapHostMemory: 1
           DeviceSupported: 1
            DeviceSelected: 1


device = 

  CUDADevice with properties:

                      Name: 'Tesla C2070'
                     Index: 1
         ComputeCapability: '2.0'
            SupportsDouble: 1
             DriverVersion: 5.5000
            ToolkitVersion: 5
        MaxThreadsPerBlock: 1024
          MaxShmemPerBlock: 49152
        MaxThreadBlockSize: [1024 1024 64]
               MaxGridSize: [65535 65535 65535]
                 SIMDWidth: 32
               TotalMemory: 5.6366e+09
                FreeMemory: 5.5344e+09
       MultiprocessorCount: 14
              ClockRateKHz: 1147000
               ComputeMode: 'Default'
      GPUOverlapsTransfers: 1
    KernelExecutionTimeout: 0
          CanMapHostMemory: 1
           DeviceSupported: 1
            DeviceSelected: 1

>> paralleldemo_gpu_backslash(.75);
Starting benchmarks with 13 different single-precision matrices of sizes
ranging from 1024-by-1024 to 13312-by-13312.
Creating a matrix of size 1024-by-1024.
Gigaflops on CPU: 34.472190
Gigaflops on GPU: 56.288799
Creating a matrix of size 2048-by-2048.
Gigaflops on CPU: 49.891778
Gigaflops on GPU: 106.760173
Creating a matrix of size 3072-by-3072.
Gigaflops on CPU: 64.997307
Gigaflops on GPU: 197.257665
Creating a matrix of size 4096-by-4096.
Gigaflops on CPU: 70.944260
Gigaflops on GPU: 266.873255
Creating a matrix of size 5120-by-5120.
Gigaflops on CPU: 84.640804
Gigaflops on GPU: 319.151358
Creating a matrix of size 6144-by-6144.
Gigaflops on CPU: 92.799236
Gigaflops on GPU: 355.467871
Creating a matrix of size 7168-by-7168.
Gigaflops on CPU: 98.141367
Gigaflops on GPU: 388.194551
Creating a matrix of size 8192-by-8192.
Gigaflops on CPU: 102.462204
Gigaflops on GPU: 405.167131
Creating a matrix of size 9216-by-9216.
Gigaflops on CPU: 98.400070
Gigaflops on GPU: 419.867571
Creating a matrix of size 10240-by-10240.
Gigaflops on CPU: 96.734765
Gigaflops on GPU: 434.993371
Creating a matrix of size 11264-by-11264.
Gigaflops on CPU: 112.294056
Gigaflops on GPU: 439.164558
Creating a matrix of size 12288-by-12288.
Gigaflops on CPU: 115.434767
Gigaflops on GPU: 440.911860
Creating a matrix of size 13312-by-13312.
Gigaflops on CPU: 115.826290
Gigaflops on GPU: 460.198654
Starting benchmarks with 9 different double-precision matrices of sizes
ranging from 1024-by-1024 to 9216-by-9216.
Creating a matrix of size 1024-by-1024.
Gigaflops on CPU: 14.479196
Gigaflops on GPU: 21.906035
Creating a matrix of size 2048-by-2048.
Gigaflops on CPU: 27.758668
Gigaflops on GPU: 70.264055
Creating a matrix of size 3072-by-3072.
Gigaflops on CPU: 35.325472
Gigaflops on GPU: 110.924771
Creating a matrix of size 4096-by-4096.
Gigaflops on CPU: 41.316066
Gigaflops on GPU: 151.816138
Creating a matrix of size 5120-by-5120.
Gigaflops on CPU: 47.203079
Gigaflops on GPU: 182.013352
Creating a matrix of size 6144-by-6144.
Gigaflops on CPU: 50.618165
Gigaflops on GPU: 203.495957
Creating a matrix of size 7168-by-7168.
Gigaflops on CPU: 53.713014
Gigaflops on GPU: 220.657206
Creating a matrix of size 8192-by-8192.
Gigaflops on CPU: 54.993392
Gigaflops on GPU: 225.368964
Creating a matrix of size 9216-by-9216.
Gigaflops on CPU: 56.978938
Gigaflops on GPU: 237.973215