A problem in MatConvNet to Compiling the GPU

Matlab2017a in Windows10 64 bites. The information of my Gpu:

                        Name: 'GeForce GTX 1080 Ti'
                       Index: 1
           ComputeCapability: '6.1'
              SupportsDouble: 1
               DriverVersion: 9.1000
              ToolkitVersion: 8
          MaxThreadsPerBlock: 1024
            MaxShmemPerBlock: 49152
          MaxThreadBlockSize: [1024 1024 64]
                 MaxGridSize: [2.1475e+09 65535 65535]
                   SIMDWidth: 32
                 TotalMemory: 1.1811e+10
             AvailableMemory: 9.6392e+09
         MultiprocessorCount: 28
                ClockRateKHz: 1582000
                 ComputeMode: 'Default'
        GPUOverlapsTransfers: 1
      KernelExecutionTimeout: 1
            CanMapHostMemory: 1
             DeviceSupported: 1
              DeviceSelected: 1

And I do this:

vl_compilenn('enableGpu', true, ...
               'cudaRoot', 'C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v8.0', ...
               'cudaMethod', 'nvcc')

then the error:

the CPU is success.But the GPU has this error. Please help me!I don't what the problem is.Please!

Answer 1

[futao zhang]

I use the VS2015+MATLAB2017A+CUDA8.0

[Ramy Mahmoud]

can you provide me with the links for installation ???

Answer 2

[futao zhang]

the picture is not clear, please see this:

1 error detected in the compilation of "C:/Users/ADMINI~1/AppData/Local/Temp/tmpxft_00001f1c_00000000-9_data.cpp4.ii". 
data.cu 
Error using vl_compilenn>nvcc_compile (line 541)
Command "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin\nvcc" -c
"F:\ZFT\trackers\matconvnet-1.0-beta24\matconvnet-1.0-beta24\matlab\src\bits\data.cu" -DNDEBUG -DENABLE_GPU -DENABLE_DOUBLE -D__SSSE3__
-gencode=arch=compute_61,code=\"sm_61,compute_61\" -gencode=arch=compute_61,code=\"sm_61,compute_61\"  -I"E:\MATLAB\R2017a\extern\include"
-I"E:\MATLAB\R2017a\toolbox\distcomp\gpu\extern\include"   -gencode=arch=compute_61,code=\"sm_61,compute_61\"
-gencode=arch=compute_61,code=\"sm_61,compute_61\"  -O3 -Xcompiler /MD --compiler-bindir "E:\Program Files\VS2015\VC\..\VC\bin" -o
"F:\ZFT\trackers\matconvnet-1.0-beta24\matconvnet-1.0-beta24\matlab\mex\.build\bits\data.obj" failed.

Error in vl_compilenn (line 486)
      nvcc_compile(opts, srcs{i}, objfile, flags.nvcc) ;

[Peter Fabri]

Be sure to compile with the right compiler version here:

https://www.mathworks.com/support/requirements/previous-releases.html

I had a similar message when I was trying to compile matconvnet beta25 in 2019a with GCC 6.5 instead of GCC 6.3 as specified to the MATLAB version in the above link... although this translated to a different error message.

After doing so, the solution by Nicholas-Schaub here might help remedy a subsequent error:

https://github.com/vlfeat/matconvnet/issues/1200

Answer 3

[Shahzad Ali]

Following combination worked for me:

• MTALAB R2017a Update 4 64-bit (v9.2.0.1226206) --- very important
• Microsoft Visual Studio Professional 2015 Update 3 (v14.0.25431.01)
• CUDA Toolkit v10.0

Once you you have installed above, this command will do the rest of magic:

>> vl_compilenn('enableGpu', true, ...
'cudaRoot', 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0', ...
'cudaMethod', 'nvcc')

It was found that the latest MATLAB release (i.e. R2019b) was the main culprit for the failure. By the way, I'm using Nvidia GTX Titan XP with Windows 10 and the good news is I can use the compiled Matconvnet in MATLAB R2019b later.