How can I solve problems in compile CUDE in matlab ?

I used Matlab 2016a , Visual Studio 2013, CUDA 7.5 and cuDNN v5.1. When I perform compile cudu in matlab.. it shows error like the following.. vl_compilenn: CUDA: MEX config file: 'C:\Program Files\MATLAB\R2016a\toolbox\distcomp\gpu\extern\src\mex\win64\mex_CUDA_win64.xml' nvcc fatal : Unsupported gpu architecture 'compute_21' Error using vl_compilenn>nvcc_compile (line 540) Command "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin\nvcc" -c

[maisun Alzorgani]

vl_compilenn('enableGpu', true) Warning: Setting the 'MW_NVCC_PATH' environment variable to 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin\nvcc' > In vl_compilenn>activate_nvcc (line 695) In vl_compilenn (line 288) vl_compilenn: CUDA: MEX config file: 'C:\Program Files\MATLAB\R2016a\toolbox\distcomp\gpu\extern\src\mex\win64\mex_CUDA_win64.xml' nvcc fatal : Unsupported gpu architecture 'compute_21' Error using vl_compilenn>nvcc_compile (line 540) Command "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin\nvcc" -c "C:\Users\Win7\Documents\MATLAB\matconvnet-1.0-beta24\matlab\src\bits\data.cu" -DNDEBUG -DENABLE_GPU -DENABLE_DOUBLE -D__SSSE3__ -gencode=arch=compute_21,code=\"sm_21,compute_21\" -I"C:\Program Files\MATLAB\R2016a\extern\include" -I"C:\Program Files\MATLAB\R2016a\toolbox\distcomp\gpu\extern\include" -gencode=arch=compute_21,code=\"sm_21,compute_21\" -O3 -Xcompiler /MD --compiler-bindir "C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\..\VC\bin" -o "C:\Users\Win7\Documents\MATLAB\matconvnet-1.0-beta24\matlab\mex\.build\bits\data.obj" failed.

Error in vl_compilenn (line 485) nvcc_compile(opts, srcs{i}, objfile, flags.nvcc) ;

Answer 1

[Walter Roberson]

http://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html

Cudnn needs cuda 3.0 or later. Your device only has cuda 2.1. You will need be able to run the software on that hardware.

You are also at pretty much the last MATLAB version that can use that hardware for gpu purposes, I seem to recall.

[Walter Roberson]

I was mistaken, CUDA 2.1 is still accepted for gpu arrays up to R2017b. However, the deep learning routines introduced in R2016b in the Neural Network toolbox need CUDA 3.0 .

https://blogs.mathworks.com/deep-learning/2017/10/20/deep-learning-with-gpus-and-matlab/