3

I have been following instructions to set up Theano to use a GPU on Windows.

The issue is I cannot follow these instructions exactly because I have a new graphics card, the GEForce GTX 980M and it only works with cuda 7.0. (The instructions suggest cuda 5.5). Everything works fine except when it gets time to run the GPU then I get an error:

nvcc fatal   : Value '2008' is not defined for option 'cl-version'

['nvcc', '-shared', '-O3', '--use-local-env', '--cl-version=2008', '-Xlinker', '/DEBUG', '-D HAVE_ROUND', '-m64', '-Xcompiler', '-DCUDA_NDARRAY_CUH=bf8d22b192850a85a00263cce1175696,-D NPY_NO_DEPRECATED_API=NPY_1_7_API_VERSION,/Zi,/MD', '-Ic:\\scisoft\\theano\\theano\\theano\\sandbox\\cuda', '-IC:\\SciSoft\\WinPython-64bit-2.7.9.4\\python-2.7.9.amd64\\lib\\site-packages\\numpy\\core\\include', '-IC:\\SciSoft\\WinPython-64bit-2.7.9.4\\python-2.7.9.amd64\\include', '-o', 'C:\\Users\\Thor\\AppData\\Local\\Theano\\compiledir_Windows-8-6.2.9200-Intel64_Family_6_Model_70_Stepping_1_GenuineIntel-2.7.9-64\\cuda_ndarray\\cuda_ndarray.pyd', 'mod.cu', '-LC:\\SciSoft\\WinPython-64bit-2.7.9.4\\python-2.7.9.amd64\\libs', '-LC:\\SciSoft\\WinPython-64bit-2.7.9.4\\python-2.7.9.amd64', '-lpython27', '-lcublas', '-lcudart']

The version of nvcc I have installed does not have version -2008 (that looks to have been deprecated by the lastest cuda 7.0 version, but 2010 is allowed). What is the best way to fix it? Should I hard code it in Theano into the file cuda\nvcc_compiler.py? I tried that and it seems to try to use 2008 version anyway. Is there a later version of Theano that would use the later nvcc version?

Improve this question
3
  • Also when I was installing cuda 7 it said that it couldn't find visual studio even though I have VS10 installed and had just installed a free version of VS12. Any idea why and how to fix? I installed VS12 to a different path to normal.
    – Thor Russell
    Commented Jun 29, 2015 at 6:17
  • Review the actual supported configurations for CUDA 7 in the getting started document and I think this will answer your question about why CUDA 7 didn't find a supported version of VS. The only "free" version of VS supported by CUDA 7 is VS 2013 community edition.
    – Robert Crovella
    Commented Jun 30, 2015 at 5:29
  • Yes thanks, I installed 2013 community and that problem went away.
    – Thor Russell
    Commented Jul 1, 2015 at 22:21

2 Answers 2

Reset to default
6

I ran into a similar problem when trying to install Theano on Win 8.1 64bit with CUDA 7.0., using a GTX 750Ti graphics card. I was able to get it working by following these instructions.

Improve this answer
1
  • Thankyou. That almost worked for me, and did work completely after I read the comment on that page about the different versions of Anaconda. I needed to run the instruction "conda install mingw libpython" Then the gpu ran fine.
    – Thor Russell
    Commented Jul 1, 2015 at 22:18
5

For me it started working when I replaced

[nvcc]
flags = --use-local-env  --cl-version=2008

by

[nvcc]
compiler_bindir=C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin

in the .theanorc file.

Improve this answer
0

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.