Running C++ code on CPU with a Cuda-enabled installation of tensorflow about tensorflow_cc

[FloopCZ]

Hi, I am glad that tensorflow_cc works for you.

I am not sure if this is what you need, but if you have already built tensorflow_cc with GPU support, there is no need to rebuild it on order to run your program on CPU. Just launching your program with empty CUDA_VISIBLE_DEVICES environmental variable should do the job:

CUDA_VISIBLE_DEVICES="" ./my-program

This solution should actually work for any program utilizing CUDA, see Accelwareblog for more tips.

Is this what you wanted?

[szm-R]

Thank you very much! That's exactly what I need!

…

On Sat, Mar 31, 2018 at 2:15 AM, Filip Matzner ***@***.***> wrote: Hi, I am glad that tensorflow_cc works for you. I am not sure if this is what you need, but if you have already built tensorflow_cc with GPU support, there is no need to rebuild it on order to run your program on CPU. Just launching your program with empty CUDA_VISIBLE_DEVICES environmental variable should do the job: CUDA_VISIBLE_DEVICES="" ./my-program This solution should actually work for any program utilizing CUDA, see Accelwareblog <http://acceleware.com/blog/cudavisibledevices-masking-gpus> for more tips. Is this what you wanted? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#74 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/APaJX2t1EYWMERPpDvE1zXKOpIUVhSbXks5tjqd5gaJpZM4TBZnu> .