[Python] How to make Tensorflow run on GPU on Windows?

Question

I may be way out on a limb here, but does any of you guys have any experience in making tensorflow run on GPU instead of CPU? I have been trying to do that on a Windows platform, but can't seem to succeed.

I found a good article on the net, but could not replicate it on my machine. Using conda, I managed to create an environment to run keras/tensorflow, but when I run IPython for my neural network excercises, it still runs on the CPU... :(

Answer 1

I of course installed the GPU (Graphics Processing Unit) version, but even when I started IPython within the virtual environment, it still processed the network's layers via the regular CPU.
I got NVidia and it is supposed to be able to cast all matrix computations onto its myriads of little processors. But apparently I can't do it on Windows... :(

Answer 2

not saying that I know what I'm talking about, but sounds like it can be fixed by running a VM with another OS if the problem is solely windows (:

Answer 3

I apologize for I don't have an answer to the main question, but:

@S Vengat: GPU=Graphics Processing Unit, aka video card.

@Maya: I think VMs offer you a virtualized hardware, like Generic video card, not your actual hardware (except for the CPU, I think). They have a "passthrough" for USB devices, and even those don't all work (I never got my Microsoft and Logitech webcams to work neither in VMWare, nor VirtualBox).

Answer 4

Install Tensorflow GPU version.
I think only very few GPUs support this.