Using the raytracer with GPU and tensor flow with CPU

[itsghaniz]

Hi, I am using windows 10 for my simulations. As windows doesn't support new versions of tf with GPU, I was wondering if it is possible to use the raytracing with GPU when the tensorflow cpu is installed.
Thanks in advance

[merlinND]

Hello @itsghaniz,

This is not officially supported. I have not tried this myself, but maybe you could:

1. Set the Mitsuba variant to a GPU backend, e.g. mi.set_variant("cuda_ad_rgb")
2. Let TensorFlow perform computations on the CPU (if it does not see any GPUs available), or force it with a with tf.device("cpu"): scope

But this has never been tested, so I would be surprised if something goes wrong.

[itsghaniz]

Hi, really appreciate the reply. I dis as you suggested. Although the code is working, but I am not able to see any use of the GPU.
Thanks

[merlinND]

Hello @itsghaniz,

I'm not sure what else to suggest since this is not supported or really expected to work.
We can leave this discussion open in case someone figures it out.