Inference slow on Windows 10 (CPU)

Hi. I have dual boot computer (Ubuntu 18 and Windows 10). Pytorch 1.1.0
I trained a pix2pixHD network, and when I try to generate an image on CPU, I get very different performance. On Windows the processing is very slow.

Ubuntu 18: 2 seconds
Windows 10: 9 seconds

This is the code I am using to measure time, at this line:

xtime = datetime.datetime.now()
fake_image = self.netG.forward(input_concat)
ytime = datetime.datetime.now()
print(ytime - xtime)

What could cause the problem?
What can I try?

Thanks!

2 Likes

hi, did you solve the problem?

1 Like

Is the problem solved?

Iā€™m having a similar issue:
I tried monodepth2 on windows and linux for inference on a single image.
The inference took 2 seconds on windows(Ryzen 9 4900 HS) compared to just 0.45 seconds on Ubuntu(i3 processor).
Does anyone know the cause of this discrepency?

1 Like