Not producing correct result about flownet2-tf HOT 4 CLOSED

Hi!

I had a similar error when I built against the wrong CUDA capabilities version (see this commit: c74bc8e).

You may need to hardcode the correct --gpu-name to match your GPUs Compute Capabilities. By default, it is sm_30. It's annoying because compiling for the wrong GPU doesn't yield any compile or runtime errors...it just outputs incorrect results as you've seen here.

Hope this helps.

from flownet2-tf.

Thank you very much for the reply!
I managed to successfully run the code on another machine but the problem is not solved. I will try out your suggestion!

from flownet2-tf.

Any update on this?

from flownet2-tf.

Your suggestion just works, thanks!

When explicitly specifying the architecture as
-GPUCFLAGS = -c -gencode=arch=compute_30,code=sm_30
the program works.

from flownet2-tf.