Building pytorch from source issue

After I updated pytorch, it doesn’t support my GPU anymore(3.5 compute compatibility) so I uninstalled and have been trying to install from source from this.

python setup.py install

Using this code, my system freezes when the build starts.
Do I need to adjust build options? I do not know what to change.

Will I be facing any issue if I install previous version of pytorch which was compatible for my gpu?

Any help appreciated. Thanks

system information:
OS: Ubuntu 18.04
RAM: 8GB
CPU: Intel® Core™ i5-5200U CPU @ 2.20GHz
GPU: GeForce 920M 2GB
NVIDIA driver: 440.33.01
cuda: 10.2

Hi,

What do you mean by the system freezer? Do you run out of ram?

Yes, cpu and ram usage goes ~100%, no other processes running. No response after 15+ mins, had to hard reboot everytime.

You’re running out of RAM during the compilation :confused:
You might want to reduce the number of threads that it can use to reduce the memory usage by setting MAX_JOBS=2 as an env variable before calling the setup.py script. Note that this will slow down the compilation quite a lot.

1 Like

Okay thanks, I’ll try it.
As an alternative, if I install the pytorch 1.2 wheel file from the older versions page, will I be facing any performance issues on model building? I’ve tried finding information on it online but not finding anything relevant.

Hi,

Older versions might not have all the most recent features, bugfix and perf improvements. So it will depend a lot on what you’re using.

1 Like