Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

require 'cutorch' takes long time #475

Closed
DmitryUlyanov opened this issue Aug 20, 2016 · 3 comments

Comments

Projects
None yet
3 participants
@DmitryUlyanov
Copy link

commented Aug 20, 2016

Hello, can you please explain why the initialization takes so long?

Previously, I noticed, that it takes long only when there are no GPU processes at all. In this case even nvidia-smi works quite long. So I was opening torch with require cutorch and moving it to background, so there was always something on GPU. In this case all other torch scripts required cutorch fast. But now this trick does not work. I also moved to different server, so it can be the reason, but can you please specify, maybe there is a way not to wait for 10 sec every time?

I updated torch and cutorch just yesterday.

Thank you.

@szagoruyko

This comment has been minimized.

Copy link
Member

commented Aug 20, 2016

it takes so long because of #431
takes less than 2s if you start it with CUDA_VISIBLE_DEVICES=0.

@DmitryUlyanov

This comment has been minimized.

Copy link
Author

commented Aug 21, 2016

@szagoruyko 5 sec for me :( Hope it will be better in CUDA 8.0. Thank you!

@soumith

This comment has been minimized.

Copy link
Member

commented Aug 24, 2016

i'm working on somewhat fixing it this week by lazy initializing devices in cutorch.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.