search

facebookresearch / pifuhd

High-Resolution 3D Human Digitization from A Single Image.
Other
9.5k stars 1.44k forks source link

Cuda OOM in Win10 #24

Closed piggybox closed 3 years ago

piggybox commented 4 years ago

I have a 2080 card with 8G memory but still ran into the following OOM problem. The env is pytorch 1.5.1 on Win 10. I pretty much closed all other apps when running this.

Error msg: CUDA out of memory. Tried to allocate 782.00 MiB (GPU 0; 8.00 GiB total capacity; 4.43 GiB already allocated; 610.25 MiB free; 5.29 GiB reserved in total by PyTorch)

Any one shred a light on how to make more memory available?

trufty commented 4 years ago

Try reducing "num_samples" and "interval" in recon.py from 50,000 to 25,000. That worked for me.

kendrick90 commented 3 years ago

I tried this to get it working on a laptop with an rtx2060 6GB and it won't work no matter how small I make the num_samples or interval :(

wingdi commented 3 years ago

I tried this to get it working on a laptop with an rtx2060 6GB and it won't work no matter how small I make the num_samples or interval :(

I have same problems win10/RTX2060 6GB/desktop , num_samples 、interval 128， still OOM...... do you solve it?

kendrick90 commented 3 years ago

No seems like it won't run. I don't know enough to go further but I found you can increase the size from 256 to 400 in the the colab. It shows some malloc lines but still outputs the file. Don't know if it's a soft crash or what but it seems to work.

On Sun, Nov 8, 2020, 7:52 AM wingdi notifications@github.com wrote:

I tried this to get it working on a laptop with an rtx2060 6GB and it won't work no matter how small I make the num_samples or interval :(

I have same problems win10/RTX2060 6GB/desktop , num_samples 、interval 128， still OOM...... do you solve it?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/facebookresearch/pifuhd/issues/24#issuecomment-723596328, or unsubscribe https://github.com/notifications/unsubscribe-auth/AMDVBYAISZIF6DEV4OWWFALSO25DDANCNFSM4OJHU3CQ .