Xingbaji/PNCG_IPC

Poor performance on laptop3080

Opened this issue · 3 comments

Thanks for such an interesting project. I can successfully run the project. The full requirement should also include "imageio", "scipy".
I just try the demo "cubic_demos" by run python cubic_demos.py.
The performance on GUI suggests that it only has 3-4 FPS, which is far from expectation.
Any help and comment? (I think I make some mistake on build.)
image
image
image
Also, when the cube rotated further, the collision seem to get failed?
[interpreter]
python 3.10
[deps]
taichi v1.7.1
meshtaichi-patcher v 0.0.19
[hardware]
Nvidia RTX3080 (laptop) , intel i9-10900HK@2.4GHz
[arch]
CUDA toolkit 12.3

Thanks for any comment and suggestion.

Thanks for your reply. About the performance, are you running in Windows? The performance is slow in Windows. If you are using GPU with other process, the speed will also be decreased. The cubic_demos.py desn't include the collision pipeline. You can try twist_demo.py for performance with collision.

Thanks for the reply.
Yes, I just noticed that cube demo is on absent of the collision handling.
Yes, I run this demo on windows; one of my labmate will benchmark the performance on Ubuntu. Thanks for this interesting project a lot!

我估计这个代码还有很多其他的问题,现在正在更新中,遇到问题欢迎随时提出。github上的回复较慢,有问题可以加微信交流,微信: shendoubixing