-
Notifications
You must be signed in to change notification settings - Fork 2.3k
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
[doc] Update install.rst and Misc. #1009
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank for fixing these!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great work! I really appreciate these fixes. Thanks 👍
docs/install.rst
Outdated
|
||
* Try adding ``export TI_USE_UNIFIED_MEMORY=0`` to your ``~/.bashrc``. This disables unified memory usage in CUDA backend. | ||
This may be because your NVIDIA card is pre-Pascal and does not support `Unified Memory <https://www.nextplatform.com/2019/01/24/unified-memory-the-final-piece-of-the-gpu-programming-puzzle/>`_. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'd say ''This might be due to the fact that ...''
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
BTW, unified memory is not supported since GTX770? Maybe I remembered a lot of things wrong about CUDA...
So if I disable the unified memory, what computation capability do I need to use Taichi?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
BTW, unified memory is not supported since GTX770? Maybe I remembered a lot of things wrong about CUDA...
NVIDIA has a fallback: https://devblogs.nvidia.com/unified-memory-cuda-beginners/ see What Happens on Kepler When I call cudaMallocManaged()?
These GPUs doesn't support memadvise
though...
So if I disable the unified memory, what computation capability do I need to use Taichi?
Taichi doesn't have a hard requirement for that, but GTX 9 series should work (without unified memory/adaptive memory pool allocation)...I haven't tested GTX 7 series (which released 7 years ago)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Oh thanks for that. I had a GTX770 for a long time and tried the unified memory ever since it was available (I think it was CUDA 6.0 but I'm not sure). I found it did a lot of unnecessary memory synchronization and I couldn't turn it off, so I just decided never use it again since then 😢
Thank you for the helpful suggestions! I fixed all of them. Let me merge this so that these changes will be reflected in v0.6.5. |
Related issue = #941
[Click here for the format server]