-
Notifications
You must be signed in to change notification settings - Fork 94
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
Add docs for UCX #294
Add docs for UCX #294
Conversation
@quasiben I have everything in for this PR, please take another look when you can. |
docs/source/specializations.rst
Outdated
Automatic Instantiation of One-Worker-Per-GPU | ||
--------------------------------------------- | ||
|
||
Using the ``dask-cuda-worker`` or ``LocalCUDACluster`` will automatically launch one worker for each GPU available on the node where that is running, avoiding the need for users to select GPUs in their application and thus reducing code complexity. |
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.
Using the ``dask-cuda-worker`` or ``LocalCUDACluster`` will automatically launch one worker for each GPU available on the node from where it was executed, avoiding the need for users to select GPUs in their application and thus reducing code complexity.
docs/source/specializations.rst
Outdated
Automatic Selection of InfiniBand Device | ||
---------------------------------------- | ||
|
||
When UCX is activated, Dask-CUDA is also capable of selecting the topologically closest InfiniBand device to each GPU, thus ensuring optimal path and improving performance even further by using Remote Direct Memory Access (RDMA) when available. See the :doc:`UCX <ucx>` page for more details. |
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.
When InfiniBand is activated, Dask-CUDA is also capable of selecting the topologically closest InfiniBand device to each GPU, thus ensuring optimal path and improving performance even further by using GPU Remote Direct Memory Access (RDMA) when available. See the :doc:`UCX <ucx>` page for more details.
Made two small suggestions but otherwise I think looks good! |
@quasiben suggestions applied, please merge at will if you have no further comments. |
Thanks again @pentschev ! |
No description provided.