Skip to content

Commit

Permalink
roll back name change
Browse files Browse the repository at this point in the history
  • Loading branch information
rjzamora committed Sep 16, 2024
1 parent d476d7d commit 16aaf85
Showing 1 changed file with 4 additions and 3 deletions.
7 changes: 4 additions & 3 deletions docs/source/examples/best-practices.rst
Original file line number Diff line number Diff line change
Expand Up @@ -44,13 +44,14 @@ We also recommend allocating most, though not all, of the GPU memory space. We d

Additionally, when using `Accelerated Networking`_ , we only need to register a single IPC handle for the whole pool (which is expensive, but only done once) since from the IPC point of viewer there's only a single allocation. As opposed to just using RMM without a pool where each new allocation must be registered with IPC.

Memory Spilling
~~~~~~~~~~~~~~~
Spilling from device
~~~~~~~~~~~~~~~~~~~~

Dask-CUDA offers several different ways to enable automatic spilling from device memory.
The best method often depends on the specific workflow. For classic ETL workloads using
`Dask cuDF <https://docs.rapids.ai/api/dask-cudf/stable/>`_, native cuDF spilling is usually
the best place to start. See :ref:`Spilling from device <spilling-from-device>` for more details.
the best place to start. See :ref:`Dask-CUDA's spilling documentation <spilling-from-device>`
page for more details.

Accelerated Networking
~~~~~~~~~~~~~~~~~~~~~~
Expand Down

0 comments on commit 16aaf85

Please sign in to comment.