[FEA] Re-evaluate Default Memory Resource for cuDF Classic
Open
feature request
Python
- Dominant language
- C++
- Stars
- 9.8k
- Forks
- 1.1k
- Avg merge
- 3d 6m
- Merged PRs (30d)
- 278
Description
In https://github.com/rapidsai/cudf/issues/12235 we eventually concluded that cuDF.pandas should use RMM managed memory pool by default. I think we should have a similar discussion for cuDF classic. The out of the box experience of cuDF is greatly improved when using some other Memory Resource: cudaMallocAsync / cudaMallocManaged
Contributor guide
Assessment
This issue has not been assessed yet.