You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
cuFile writes to a NVMe, not into GPU Memory. The current APIs that we have already can write directly into GPU memory. We copy the data out to host memory because that is what Spark wants so it can serialize the data out to disk if it feels like it or keep it cached in host memory. Where it is cached is controlled by the user.
Look into if we can leverage cuFile to write parquet file directly to GPU memory
rapidsai/cudf#7444
The text was updated successfully, but these errors were encountered: