In particular, thanks to Red Hat's Xiao Ni is improved RAID10 discard request handling. The change with a set of five SSDs in a RAID10 array on a test system dropped the mkfs.xfs time for creating an XFS file-system taking 4 minutes 39 seconds to less than 1 second... Quite a noticeable difference in that scenario.
Besides that there is other related RAID10 discard improvements and separately a number of MD RAID5 improvements too that were pushed into the block code for the Linux 5.10 cycle.
The latest block updates can be found via the for-next Git branch.