The slow random read figures are because the Agnostics benchmark follows a trimmed-down version of the SD A1/A2 test procedure - or at least as close as I can make out. The spec calls for a 10 minute continuous operation of each phase, which is kinda dumb when you want a quick evaluation of storage performance. So it's time-limited for brevity, but when run in this manner, a majority of sequential and random writes will have hit the SLC cache (hence the very fast sequential write number) and when the random read test is performed, the drive is busy streaming those out to the backing store.Very different results than those announced... ???leepspvideo has done a video:
https://www.youtube.com/watch?v=uwKJcVFZ-xk
The USB drive random read specification is for "cold" sequentially written data, which closely aligns with use-cases such as OS boot.
Statistics: Posted by jdb — Sat Jan 31, 2026 10:21 pm