[gdal-dev] In GDAL Warp What WM Should I Choose

[Thet Lin Thu]

I am trying to stitch a global elevation raster dataset made of 14,000 files
into 1 single 120GB file. I already have the VRT generated and have tried
applying both gdal warp and gdal translate on it with varying results in speed.
Although I am not warping anything (same projection for the input and output) It
seems like gdal warp is faster because I can set a larger memory cache and can
enable -multi for multi core processing.
My question is, what is the correct -wm to set? I have an AWS machine with 700GB
or so of RAM. The total dataset in question is only 120GB (deflated compressed).
I have set GDAL_CACHEMAX at 500GB as well. I am using the most latest  GDAL
3.3.1 in the full Ubuntu docker environment.
I saw in many places about the ambiguities of wm, especially this:-
https://github.com/OSGeo/gdal/issues/1345

That issue was in 2019 and seems like a few things happened around 3.2 and 3.3
in 2020 with regards to warp performance and PROJ.
So based on the assumption that I have any amount of RAM I can procure, is the
best value for wm around "100" still as per the github link?
My apologies if I have interpreted some of the articles I have read in dribs and
drabs across the internet. Thank you for the help.
Thet
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20210826/b4363d97/attachment.html>