I have a directory of 467 aerial photos that I want to merge into a geotiff (or some other way of looking at entire dataset). My problem is at the moment my workflow takes a long time - the final merge step takes about half an hour. This is problematic because I have millions of photos in total, this is just a sub-set. My current workflow is:

1. create worldfiles for the jpgs using georeference info I have
2. gdal.Warp to create .vrt files for each jpg
3. gdal.BuildVRT to make a single merged vrt file
4. gdal.Translate or gdal.Warp to make the final merged geotiff (Warp is slightly faster than translate but not by much)

Steps 1-3 do not take long, it's the final step 4 that is problematic. I have tried some things like setting NUM_THREADS, increasing the GDAL_CACHEMAX, but they haven't had a big effect.

Are there better ways to do this? Are there better ways to achieve my overall aim, which is viewing all the photos at once? Geotiff is useful because overviews makes them quite quick, once the file is generated...