Hello, I am trying to convert a full JPEG2000 image (Sentinel 2) to a GeoTIFF (10000x10000 pixels).
My config is the following: GDAL 3.1.0, released 2020/05/03, freshly compiled on my computer (Ubuntu 18.04). I use the following env variables: OPJ_NUM_THREADS=8 GDAL_NUM_THREADS=1 GDAL_CACHEMAX=1000 The command line is: gdal_translate -of GTiff -co COMPRESS=NONE -co TILED=YES T15RTN_20190328T170201_B02.jp2 /mnt/ramdisk/test-openjpeg.tif As stated in its name, /mnt/ramdisk is in memory, so there should not be any I/O lag. I observed the following pattern: - decompression at full speed, using all my laptop cores (about 5sec); - a single thread starts, to write the GeoTIFF (about 5sec as well). I would have expected the GeoTIFF creation to be almost instantaneous given the fact that there is no compression. If I set GDAL_NUM_THREADS > 1, I do not manage to reach full CPU speed (all cores are used, at about 50%), and it takes about 10 secs as well. If I use mapbox/rasterio to read the raster into memory then to write it back to a GeoTIFF file, the whole process takes 5s, the GeoTIFF write being almost instantaneous. I am puzzled by this behavior... Is there something wrong with my GDAL settings ? Is there a way to speed up the GeoTIFF creation ? Why is it faster using python ? Thank you! Thomas
_______________________________________________ gdal-dev mailing list [email protected] https://lists.osgeo.org/mailman/listinfo/gdal-dev
