Describe the bug
The "Full pipeline" benchmark table in the README has stale numbers for the CPU rows.
Current README:
| Backend |
Time |
| NumPy |
2.7s |
| rioxarray (GDAL) |
418ms |
Measured now (3600x3600 Copernicus DEM, read + reproject EPSG:3857 + write GeoTIFF, median of 5 runs):
| Backend |
Time |
| NumPy |
~784ms |
| rioxarray (GDAL) |
~749ms |
The NumPy number is off by 3.4x, likely from before the reproject and geotiff reader were optimized. The rioxarray number shifted too, probably just hardware/version drift.
GPU rows (CuPy 348ms, Dask+CuPy 343ms) can't be verified without a GPU, so leaving those alone for now.
Expected behavior
README benchmark numbers should reflect current performance.
Describe the bug
The "Full pipeline" benchmark table in the README has stale numbers for the CPU rows.
Current README:
Measured now (3600x3600 Copernicus DEM, read + reproject EPSG:3857 + write GeoTIFF, median of 5 runs):
The NumPy number is off by 3.4x, likely from before the reproject and geotiff reader were optimized. The rioxarray number shifted too, probably just hardware/version drift.
GPU rows (CuPy 348ms, Dask+CuPy 343ms) can't be verified without a GPU, so leaving those alone for now.
Expected behavior
README benchmark numbers should reflect current performance.