Hvplot with datashader - replicate tutorial example with larger dataset

I’m replicating one of the geo-examples (Geographic Data — hvPlot 0.8.0 documentation) that also uses datashader (rasterize=True). Using the same dataset (rasm) it takes 1-2 minutes to render the image with a 6-core CPU and 64 GB RAM, so this is already surprising and I’m wondering if it’s normal? I expected this to be much faster.

Additionally, I replicate the example with a different (bigger) dataset (size: 1 x 22179 x 26838) and it takes several minutes to render. The memory usage also goes through the roof, averaging 32 GB for a long time and reaching 52 GB.
The file loaded has a 2.5 GB size on disk and is rendered immediately in QGIS, without considerable amount of memory usage.

See the code and the dataset metadata in the screenshot.

Should I do anything differently?

1 Like