# Intake vs mfdataset

**URL:** <https://forum.access-hive.org.au/t/intake-vs-mfdataset/4445>\
**Category:** Technical\
**Tags:** python, help, intake\
**Created:** [29 April 2025 23:33 UTC](https://forum.access-hive.org.au/t/intake-vs-mfdataset/4445 "2025-04-29T23:33:06Z")\
**Posts on this page:** 1\
**Showing post:** 16

<div class="post-metadata">

**Author:** ![dougiesquire](https://sea2.discourse-cdn.com/flex020/user_avatar/forum.access-hive.org.au/dougiesquire/32/16_2.png) [@dougiesquire](https://forum.access-hive.org.au/u/dougiesquire)\
**Post date:** [5 May 2025 06:23 UTC](https://forum.access-hive.org.au/t/intake-vs-mfdataset/4445/16 "2025-05-05T06:23:24Z")

</div>

> [@dougiesquire](#):
>
> That said, `chunks="auto"` is almost certainly a better starting point than the default NetCDF chunking on disk, so adding `chunks="auto"` to the [COSIMA Tutorial](https://github.com/COSIMA/cosima-recipes/blob/main/Tutorials/ACCESS-NRI_Intake_Catalog.ipynb) sounds sensible to me. @JuliaN or @jemmajeffree, would you like to do this?

Actually, I just remembered I tried [in the past](https://github.com/intake/intake-esm/pull/633) to change the default in Intake-ESM to `chunks="auto"` and it doesn’t work well in the general case because dask auto chunking does not work with object dtype (e.g. cftime arrays).

---

_[View the full topic](https://forum.access-hive.org.au/t/intake-vs-mfdataset/4445)._