How much smaller will the file be?

It depends entirely on the data — columns with repeated values compress dramatically, high-cardinality text much less. A substantial reduction is common but we would rather not quote a figure we have not measured on your file.

Which compression should I pick?

zstd is a good default — strong compression and fast to read. Snappy is faster to write and larger; gzip is the most widely compatible with older tooling.

Do I need Python or Spark?

No. DataTray writes Parquet natively.

Free converter

CSVParquet

CSV to Parquet Converter

Going the other direction, usually because the CSV has become unwieldy. Parquet is columnar and compressed, so the same data often lands at a fraction of the size and reads far faster afterwards.

Windows 10/11 (x64) · Free · No account

Drop in your files — Parquet is written with zstd compression by default; snappy, gzip and none are available.

Why convert

What you gain by converting

A CSV stores everything as text, row by row, with no types and no compression. Parquet stores it by column, typed and compressed, which is why analytics tools read it so much faster — and why a large CSV converted to Parquet usually shrinks substantially.

Shrinking a large CSV that is slow to work with.
Preparing a file for a data warehouse or a pipeline that expects Parquet.
Recording the column types once so nothing has to re-guess them.
Partitioning a big export by a column into a directory of parts.

How it works

Convert CSV to Parquet in a few clicks

1
Open the CSV
Delimiter, encoding and header detected and shown for correction.
2
Set the types
This is where the value is — Parquet records the types, so getting them right once means nothing re-infers them later.
3
Choose compression and write
zstd by default. Optionally partition by a column into a directory of parts.

The details

CSV to Parquet, at a glance

Compression: zstd (default), snappy, gzip, or none
Types: Explicit per column and stored in the file
Partitioning: Optional, by a column's distinct values
Python: Not required

Questions