If I have servers with good, modern GPUs, is there a way to configure Trino to leverage that?
Answer: no, Trino doesn’t support using GPUs, nor is it planned.
More details:
Trino runs as application on the Java Virtual Machine. Specifically it is using Java 17. Upcoming version of the JVM and the related libraries have improved support for GPUs and leveraging these is being considered for future releases.
However keep in mind that GPUs and the parallel processing that is accelerated by them is not a typical workload for query processing as done by Trino, so the benefits will be small and isolated to specific use case.
On the other hand there are constant performance improvements going into Trino as can be seen from the release notes, and we have some dedicated projects and refactorings in progress for performance improvements. Examples are the removal of the old Hive/Hadoop legacy libraries, the implementation of a custom Parquet reader and writer, and various measure in Project Hummingbird (see Project Hummingbird · Issue #14237 · trinodb/trino · GitHub)