[scj13]

NVIDIA published this blog post showing how to run GPU-accelerated pandas in the cloud using Coiled. In the example they do some typical data wrangling (read in parquet files from S3, do some filtering, aggregation, return a result) and saw ~9x speedup.

They used Coiled to manage the cloud infrastructure, a UX-friendly tool for Python devs who want to use the cloud, without needing to directly manage EC2 instances.

Whether you _should_ be running pandas on cloud GPUs is a whole other question (and probably pretty specific to the use case).