A community-garden inventory exports a wide parquet of daily soil moisture and plot IDs. The analyst is about to open it in pandas and then move it to the GPU. How should that table land on a single GPU?
Select an answer to reveal the explanation.
Short Explanation
A wide parquet that one GPU can hold should land with cuDF in a single ingest step. Opening it in pandas first is a host detour. Dask cuDF waits until the table exceeds that GPU, and TensorRT-LLM does not read moisture readings.
Full Explanation
cuDF can read common tabular formats directly onto one GPU. Opening the parquet in pandas and then copying it across adds a host detour that the official ingest path does not need. Dask cuDF waits until the table exceeds that GPU. A CUDA kernel or an inference optimizer is the wrong layer for a single-device read.