What is the problem the feature request solves?

The publish benchmark for Spark with comet vs spark OSS, the memory for executors when using Comet for acceleration is double (64G vs 32G) the one for Spark OSS. Could we provide documentation to explain how comet impact the memory usage?

Describe the potential solution

Clarifying the need would provide a better user experience when setting up/configuring their Spark job.

Additional context

No response