How do I optimize my usage of oclvanitygen?

I’d like to extract the most performance possible from oclvanitygen. Can anyone tell me what the best practices are with this tool?

Examples may include:

• Use operating system X, or video card Y
• Ensure that the file specified with the -F option has more than 7 characters per row
• Ensure that there are no more than Z quantity of lines in the file
• Set a custom grid size with the -g option (what is this?)
• How many work items per thread ( -w option)
• What custom options should I use (-d option)