Look at your use of tempdb:
Estimate the size of the temporary tables and worktables generated by your queries.
Look at the number of pages generated by select into queries.
These queries can use 16K I/O, so you can use this information to help you size a 16K pool for the tempdb cache.
Estimate the duration (in wall-clock time) of the temporary tables and worktables.
Estimate how often queries that create temporary tables and worktables are executed.
Try to estimate the number of simultaneous users, especially for queries that generate very large result sets in tempdb.
With this information, you can a form a rough estimate of the amount of simultaneous I/O activity in tempdb. Depending on your other cache needs, you can choose to size tempdb so that virtually all tempdb activity takes place in cache, and few temporary tables are actually written to disk.
In most cases, the first 2MB of tempdb are stored on the master device, with additional space on another logical device. You can use sp_sysmon to check those devices to help determine physical I/O rates.