waitrequest and readdatavalid

wait_request is asserted by the slave, or by adapter logic created by the sopc system and attached to the slave, when the bus can not accept any more transactions. look at what the maximum number of pending transfers is for both the master and the slave, and that will most likely correspond with when the waitrequest is asserted.

Note that SDRAM has a variable read_request to read_valid length, and 640ns doesn't seem that high depending if that row is already open or not. However, when you pipeline it and have multiple requests posted at the same time it hides some of that latency.