I am investigating the same thing. My GTP is setup to use:
--line rate of 312.5 MHz
--No 8b/10b
--Oversampler is of course used
--RECCLK is NOT used
--dynamic sampling is used
Reading the user guide it seems that there is no way to set the receiver to do both oversampling AND lock to ref (page 139 - PMA_RX_CFG settings). I need this functionality since I am using a slow line rate AND my data can have LONG runs of no data transitions. The RX signals can stay idle for an indeterminate amount of time before any useful data comes across. Therefore I don't want to use a recovered clock - since there might not even be one!
I had a V4 MGT setup to do oversampling AND it locked to the reference clock. This worked fine for my protocol. Unfortunately the line rate is too low and not supported in the MGT's. Even though it worked fine for our eval board, it may or may not work on other FPGA's. We got a V5 eval board to test the GTP since they can run the line rate we need.
Perhaps I am not understanding the GTP user guide (I admit I haven't scrutinized it yet - I have just started this project) and there is no way to do what I need.
I am currently using the wizard-created instantiation/attributes and I am seeing the data recovered AFTER some number of bits (like the above post). This is analogous to losing frames of data.
I have a solution that uses 4X oversampling in the V4 fabric, but we wanted to evaluate another fallback option. Is it possible to use the V5 GTP for this protocol?