Concurrent memcopy (all Fermis or only Tesla ?)

I’ve just finished nvidia’s webinar on concurrent kernels and memcopys.

And because there was some misunderstanding I would like to know:

Does all Fermi architecture GPUs support concurrent memcopys (H2D/D2H at once).

Because the conference leader told us that all Fermi cards support concurrent memcopys and the guys answering questions wrote that only Tesla cards support concurrent memcopys and it will be serialized for GeForces and Quadros.

Who’s right ?

Only Tesla 2xxx cards and Quadro Fermi cards have dual DMA engines, that you need to overlap in both directions.

Ok, thank your for your fast response.