fyi:
To further clarify some point that Masoud made . . .
2- 6500 has a shared bus back plane and a crossbar backplane.
If a line cards uses the shared backplane, maximum 32G for all line cards.
I believe the 32G (classic bus) is 16G duplex.
If line cards uses cross bar, each line card can have maximum 8G (CEF256 card) or 20G (CEF 720 with one connection to cross bar. check my picture) or 2*20G(two connections to cross bar) per slot.
There's one dCEF256 card with dual 8G fabric connections, the 6816, others, besides having the single 8G fabric connection, also connect to the classic bus. (NB: I believe, by default, CEF256 cards will use their classic bus connection for egress traffic, they need to be configured to prefer to use their single 8G fabric connection. I further believe, classic bus cards communicates with fabric cards by traffic "hopping" through the supervisor. [Obtaining optimal performance, when there's both classic bus and CEF256 cards in a chassis, depends on your traffic flows. I.e. whether it's better to have your CEF256 cards use their fabric connection, or not.])
The sup2T, in an -E chassis, support cards with dual 40G fabric connection. It also supports dual 20G in all ports of the 6513-E chassis.
Fabric connection are duplex. So, for example, a 6704 4 port 10g card's dual 20G fabric connections, appears to provide just enough bandwidth to not block the card. (NB: years ago, Cisco noted only the [discontinued] 6702 dual 10g card, with its dual 20G fabric connections, was true wire speed.)
From the above, you might now fully appreciate, getting the most out of a 6500, or using the "right" components for what's needed, can require a good understanding on how all its hardware options work together. So far, not discussed, even the capabilities of line cards differ at the port level. How they, for example, impact usage, can be even more important than how the card "connects" to the chassis, and whether or not it has a DFC.
