A closer inspection...
Back when we reviewed GeForce 7900 GX2, we took the time to take the card apart in order to explain how the technology works from a physical perspective. In essence, the GeForce 7950 GX2 uses many of the same principles as the GeForce 7900 GX2, but the implementation is a little more refined in order to get Quad SLI ready for the mass market.
The card still uses a PCI-Express switch between the two PCBs, but there is only a single SLI connector, shared between the two PCBs. Afterall, both GPUs on the GeForce 7950 GX2 communicate via the on-board PCI-Express switch. When using two GeForce 7950 GX2s in Quad SLI, there is only a need to combine data from the two cards and not between all four GPUs because this is basically SLI on a single video card - the two GPUs act together as one. This is a change from the GeForce 7900 GX2 which - when running in Quad SLI - required two SLI bridge connectors, thus creating one rather complex loop.
As a result of using a 48-lane PCI-Express switch between the two PCBs, there is a need for a bridge chip in order to make the GPUs work together in tandem. Typically, a pair of GPUs working in SLI mode send data across the SLI bridge connector and PCI-Express bus. Obviously, there is no physical PCI-Express bus between the two GPUs on the GeForce 7950 GX2, so NVIDIA had to make its own using a PCI-Express to PCI-Express bridge chip. This is very similar to the AGP to PCI-Express (and PCI-Express to AGP) bridge chip that we have seen on a number of NVIDIA products where the SKU wasn't using the GPUs' native bus interconnect.
The PCI-Express switch between the two PCBs uses a slightly different connection on GeForce 7950 GX2. GeForce 7900 GX2 used a pair of connectors to send data between the two GPUs, now there is only a single connection in between the boards. This is the same size as the larger of the two connectors on NVIDIA's previous dual-GPU video card implementation.