Physical Integration
To begin with, Quad SLI means a new board SKU in order to get four NVIDIA GPUs working in your system. Called the GeForce 7900 GX2, the new board occupies just one PEG16X slot, but marries two PCBs - each with a G71 GPU and 512MiB of memory - in dual-slot configuration to do 2-GPU SLI on a single board product. Here's a Quad-SLI system in the flesh.
Click the image to see the enlarged version. You can clearly see the two seperate GX2 boards, each with a pair of PCBs. Each PCB has its own power connector, heatsink cooling an NVIDIA G71 GPU and 512MiB of memory. Therefore a Quad SLI system using 7900 GX2s has four G71s and 2GiB total framebuffer space.
Each board needs two power connections, in addition to the power that can be supplied via the single bus slot. You also need to connect the GX2 boards together with inter-card connectors, two per board. I'll explain the reason why in a coming page, suffice to say it's related to the new rendering modes available when four GPUs take part in accelerated 3D rendering.
The boards are big, measuring the width of a full ATX mainboard, so a suitable chassis will be needed to house the pair. Each sub-board is going to ask for about 8.5A at peak load. So you want over 600W on the PSU to be completely comfortable (around 100W per sub-board and around 150W for the rest of the system).
Realistically, with the way power outputs are calculated and supplied on a modern ATX2.0 power supply, that's going to translate into NVIDIA and system integrators advertising ~800W as the minimum needed for Quad SLI, regardless of the actual reality.
You don't need a true x16 SLI system to run Quad SLI, with 'basic' SLI core logic up to the task, but the reality is that system integrators will all integrate the first Quad SLI systems using x16 versions of nForce4 or 5, and that buyers of the boards at retail will likely own proper x16 systems already, or will buy a proper x16 board to go with the cards.
Hopefully by now you'll be wondering how Quad SLI actually works, with four GPUs communicating with each other. Here's how.
