Here's How SLI Works
Motherboards using SLI-compatible chipsets have two x16 PCI-Express connections instead of one. But since the chipset logic provides only a total of 16 PCIe lanes, motherboard makers divide up the 16 into eight lines per physical x16 slot. Although this impairs the theoretical bandwidth, in practice it has no noticeable impact on performance.
In order for the two graphics cards to work in SLI mode, they must be connected to each other by means of a jumper. This jumper is usually included with the motherboard, since the distance between the two PCIe slots is left to the motherboard manufacturer. Once that is done, when the computer is started up, and the NVIDIA driver recognizes that it's SLI compatible. After activation in the driver, a further restart is necessary, and then you can really get going.
The computing load is distributed through a horizontal division of each individual frame. Of course it can happen that numerous complex objects are located in one half of the picture while the second half has simpler elements that can be rendered quickly. In these cases, the division ratio between the two image halves can be adjusted dynamically, in order to achieve the most balanced utilization possible - and thus, the highest performance.
SLI, however, also has a disadvantage for developers and graphic designers: it only works on one display, at least for right now. This means anyone using two monitors or TFT displays has to deactivate the SLI mode for normal work in Windows and reactivate it when needed, which always requires a restart. That's naturally a major disadvantage when running a P2P client, since it would have to search for its connections over and over again.
Alternatives? ATI Multi VPU
NVIDIA's archrival ATI will be introducing its own technology for powering two graphics cards at once at Computex at the end of May. Called CrossFire, the term was chosen as both a pithy challenge to NVIDIA as well as a snappy product name. ATI's solution is touted as being far more flexible, because the solution for connecting the two graphics cards is more user-friendly. Furthermore, in ATI's setup, the two cards do not have to be identical and according to our findings, the possible acceleration could include all 3D applications.
That all sounds enticing; however, the improvements over NVIDIA also involve optimizations that ATI has introduced in order to combat its time disadvantage vis-à-vis NVIDIA. After all, SLI has been around since the end of 2004.