Since Nvidia acquired the rights to the original Scalable Link Interface (SLI) technology as deveopled by 3dfx (then called scan line interleaving), the concept of using multiple graphics card has been somewhat forgotten about. Only a brief sojourn into the realm of dual chips on a single card by ATI has deviated from the normality of a single graphics processor. However, with the introduction of PCI Express, Nvidia looked again at SLI – now called scalable link interface - and subsequently released their a new dual-card technology in it’s wake.
Similar to the preceding 3dfx SLI, which used two PCI slots to house Voodoo2 cards, Nvidia’s new SLI model uses two PCI Express x16 slots on a compatible motherboard. In old systems, PCI slots were numerous, with most boards having between three and six slots available. On modern boards, however, there is often only a single PCI Express x16 slot, so to use Nvidia SLI, a special SLI-compatible motherboard is required. Nvidia SLI does not work on AGP interface and was designed solely around PCI Express architecture.
Nvidia’s SLI requires that both the graphics cards use the same graphics processor, although the actual board manufacturer can be different. This means that using a Geforce 7600-based card can only be used in a SLI configuration with another similar 7600 card, for instance, although one could be manufactured by Asus, and the other by Gigabyte.
Depending on the speed of the graphics cards, the faster clocked card will require to throttle back it’s speed to match that of the slower card, while if one card has a greater amount of onboard memory, then the card will disable the extra RAM in order to create an equal quantity with the sister card.
PCI Express x16 slots have 16 PCI Express links, known as lanes. Each of these lanes is potentially capable of communicating 500MB of data per second, which consists of 250MB from the motherboard to the graphics card and the same amount in the opposite direction. In contrast, the original PCI could only deliver around 100MB. This means that a PCI Express x16 slot has the ability to transfer 8GB of data per second.
Earlier models of Nvidia SLI saw the capacity divided between both cards – each receiving eight lanes and a top transfer speed of 4GB/sec which had to be controlled via a setting on the motherboard, which often consisted of moving a jumper pin or inverting a small card from a slot on the motherboard between the two graphics cards. In earlier models, it was also necessary to link the two cards together in a direct connection. Nvidia claims this connection is capable of transfer speeds up to 1GB/sec between the two cards.
In a Nvidia SLI system, there is a card hierarchy in place: one card is the master, the other is the slave, and the two work together in three ways. Whenever you load a game onto your PC, the driver which controls when the graphics cards should enter SLI mode examines the NVAPPS.XML file which contains information on the best SLI mode to run for the game in question.
The first mode is called ‘split frame rendering’ or SFR, which works by dividing the frame to be rendered into two parts. Instead of merely splitting the screen in half, however, SFR works to ensure the frame is split into two equal portions, so that both cards have an equal workload. For example, in an outdoor area of ‘Crysis’ where there might be a large expanse of sky at the upper end of the monitor, then SFR will divide the screen at a lower point where there may be much more complex rendering. Once the frame has been completely rendered, the slave card sends its image to the master card which combines the two images together and sends the completed frame to the monitor for display.
The second mode is called ‘alternate-frame rendering’ or AFR. This mode works by having each graphics card produce alternating frames of a game: one card renders the odd-numbered frames, while the other renders the even-numbered frame. Frames rendered by the slave card are sent to the primary card for display via the SLI connector. Because successive frames in a 3D game, such as a first-person shooter, are often only slightly different, AFR mode ensures that the workload is evenly balanced between the two cards.
The third mode is known as SLI anti-aliasing and works to an entirely different instruction. Whereas AFT and SFR concentrates on using both cards to render frames, SLI anti-aliasing concentrates more on using the two graphics processors to produce a more thorough anti-aliasing overlay of the finished image, applying either 4x or 8x anti-aliasing using different sample patterns. Anti-aliasing is the process of removing jagged edges from the 3D image in order to produce a more clear and realistic image. With each card performing anti-aliasing independently, this means that it is possible to perform greater levels of anti-aliasing than would be possible through a single card.
However, Nvidia haven’t stopped there and since the initial launch have refined and improved its SLI model. Nvidia’s first major update allowed owners of budget and mid-range Nvidia cards, such as the 6600 series, to run SLI without the need of a connector to link the cards together. Instead, the image data that would normally have been sent across the connector would be sent via the motherboard. Following on from this update, Nvidia then lauched it’s nForce 4 SLI x16 motherboard chipset, which provided more PCI Express x16 lanes and allowed for both slots to run at full speed in SLI mode rather than having to be split between the two slots.
Perhaps the most significant update to Nvidia’s SLI model was the introduction of QuadSLI, which uses two PCI Express x16 slots but allows four graphics processors to be used. However, such a configuration requires a special motherboard and graphics cards to be utilised properly. Special Geforce 7950 GX2 graphics cards, which each contain two graphics processors and own memory and are designed for Quad SLI, can be linked together to use several new SLI modes, including a method to produce alternate frames, with each frame being split across graphics processors using SFR and providing anti-aliasing upto 32x.
As graphics technologies continue to evolve, and games programmers continue to place added demands on graphics card hardware, it’s likely that SLI will continue to develop and with good reason. With unparalleled graphical muscle afforded by interlinking multiple cards, providing your motherboard is compatible with SLI, it provides an inexpensive alternative to boosting the graphical prowess of your PC.