LOSTCIRCUITS

The resurrection of SLI, made possible by PCI Express offers new dimensions for 3D processing - in both performance and price categories whereas the upper midrange of the price range still faces a painful void. That particular nitch is where the GeGorce 6600GT is aiming, especially in SLI configuration. From an SLI standpoint, the 6600GT is especially interesting since its single card performance - even though excellent for the price - offers enough headroom for improvement by doubling the resources. Therefore, the 6600 GT appears, in fact, a better candidate for SLI evaluation than its higher-end brethren, which may or may not run into system and CPU limitations before they can play out their full combined force.

Over the past few months we have been confronted with the wildest arguments about the pros and cons of SLI - even from official sides. Thus, in view of the still prevailing general confusion about some of the issues, particularly regarding the design and functionality / necessity of some of the components, we take a more in-depth look at bandwidth, redirectors and video link components, using multimeters and simple benchmarking approaches. In the end, there is some redundancy in the board design and some recommendations for simplification and, of course, a ton of benchmark results.

As of lately, SLI - short for Scalable Link Interface - has developed into the holy grail of high-end graphics processing. Those old enough will probably remember a different type of SLI - in that case used as acronym for scan line interleaving and being the quintessential graphics solution some 7 years ago – time flies, indeed.

SLI and SLI

The original SLI solution that overnight made 3dfx Voodoo2 the most powerful 3D accelerator used a pair of PCI graphic cards connected by a VGA pass through cable to supply even or odd scan lines from the first to the second card. The effect was that a complete picture was generated by two separate graphics cards, using interleaving and superimposition of the different scanline sets.

Two eVGA GeForce 6600 GT in SLI configuration running in the ASUS A8N-SLI - powered by an Athlon64 4000+, click for larger image.

At the time, scan line interleaving was common anyway, the limited amount of local frame buffer (2-4 MB) was not enough to store the entire picture. Therefore, above a certain resolution, single graphics cards needed to be configured to interleave mode to run two passes for every frame. This was usually necessary above 800 x 600 pixel dimensions – even in 2D desktop mode. The main technical hurdle for this kind of operation was managing of the clock signals in order to provide the necessary synchronization between the interleaved output fragments. .

The acronym used now, namely SLI, is the same as 7 years ago. The technology, however has changed dramatically. Scanline interleaving is a thing of the past, the current lingo refers to Scalable Link Interface - arguably one of these descriptions bent into matching the acronym in order to capitalize on its historic market value.

SLI technology has been more or less covered all over the place. We’ll stick to the very minimalism of what is involved but add a few things – LostCircuits style. Basically, as mentioned above, SLI requires two different components, that is, availability of the data as well as cross talk between the two graphics adapters so that one knows what the other is doing.

Next Page:    => PCIe and the Bandwidth Hype =>