Multi-GPUs

If you're the least bit interested in graphics cards, we're sure that you've heard the terms SLI and CrossFire bandied about recently. Touted as a means of achieving maximum 3D performance by, effectively, using two or more graphics cards in tandem, multi-GPU technology is here to stay. With that in mind, let's take a closer look SLI and CrossFire; the two competing multi-GPU solutions from NVIDIA and ATI Technologies, respectively.

A few words about how and why multi-GPUs technology can and does work. Due to the inherently parallel nature of graphics cards, GPU designers have sought to broaden the processing ability of their respective cards. Right now, taking it down to the most basic level, NVIDIA's GeForce 7800 GTX GPU has 8 vertex shaders, uses 24 parallel pixel/fragment processors (six quads, actually), and offloads then to 16 ROPs. These big numbers allied to a 430MHz default core speed, produces some mighty fillrate and shading power. Thinking about graphics, wider is most definitely better and faster. The physical cost of housing all the requisite parts of GeForce 7800 GTX's wide-and-fast architecture, however, comes at a cost of no less than 300m transistors. Both NVIDIA and ATI would love to design super-wide (and super-fast) GPUs, but power draw, associated heat, and pure cost would make them financially untenable right now.

So how do you make something that's already fast that bit faster? From reading the above, the answer is kind of obvious. The key word is parallelism, and both 3D heavyweights have architected multi-GPU technologies that attempt to efficiently harness the power of two or more present-day graphics cards and output the combined rendering as one. NVIDIA's SLI (Scalable Link Interface) and ATI's CrossFire both aim to do just that, yet do so in slightly differing manners. We'll dissect each and you can then make your own mind up on which is best.