What makes a PC tick? Is it the processor, graphics card, RAM, or a plethora of other peripherals. It's all those and more, but none would be much use without a means of connecting said components in an efficient manner. That's where your humble motherboard comes in. Its job is to ensure that all devices can communicate correctly, and the beating heart of any motherboard is the chipset it's based upon. The choice of chipset and motherboard is probably more crucial than any other when deciding upon what kind of system you'd like, so a basic knowledge of which chipsets are available and what they offer you is paramount.
Carrying on from the above, the chipset, or core logic, found on each board is really what makes your PC tick along, and over time chipsets have become increasingly more complex as newer technologies are amalgamated into designs, leading to transistor counts as high as 60 million, which, in turn, has lead to the practical need to separate the chipset into 2 bridges, appropriately named north and south. In modern chipsets the northbridge has handled the I/O between CPU, system RAM, and graphics card, and the southbridge has been tasked with effectively routing peripheral, expansion and storage concerns, with inter-chipset communication undertaken via a dedicated, often proprietory, interconnect.
The first choice a chipset design team needs to make is whether to run with a single or physically bridged chipset (that is, with discrete north and south bridges), and the choice is defined by which CPU it is being designed for. Take AMD's CPUs with integrated memory controllers as an example. By situating the memory controller right on the CPU die itself and eliminating the need for a dedicated memory controller hub, along with a fast HyperTransport interface that communicates with a chipset, designers have been able to harness advanced, low-micron manufacturing processes to release single-chip core logics. NVIDIA's basic nForce4 springs to mind, although competitors such as VIA have stuck with the tried-and-tested two-bridge design, perhaps with one eye on easy bridge replacement as newer iterations are released. The choice for consumer-level Intel-based CPUs is more straightforward, thanks to the need to situate a discrete memory controller hub on a northbridge.
The technology and players
The motherboard chipset business can be a confusing place, then. The simplest method of delineating the features and benefits inherent in a selected number of modern high-end consumer chipsets is to take a look at what each offers, and, from there, why one may be better than another for your purposes.
The market trend
There's an unmistakable trend with modern premium chipsets and it's easy to spot. The rise of multi-GPU support is manifested in no less than five of the six listed chipsets supporting multi-GPU technology in one form or another. All chipsets have at least 16 PCI-Express lanes on their respective northbridges that can be toggled to offer each card, in a dual-card setup, 8 PCIe lanes each. That's still a total bi-directional bandwidth of 4GB/s for each card, and whilst NVIDIA, in particular, is pushing for double the graphics-orientated lanes with its latest iteration of X16 chipsets, dual-card support isn't hindered by the incumbent 8.
What's also telling is that no chipset, right now, supports all multi-GPU technologies. Intel's i975X and ATI Xpress200 CrossFire, naturally, offer CrossFire support and it doesn't take a genius to work out just what compatibility NVIDIA's chipsets carry. Perhaps the most interesting of this sextet is VIA's K8T900, which carries support for S3's multi-GPU MultiChrome technology, an as-yet unproven solution that has obvious disadvantages when directly compared against the established systems from NVIDIA and ATI.
Another recent trend has been to include SATA2, and by inference NCQ, support on modern southbridges. All bar the ATI Xpress200's SB450 support SATA2 and varying flavours of RAID. 4 SATA2 ports is the norm, although premium motherboards usually add a discrete controller with extra ports on offer, and most offer the ability to migrate between one form of RAID and another without sacrificing data integrity. Further, chipset designers have begun adding in southbridges with high-definition audio capability, yet NVIDIA bucks the trend by only supplying its chipsets with the basic AC'97 standard.
The choice
The choice of chipset is largely dictated by your graphics card needs. Want to run two GeForce cards in SLI configuration on an Intel motherboard? You can, of course, but the choice is then limited to NVIDIA's nForce4 Intel Edition. Want to run CrossFire on an AMD platform? Again, you can, but not on NVIDIA's offerings. The lack of an open standard whereby all multi-GPU technologies work on all chipsets ensures that you'll be tied into a particular brand if huge framerates is your primary need.
Carrying on from that, there is no chipset that does it all. Careful evaluation of basic needs will push you, the prospective buyer, down a certain path that may close doors on the use of certain equipment. The above table shows that each chipset has merits that others do not, and vice-versa. Users looking for a high-end chipset to base their LGA775 CPUs around will probably opt for Intel's i975X. It hits most of the right performance notes, will be compatible with all Intel CPUs, and is available today from a number of vendors. Multi-GPU support in the form of CrossFire doesn't hurt matters either.
The situation for AMD-based CPUs is a little trickier. ATI has now released a multi-GPU-compatible chipset of its own, and VIA has weighed in with its K8T900. On balance, though, NVIDIA has shown an aptitude of getting newer technologies (barring HD audio) amalgamated earlier than its competition.
