Several months after NVIDIA's initial nForce Pro announcement, boards are starting to trickle onto the market. We've collected a trio of offerings from this initial wave of products, including Asus' K8N-DL, Iwill's DK8ES, and Tyan's Thunder K8WE, and run them through a punishing gauntlet of tests to expose their strengths and weaknesses. Join me as we explore the first nForce Pro 2200 implementations to hit the market and determine which is right for your Opterons.

The nForce Pro 2200

Introducing the nForce Pro 2200
Before we dive into these three nForce Pro implementations, it's worth taking a moment to examine the chipset itself. "Chipset" is usually a misnomer when referring to NVIDIA's core logic solutions for AMD processors, since the recent nForce3 and nForce4 families have been single-chip designs. The new nForce Pro lineup actually consists of two chips, though: the nForce Pro 2200 and nForce Pro 2050. Although they're designed to work together, the nForce Pro 2200 and 2050 are far from a standard north/south bridge pair.

You wouldn't guess it from the name, but the nForce Pro 2200 is essentially an nForce4 Ultra for Opteron. The chip sports a 16-bit/1GHz HyperTransport link, 20 lanes of PCI Express, Gigabit Ethernet with a TCP/IP offload engine, a couple of ATA/133 channels, and four Serial ATA II ports with support for both Native and Tagged Command Queuing. Like the nForce4, the nForce Pro 2200 also supports multiple RAID 0, 1, 0+1, and JBOD arrays spanning both ATA and SATA drives.

Taken alone, the nForce Pro 2200 doesn't really offer anything above and beyond the nForce4 Ultra. However, the 2200 can team up with the nForce Pro 2050 to bolster an already impressive feature set. The 2050 adds an additional hardware-accelerated Gigabit Ethernet MAC, 20 more PCI Express lanes, and another four Serial ATA II RAID ports. The additional PCI Express lanes allow a 2200/2050 combo to provide a real sixteen lanes of PCI-E to each graphics card in an SLI configuration, and NVIDIA's RAID implementation is flexible enough to span arrays across drives connected to both the 2200 and 2050.

The nForce Pro 2050 is more of a companion chip than a traditional south bridge, and it doesn't actually hook directly into the 2200. Instead, the 2050 interfaces with an Opteron's non-coherent HyperTransport link, cutting out the middleman and eliminating a potential interconnect bottleneck between the 2200 and 2050. nForce Pro implementations aren't limited to a single 2050, either. The chipset actually supports up to three 2050s alongside one 2200, although only dual-Opteron systems will have enough spare HyperTransport links for such a configuration.

Comparing the specs
Although all three boards use the same nForce Pro 2200 chipset, each implementation is different, and only one doubles up with the 2050. Let's compare the spec sheets of the three boards to see what they have in common, and where they differ.

Although all three boards support dual Opteron processors and make use of the nForce Pro 2200 chipset, there's more variation between the boards than one might expect. I'll be exploring each board's unique characteristics, quirks, and common traits over the course of the comparison.