A whole new set of chips
The nForce 780a SLI's motherboard GPU sits at the heart of a brand new MCP chip fabricated using 65nm process technology. Of course, this just wouldn't be a new nForce without a little love from Nvidia's silicon back-catalog, so the chipset also includes an nForce 200 chip first introduced with the nForce 780i SLI. The nForce 200 provides the bulk of 780a's PCI Express 2.0 connectivity, serving up a total of 32 lanes split between one x16 link and a pair of x8s for three-way SLI. These lanes could also be split evenly between four x8 links, but Nvidia doesn't yet offer an SLI configuration that requires four slots.
The nForce 200 packs a couple of features designed to improve performance with SLI configurations. The chip supports a posted-write shortcut that allows commands to be passed between graphics cards without hitting the processor or system memory, and a broadcast function enables a single command sent from the CPU to be replicated across all graphics cards, saving the CPU from having to generate and send multiple commands.Both of these capabilities should reduce the number of commands needed to manage multiple graphics cards, decreasing congestion on the chipset's link to the CPU. They should also trim traffic on the 16-lane PCI Express 2.0 link that connects the nForce 200 with the 780a SLI MCP. This link delivers a full 16GB/s of bi-directional bandwidth, which Nvidia says is plenty even for three-way SLI. However, it is worth noting that 16GB/s of interconnect bandwidth is only half of what would be needed to saturate all 32 of the nForce 200's PCIe lanes.
In addition to the 16 lanes it has running to the nForce 200, the nForce 780a SLI MCP has three lanes of PCIe reserved for x1 slots and peripherals. That still leaves the 780a a few lanes short of the 790FX's effective 38 lanes of PCIe 2.0 connectivity, but the gap closes to only two lanes when you consider that the AMD chipset will lose at least one lane to a Gigabit Ethernet controller. The 780a SLI has an integrated GigE MAC, so auxiliary networking controllers aren't required. This single GigE controller is a bit of a step back for Nvidia, though; previous high-end nForce chipsets have included dual Gigabit Ethernet controllers.
To make chipset specs a little easier to digest, we've distilled them into a handy chart comparing the 780a SLI with AMD's 790FX.
|AMD 790FX||Nvidia nForce 780a SLI|
|Processor interface||16-bit/2GHz HyperTransport||16-bit/2GHz HyperTransport|
|PCI Express 2.0 lanes||38||35|
|Chipset interconnect||PCIe 1.1 x4||PCIe 2.0 x16|
|Serial ATA ports||4||6|
|Native Command Queuing||Y||Y|
|Max audio channels||8||8|
As you can see, both have a HyperTransport 3.0 processor interface and support for their own brand of multi-GPU graphics, be it CrossFire or SLI. Don't pay too much attention to differences in chipset interconnect bandwidth, though. The 790FX uses a traditional north/south bridge layout, while the 780a's MCP and nForce 200 chips are arranged in almost an opposite configuration. Nvidia needs all the interconnect bandwidth it can get to keep the nForce 200's PCI Express lanes fed, but the peripherals integrated into the 790FX's SB600 south bridge component aren't nearly as bandwidth-hungry.
The 790FX's older SB600 south bridge definitely looks overmatched in the storage department, where the nForce offers two more Serial ATA ports and support for RAID 5 arrays. Nvidia has added AHCI support to its latest storage controllera feature that AMD still hasn't managed to get working right with the 790FX.
12 USB ports round out the nForce MCP, giving the 780a another edge over the 790FX. The 780a also has full support for the Enthusiast System Architecture specification that Nvidia has proposed to the USB-if governing body.
For all you core logic geeks, here are a couple of close-up nudies of the 780a's MCP and nForce 200 components.