North Bridge (cont’d)

QuickSync

NVIDIA’s QuickSync technology is designed especially for asynchronous bus operation (i.e. running the front side bus at 1066MHz while the memory bus runs at 667MHz, or when overclocking one bus more than the other). When the two buses are out of sync with some memory controllers, the additional latency required for receiving data from memory and sending it to the CPU (or vice versa) can actually hamper performance. QuickSync is designed to handle memory requests and data between the FSB and memory in the shortest amount of time, ensuring the best possible performance even when the two buses are out of sync. NVIDIA accomplishes this by accelerating the internal paths between the FSB clock domain and the memory clock domain just as the FSB and/or memory bus speed is increased by the user (either via overclocking or by using faster components).

Each of the SPP’s twin memory controllers can support up to two DIMMs, for a maximum of up to four DIMMs on the motherboard. If 4GB DIMMs are used, the chipset can support up to 16GB of memory. Another interesting aspect of the SPP’s memory controller is that it can run in “ganged” mode, allowing it to act as one 128-bit memory controller even when the DIMM sockets are populated with memory modules of different configurations.

HyperTransport connects the chips

Besides the memory controller, the other main feature of the nForce4 SLI SPP is its PCI Express subsystem. Like the nForce4 SLI chipset for the AMD platform, NVIDIA provides twenty lanes of PCI Express for the Intel Edition of the chipset. Sixteen of these lanes are dedicated to the graphics processor, with three lanes reserved for the x1 PCI-E slots.

Just like the original nForce4 SLI, when running in SLI mode on the Intel Edition chipset, the two x16 graphics slots operate as two x8 slots for optimal performance. The uppermost card is designated as the “Master” card, while the graphics card beneath it is slaved to it. In testing, NVIDIA found this configuration yielded the best performance.

Of course, if the second card is removed, or if you don’t have two SLI-ready NVIDIA graphics cards, the chipset will devote all sixteen PCI Express lanes to the primary PCI-E slot.

Processor support/FSB

nForce4 SLI Intel Edition supports all of the latest Intel processors, including Intel’s dual-core “Smithfield” chips, such as the Pentium Extreme Edition 840, which was first introduced on Monday. Front-side bus speeds up to 1,066MHz are also fully supported by the chipset.

It is in this regard that NVIDIA engineers appear to be particularly proud. Over the past few weeks we’ve been told on multiple occasions just how scaleable the nForce4 SLI Intel Edition was designed to be, with slides cryptically referring to bus speed support of “1,066MHz FSB and beyond”.

However, while dual-core processors are supported, we were told that the chipset doesn’t support SMP. Instead NVIDIA is reserving dual processor support for a workstation-class part, akin to the nForce Professional family for servers and workstations on the AMD platform.

In terms of complexity, the SPP is fairly impressive, being built on a 0.13-micron manufacturing process and containing a whopping 61 million transistors! This is just two million transistors shy of NVIDIA’s GeForce4 GPU.