The World’s First 32nm Processor Will Include (Low-End) Integrated Graphics

Intel is keen to bring its 32nm CPUs to market, despite the weakness of the economy. In order to do this, they have delayed and replaced the first proposed 2 core Nehalems, Havendale (desktop) and Auburndale (laptop), with 32nm versions called Clarkdale and Arrandale, due by year’s end. It is possible they will be marketed as Core i3 or Core i4.

The first mainstream Nehalems, 45nm, 4 core, called Lynnfield and Clarksfield, are due in Q3, likely under Core i5 branding. A new Core i7, codenamed Gulftown, at 32nm with 6 cores, is in the works for release early next year. The mobile Arrandale will be discussed in another article soon.

This article deals with the desktop Clarkdale processor.

Clarkdale will be a 2 core, Nehalem microarchitecture CPU - the first to be made on a 32nm process. Like the rest of the Nehalem family, Clarkdale uses Hyper Threading, which allows each core to handle two threads. The cache design also appears to be the same, at least on a per core basis, giving the Clarkdale 4MB of shared L3 cache.

It also has an integrated, DDR 3, memory controller as does the rest of the Core iX branded chips. The Core i7 has a triple channel controller, though, while Clarkdale (and Lynnfield) are dual channel. It also brings the graphics controller onto the CPU (like Core i5, but not Core i7). Even the integrated graphics are brought onto the processor in the Clarkdale, about which Intel is pretty excited, though the silver cloud definitely has a grey lining. More on Intel’s graphics on the CPU later.

Further differentiating it from other Nehalems, Clarkdale will use its own socket, LGA 1155. We knew before Core i7 launched that later, and lesser, Nehalems would use a different socket than the X58’s LGA 1336. Now, we have learned that upcoming 2 and 4 core Nehalems will require different sockets: LGA 1156 for Lynnfield; LGA 1155 for Clarkdale.

With the memory and PCI-e controllers on the CPU itself, and even integrated graphics in the Clarkdale’s case, the Clarkdale and Lynnfield processors perform most of the functions traditionally assigned to the northbridge. As a result, the Ibex Peak platform those CPUs use, (which will come to market as Series 5), does not have one.

Because the Ibex Peak platform has the graphics speaking directly to, or even on the CPU itself, it only needs to handle CPU communication to the southbridge (which is now essentially the whole chipset). As this connection doesn’t need as much bandwidth as it would if it carried graphics the way it does on the Core i7/X58, the Ibex Peak platform uses Direct Media Interface (DMI). This is instead of the Quick Path Interconnect on the X58 (Tylersburg) platform.

Intel’s DMI is currently used to connect the north and southbridges at 2 GB/s, on both the Nehalem Core i7/X58, and Core 2/Series 4 chipset. Because the Clarkdale has onboard graphics, it needs a way to get these to the display output. A second connection between the CPU and southbridge, which Intel calls Flexible Display Interface, does this.

The diagrams above show, from left to right: a traditional Series 4 platform, with FSB connecting CPU and northbridge, and DMI connecting northbridge and southbridge; a Core i7/X58 or Tylersburg setup, with QPI connecting CPU to northbridge, and DMI connecting north and southbridge; finally, a Series 5 or Ibex Peak platform for Clarkdale, with no northbridge, and DMI and FDI connecting the CPU and southbridge directly.

On the next page, performance and prices.