Ibex Peak – The Platform for Intel’s Mainstream Nehalem Processors

Welcome to Ibex Peak

Starting with Lynnfield in the third quarter and Clarkdale as we move into next year, Intel will be bringing Nehalem processors to the mainstream desktop market. Their Core i7 enthusiast-oriented chips, launched last November, are great performers and reasonably priced considering their power and newness. They have failed to gain the popularity Intel was hoping for, though, but this is likely related to general economic issues.

Intel is now preparing Nehalem’s – admittedly reduced in grandeur – introduction to the mainstream. This includes a new platform, called Ibex Peak, which is quite different from not only the preceding Core 2 platforms, but even the X58/Tylersburg platform which runs the Core i7 Nehalems.

X58 and Core 2 Compared

Intel 4 Series Platform with FSB and DMI

Core 2, and previous Intel architectures, used a northbridge that contained a memory controller and a PCI-e graphics controller. It communicates with the CPU using the Front Side Bus, and with the Southbridge (which controls things like hard drives and USB ports) via DMI, or direct media interface. This configuration, used on motherboards with Intel’s LGA 775 socket, is pictured at right.

Core i7 moved the memory controller onto the CPU itself. It communicates with the X58’s graphics controller (now the north bridge’s main job) by Quick Path Interconnect (QPI) at up to 6.4 GB/s. The north and south bridge are still connected by DMI. This configuration, which is pictured below, uses an LGA 1336 socket.

Intel X58 Platform with QPI

X58 and Ibex Peak Compared

Ibex Peak with FDI for Clakdale

Lynnfield, likely to be called Core i5, and Clarkdale CPUs, will use the Ibex Peak platform. In addition to the memory controller being on the CPU, as in Core i7, the graphics controller also comes on board the processor. In the case of the 32nm Clarkdale, so does integrated graphics functionality.

Since the work that was done by the northbridge is now being done on the CPU, the Ibex Peak platform doesn’t have one. And since the processor doesn’t need to bus graphics data to a bridge, it doesn’t use QPI. The processor communicates directly with the southbridge, using the lower bandwidth (2 GB/s) DMI. We see this at left. We should also note that while both CPU families have integrated DDR 3 memory controllers, the X58’s above is triple channel, while Ibex Peak’s is the traditional dual channel.

In addition to splitting the enthusiast Core i7/X58 platform from the mainstream Ibex Peak platform, Lynnfield and Clarkdale each get their own version of the Ibex Peak platform.

Hard to Follow the Plot: Twin (Ibex) Peaks

The main theoretical difference between Ibex Peak in Lynnfield and Clarkdale is the absence or presence, respectively, of what Intel calls Flexible Display Interface. Since Clarkdale has graphics integrated on the CPU, it uses FDI to get the display signal to the southbridge, whence it can be sent to an output on the motherboard where you plug in a monitor. Havendale CPUs do not include graphics, and are meant to deal with graphics cards (which have their own display output), so they don’t have FDI.

The main practical difference is a biggie. It looks like Lynnfield will use an LGA 1156 socket, and Clarkdale will use an LGA 1155 socket. Chips using an LGA 1157 will be sold under the Xeon brand as server products. Thus we don’t worry about them here.

Further Divisions: Kings Creek and Piketon

In addition to the integrated graphics/no integrated graphics split, which have different sockets but share the Ibex Peak name, is a split between the home, and office chips. The former will be called Kings Creek, and the latter Piketon. The main difference appears to be in remote administration features: stuff that makes a System Administrator’s job easier like being able to update the Service Pack on dozens of computers at a time.

Piketon will continue Intel’s chipset naming conventions, coming to market as Q prefixed, meaning business oriented, chipsets. The second character, a numeral, indicates the series. The 4 series was the last for Core 2; the Tylersburg X58 and Ibex Peak chipsets are collectively the 5 series. For reference, the X instead of Q is for eXtream, and the final 8 indicates, somewhat redundantly, the top of the line.

Next Page: Q55, Q57, and the P, Q, and H 5 Series

Nehalem for Business: Q57 and Q55

While future product plans are never carved in stone, and many deductions are based on data not confirmed by Intel, now we are getting closer to conjecture:

Piketon chipsets will likely be Q5n, n being a numeral to indicate the board’s features relative to other boards sharing the first two characters. Traditionally, 5 was the best of the mainstream, but Intel has slotted in a 7 designation below the X58.

Q57 and Q55 motherboards should be out next year, coinciding with the planned release of the Clarkdale processor. If we assume a repetition of historical patterns, an economy Q53 chipset will become available eventually. Then again, Intel may just go with 5 and 7.

Since business computers often don’t need dedicated graphics, most (if not all, considering the planned timing of their arrival), of these boards will have LGA 1155 sockets for processors that integrate graphics, and the associated FDI path to the southbridge.

That could be a problem for business users because, though business applications rarely demand dedicated graphics, they can sometimes benefit from multi-core processors. Currently, the only LGA 1155 CPUs in the works are two-core, with the four-core Lynnfields (described here) starting off in LGA 1156.

Nehalem for Mainstream Home Use: Series 5 P H Balance

Returning to the Kings Creek side of Ibex Peak (sounds like a nice walk), Intel hangs on to the P prefix for its mainstream boards, and as we mentioned above, chipsets for Nehalem are the 5 series. They are also going with the 5 and 7 suffixes for the time being. Chipsets with integrated graphics (other than the Q business line) used to have a G prefix, but with those graphics now on the CPU, Intel has dropped that.

The integrated graphic based chipsets will be H5-, H possibly referring to the Havendale chip with which it was to debut, until it was dropped in favour of waiting for the 32nm Clarkdale (described here). They should have used F, for Flexible Display Interface. This would require the assumption that FDI, and by extension the LGA 1155 socket, will only appear on the H5s (and Q5s); and all P5 boards will use LGA 1156 sockets and CPUs without integrated graphics.

We’re tempted to make that assumption since it appears that FDI will only be on Q and H boards, and P boards have never had integrated graphics. Also, the P55 will be the first Ibex Peak to market, apparently coinciding with the non-graphics bearing Lynnfield’s arrival. Then again, maybe we are just desperate to make sense of all these sockets; and are trying to overly simplify things by associating X58, P5-, and H5-, with LGA 1336, 1156, and 1155 respectively. It would make sense, but that is no guarantee.

It looks like the H boards will have one PCI-E 2.0 x16 slot for graphics, and the Ps will have two, though only at x8 each when using two graphics cards. PCI-E lanes are explained here.

Which, If Any, to Buy?

Intel isn’t making upgrading to Nehalem all that tempting this year. These chips, presuming they maintain Core i7 type performance (scaled down with price, of course), are certainly impressive. But most of Nehalem’s benefits have to do with multi threaded applications. Day to day use and even games won’t really see a pop from this for the time being.

Nehalem shines running the type of professional software you would usually find on top end PCs or even workstations. These machines obviously would be better fits for the top of the line Core i7.

Many users would love to upgrade to Nehalem without meeting the Core i7/X58’s entry fee. That leaves the Ibex Peak platform. Most people who rely on integrated graphics, in the home or office, don’t run out and buy the latest tech, particularly in a sour economy. Factor in that the graphics included in Clarkdale are a warmed over XMA 4500, and we can largely discount the popularity of that chip and the LGA 1155 socket (first pictured here): first 32nm CPU in the world or not. At least until some more processors are announced.

That leaves Lynnfield, which will debut earlier than Clarkdale (roadmaps here), and be quad-core, but will be 45 instead of 32nm. While not having the northbridge bottleneck will be an improvement, most people will hold off on an upgrade until 32nm, LGA 1156 socket CPUs, are available, or at least announced. Again, Intel is fighting a tight fisted economy.

Some might be put out enough by the lack of full bandwidth for multi-graphics card setups, or turned on enough by the triple channel memory, and consider a full on i7/X58 rig. Particularly if we consider that Intel is planning Gulftown: a six-core, 32nm CPU, that we hope will fit in existing LGA 1336 motherboards, due early next year. That gets your budget back up though, and we are back to the lack of consumer demand.