Curious to see what powers the Nehalem and Core i7 architecture? We reveal the tech behind the fastest Intel-powered computer on the planet.
Intel’s latest, fastest and most feature-rich generation of desktop CPUs, formerly codenamed Nehalem and now branded Core i7, has finally arrived (and we reviewed it in more detail, with benchmarks here).
After more than a year of anticipation we finally have hardware to test, but can it live up to the hype?
At first glance, the three new chips that make up the initial Core i7 family look disappointingly ordinary. They’re all quad-core parts, manufactured on a 45nm process and running at speeds from 2.66GHz up to 3.2GHz.
On paper, nothing seems to set them apart from the existing range of Core 2 Extremes. But as soon as you set eyes on one it’s apparent that Core i7 represents a break with the past: the new CPUs are larger than their forebears in the Core family, and to accommodate them Intel has introduced its first new desktop socket in four years. LGA 1336 (also known as Socket B) uses the same ZIF design as the familiar LGA 775 architecture, but, as the name suggests,
it incorporates many more contacts.
One reason for this expansion is that with Core i7 the CPU takes over memory controller functions that were previously handled by the north bridge.
In place of the old front side bus it now has a dedicated high-speed connection directly to the system RAM, just like the HyperTransport used by AMD processors. Intel calls its new bus the QuickPath Interconnect (QPI).
RAMping up
It’s worth noting that Core i7 is Intel’s first DDR3-only platform. The memory controller is now on-die – as with Phenom – which makes for a bigger chip and a bigger socket. Curiously, it uses a three-channel memory controller, so you can expect to see DDR3 DIMMs start appearing in three packs.
You don’t have to install modules in threes, but if you do you can expect a small performance benefit, as with existing dual-channel controllers.
Instead of the familiar 775 pin from Intel’s Core 2, Core i7 has a whopping 1366 pins – the additional pins are needed for the three 64-bit DDR3 channels. Because the CPU now includes pins for the memory channels, future changes to memory will mean changes to the socket and number of pins.
As a first example of how the memory channels will affect pin and packaging of CPUs, expect a two-channel version of Nehalem later to create lower-priced products, using the LGA1156 socket.
Multicore and cache
 |
| Multicore and cahe |
These big architectural changes are backed up by some less visible, but no less significant, advances. Until now, Intel’s quad-core processors have been constructed from two dual-core dies, but now Core i7 brings together four cores on a single die. It’s also Intel’s first processor design to use an L3 cache, shared between all four cores.
The L3 cache in the initial Core i7 processors is 8MB, although its size will vary depending on the number of cores. Multi-threaded applications that are being worked on by all cores will enjoy the large, shared L3 cache. The L2 cache now acts as a buffer to the L3 cache so you don’t have all of the cores banging on the L3 cache, requiring tons of bandwidth.
When it comes to cache, Core i7 once again comes out resembling AMD’s Phenom more than its Intel predecessors; although, as usual, Intel has been far more generous with cache RAM than its rival, equipping the i7 with 8MB of L3 while AMD’s quad-core parts get just 2MB.
The icing on the i7 cake is a pair of logic features found on neither the Phenom nor the Core 2. The first is Intel’s Hyper-Threading (HT) technology, as implemented in the Pentium 4 and, more recently, the Atom processor.
This allows each CPU core to present its spare execution capacity to the OS as a second core, which can speed things up whenever more than four process threads need to be serviced simultaneously. Since most applications are single-threaded, it’s unlikely to make a big difference in everyday use; but it’s fun to see eight CPU graphs appear in the Windows Task Manager.
The second is a neat new feature called Dynamic Speed Technology, which allows the processor to detect when load is unevenly balanced and automatically boost the speed of the cores with the most work to do.
Idle cores are clocked down to keep the chip’s overall power consumption within tolerance. Don’t expect to see dramatic overclocking: from early technical documentation, it looks like the biggest dynamic clock rise you’ll see is 266MHz – but every little bit helps.
Nehalem architecture is designed to be scalable and modular, so expect dual, quad and eight-core versions released in 2009, for the mobile, desktop and server markets respectively. As in the past, there’s some overlap between laptop and desktop processors.
 |
| Motherboard diagram (click on image for larger size) |