One of the reasons behind the changes of architectural Nehal is the server market. The architecture of the Core 2 Duo and is derived from the Banias, which was developed as a low-power chip for laptops. They provide a link performance / consumption are very good and competitive for the AMD processors, but the legacy mobile weighs in some situations, particularly with respect to virtualization and performance in the database, two major clusters within the area of servers, in which AMD still remained strong on Penryn.

With Nehal, Intel worked to eliminate these bottlenecks in performance, creating an architecture that can be aided by increases in caches and other improvements to be introduced with the migration to the architecture of 32 nanometers in 2010.

The first processors based on the received Nehal codenome Bloomfield. Initially, it gave rise to only three processors, the I7-920, I7 and I7-940 XE-965, which were followed by I7 and I7-950 XE-975:

Core I7-975 XE: 3.33 GHz, 8 MB, DDR3-1333, the TDP 130w, unlocked multiplier
Core I7-965 XE: 3.20GHz, 8MB, DDR3-1333, the TDP 130w, unlocked multiplier
Core I7-950: 3.06 GHz, 8 MB, DDR3-1066, the TDP 130w
Core I7-940: 2.93GHz, 8MB, DDR3-1066, the TDP 130w
Corei 7-920: 2.66GHz, 8MB, DDR3-1066, the TDP 130w

As you can see, the three processors offer basically the same characteristics, changing only the clock, which by the way still parked at around 3 GHz, that mark was reached at the time of the Pentium 4. This is due to the simple question of heat dissipation. It would be possible to produce processors with much higher clock with current technology, but the consumption and heat dissipation would be impractical, as was made clear at the end of the Pentium 4 era.

A good example is Intel's decision to use transistors in static CMOS Nehal, instead of using domino logic circuits, as in all previous cases (up to the Penryn). In summary, the domino logic circuits gives capable of operating at much higher frequencies, but that in turn consume more energy, while the CMOS circuit results in low consumption.

That will not back any benefit in terms of performance (rather, it limits somewhat the frequencies supported, who want to resort to more aggressive overclocks), but it is positive in terms of efficiency, allowing the processor operating within the mark the same TDP of 130 watts when in full-load.

In other words, Intel has chosen to focus on optimizing the architecture and to reduce the consumption of processors, instead of simply trying to win on the basis of brute force, increasing the clock.

As with other Intel processors, the TDP specification indicates only the maximum consumption of processors, serving as a reference for manufacturers of coolers, fountains and the motherboard, so just not very significant. In situations of normal use, the management of energy resources into action, keeping the consumption at levels much lower.

Another important observation is related to the frequency of memory. Intel was very conservative to adopt the DDR3-1066 as standard on XE models outside the range, although the majority of modules on the market bear much higher frequencies.

This has a reason: the concern to popularize the modules quickly DDR3, thus opening the way for the growth of the platform. The popularization of any new technology is the growth of production and decrease the cost and DDR3-1066 modules are much simpler and cheaper to produce than modules faster.

Another reason behind the decision has to do with tension. Almost all modules capable of operating at 2 GHz or more uses voltages above 2 volts, well above the rated voltage of DDR3 modules that is only 1.5V. As in the case of processors, increasing the voltage allows the memory modules are capable of operating at frequencies a little higher, but disadvantages as behind the increase in energy consumption and a substantial reduction in the lifetime of the modules.

The race around faster DDR3 modules for the enthusiast public has made the manufacturers would sell overclocados modules, focusing on the frequency and performance benchmarks rather than the life of the modules.

With I7, Intel decided to "start over", encouraging manufacturers to produce modules slower, but capable of working within the rated voltage. Naturally, you can use modules faster or even to overclock the memory, but Intel has warned that the use of voltages above 1.65V processors in the platform Core I7 can damage the memory controller after a while of use, unused processor.

Anyway, we Lynnfield, which will be a low-cost Core I7. Many of the resources released by Nehal, including the ability to use 4 processors (no less than 16 nuclei) are intended for the server market, where growing demand for machines capable of rodarem hundreds of virtual machines simultaneously and process databases with several gigabytes in a timely fashion.

On a desktop, the use of multiple processors does not make much sense (after all, few applications are able to use the full potential of 4 cores, what about 16), which makes the discussion more falls on the earnings performance the Core I7 on previous processors. The changes introduced by Nehal are positive, but the most important has been the cost.

The Lynnfield is a desktop version of Nehal with a memory controller, dual-channel support and without the use of multiple processors. With at least one channel to the memory controller, the number of contacts was reduced, leading to more Intel to develop a new socket, the LGA1156, which is the direct successor of the current LGA775. These changes simplify the architecture, making the chip (and hence the parent plate) more affordable and eliminating the temptation to use three modules of memory instead of two.

Another important change is the inclusion of 16 lines PCI Express within the processor, which makes the Lynnfield perform all functions that are executed on the platform I7 by X58.

Although not bring change with respect to performance, this change has allowed Intel to simplify the platform, adopting a single-chip solution for the chipset.

Another small advantage of integration is a slight reduction in total energy consumption, as to be moved into the processor, the PCI Express controller transistors have been fabricated using the same technique of 45 nanometers it.

You could ask that the wait to incorporate the Intel PCI Express in Lynnfield lines, rather than incorporating them already in Bloomfield. The answer is a simple detail: the Lynnfield has only 16 lines, while Bloomfield and X58 offer 36 lines in total. If incorporated the 36 lines directly on the processor, the Bloomfield would have a much greater volume of transistors and Intel would have difficulty in keeping the TDP of 130 watts to the processor, hence the use of X58.

Other features of the internal processor, such as caches and support for SMT (with the exception of the 2.66 GHz model) remain the same, making the clock for clock performance of Lynnfield and Nehal is very similar. The bear also Lynnfield Turbo Boost, often not yet confirmed.

Of course, Intel needs to differentiate the two platforms, so the models based on the use Lynnfield Clocks lower, justifying the reduction in cost.

The Lynnfield will be sold from September. The launch marks the end of the line for the current LGA775 processors, which will be relegated to low-cost lines and eventually discontinued.

That is no reason to fail to mount a micro with a Pentium or Celeron And since low-cost processors based on the new architecture should be launched only at the end of 2009 and does not popular before 2010. However, if you are thinking about buying a Core 2 Quad based in Penryn, it is advisable to postpone the purchase.

The first cards will be based on P55 chipset, which is basically an updated version of the old ICH10 chip that was used as the south bridge chipsets earlier, which is connected directly to the processor and through a link DMI. Following the change, the chip is now called the Platform Controller Hub (PCH), indicating a redistribution of duties.

One change is that the cards will be much simpler, without the traditional north bridge chipset and heatsink in the middle of the plate:

The additional slot among the slots of memory and the source connector (you can find on many boards with the P55 and X58) will install a memory module NVRAM, to act as a cache for additional operations to access the disk. He is a kind of fueled version of ReadyBoost, which serves as an alternative to improve the performance of access to disk, without having to buy an SSD. This is a proprietary technology of Intel (named Braidwood), so do not wait to see her signs for AMD.

A careful reader might ask why the continued use of the Intel chipset separate rather than integrate all the interfaces directly in the processor, as it did with the PCI Express lines.

From the technical point of view, would not have many problems in doing this, since the P55 has more than 5 million transistors, which is less than 1% of the transistors of the processor.However, the idea would have a practical problem, which is the issue of trails.

The south bridge chipset, which is basically that remains at P55, serves as a "hub" for the trails that go to the slots and ports of the motherboard. We have so few tracks (corresponding to bus DMI) between the processor and the P55 which is in all other branch. Removing the chip, these tracks need to go directly to the processor, which greatly increased the number of contacts in the socket and complicate the layout of the plates.

Another reason, perhaps more important than the technical issue is that the Intel chipsets make much money selling a rental that would be lost if they were removed completely.Although the P55 is brutally simple that the P45 (used in signs for the Core 2 Duo), Intel to sell to manufacturers for about the same price (around $ 45), resulting in a profit margin fabulous.

As commented in the beginning, the Nehal is the first in a series of new processors based on the new architecture, a family that includes several other models, to be launched over the coming months:

Nehal-EX: This is the version for servers. He is a native processor 8-core with 24 MB of L3 cache, which supports the use of 2 or 4 processors (ie, up to 32 cores per card), taking advantage of the use of QPI. Due to the increase in the number of tracks necessary to perform the communication between processors, it uses another new socket, the LGA-1567.

Clarkdale: This will be a dual-core version of Nehal produced using a technique of 32 nanometers and a chipset with integrated video (more a function of the chipset moved to the processor), serving as a direct replacement for the Pentium and Core 2 E Duo

Thanks to the technique of production of 32 nanometers, the cost of production as well electrical consumption will be lower than the current processors, which should lead to Inter to make the change rapidly. The expectation is that Intel announced the first processors based on the platform before the end of 2009.

Two observations with respect to the video chipset is that it is a derivative of the GMA X4500, chipsets used in current (and not a derivative of Lahhabee) and it will be a separate chip included within the packaging of the processor and not really an internal component, included on the same silicon chip.

Arrandale: This is a mobile version of Clarkdale, also dual-core, but with a lower TDP. He is to replace the Core 2 Duo in notebooks, complementing the new generation of the Atom (based on Pinetrail), which advances on the netbooks.

Clarksfield: This version is the quad-core for notebooks, also produced using the technology of 32 nanometers. The main difference for desktop processors to be the TDP and the frequency of operation and lower.

There is a good dose of controversy surrounding the use of a quad-core processor in a notebook, as they are predominantly used for work productivity, and some games, tasks where dual-core processors are typically the best option. Nevertheless, Intel seems determined to popularize the quad-core processors on all fronts, including the notebooks there.

The good news is that the use of the PCU and the Turbo Boost to ensure that the chip performance and maintain a balanced energy consumption, even in applications that do not benefit from 4 cores. In other words, the two additional cores may not help much, but at least also will not interfere.

Gulftown: This is the successor of the codenome Bloomfield, produced using the technology of 32 nanometers for the LGA-1366 boards. He will keep the use of 4 cores, but will bring more cache and possibly other improvements related to performance.

The migration to the technique of 32 nanometers resulting in transistors with little more than half the size of those produced using the technology of 45 nanometers, which offer enough space for adding new components. This is where the modular architecture of Nahal really begin to pay dividends.

Initially the Gulftown replace the XE models of the series as high-performance processor, eventually gaining more affordable versions. It also announced a version with 6 colors, to be launched in early 2010.

These versions of 32 nanometers of Nehal are collectively called Westmere, which is the code name of the new platform, similar to the Bloomfield, Lynnfield & Cia. are sub-versions in the family Nehal.

2010 will be a version with integrated GPU, the Sandy Bridge. Unlike Clarkdale, where the GPU is only a second chip in the packaging of the processor, the GPU in the Sandy Bridge will be moved into the nucleus, resulting in a more elegant solution.

Along with Sandy Bridge, will launch the first products based on Lahhabee, the new architecture for parallel processing that Intel has announced since 2007.

The first sample was a demonstration of the architecture of a chip with 80 cores, developed with the objective of providing 1 Teraflop of processing power. Each of the 80 cores is a relatively simple chip, optimized for processing floating point instructions. Each chip has a "router" that connects to the neighbors. This structure allows the instructions are distributed between the nuclei fairly similar to what happens inside a cluster with multiple machines. The main difference is that everything is done within a single chip:

Over time, it was shown that this chip is a massively parallel version of the previous Lahhabee, which Intel wants to enter the field of high performance 3D graphics cards, competing with nVidia and AMD / ATI.

As in the case of Atom, the inspiration for the new architecture came from the old Pentium.Each of the processors Lahhabee is essentially a modernized Pentium 1, with two units of execution and ability to handle 4 threads (instead of just two, as in Nehal) each. They are complemented by a vector-processing unit, comprising 16 separate units, capable of processing floating point instructions for 32-bit and 256 KB of L2 cache, which complement the 64 KB of L1 cache. Unlike traditional processors, which are optimized for the processing of sequential instructions, he specializes in parallel processing, and 3D chipsets.

Developing a CPU optimized for graphics processing may seem strange, but this is basically the same thing as ATI and nVidia have been making since the introduction of programmable shaders, two generations ago.

When looking at the block diagram of a G80 (used in GeForce 8xxx), you notice that it is composed of 8 clusters of processing units, vectors, which are the basis of the chipset. Each has its own cache and communication between them is done in a way not very different from that developed for the Intel Lahhabee:

A single unit would not say a performance of note, but by combining several dozens of them (have announced plans to use 16 to 32 units in the initial generation of products) will have the Intel chip in the hands of a very powerful video.

The first versions will be nothing more than 3D PCI-Express cards regularly, which compete with the release of nVidia and ATI, but the accelerator should also soon replace the integrated video chipsets, resulting in processors with relatively powerful 3D accelerators included directly in nucleus.