'New Old' Core i7 Architecture
Nehalem pays so much attention to power management as if Intel had known about the world economic crisis beforehand. The new architecture approaches it in a traditionally profound way: the entire power management system is implemented as a separate unit called PCU (Power Control Unit). It acts as a primitive processor in a processor.
This microcontroller supports a much more complex power management scheme than older processors from Intel: frequency and voltage for each core are controlled separately based on information about temperature and consumed power. Thus, each core can switch to low power consumption individually. Memory controllers and QPI buses go to the low power consumption mode, when all cores are idle. In fact, AMD did a similar thing with its Cool'n'Quiet 2.0 technology (Phenom) for one exception: C'n'Q 2.0 allows a processor to interact with the on-board power converter, while Nehalem's power management is implemented entirely inside the CPU and requires (or allows) no additional devices.
Intel was approaching the overclocking idea slowly but steadily: at first overclocking functions appeared in its motherboards, now they are integrated into processors. :) Seriously, it's Nehalem's PCU that allows to implement another interesting feature: this processor can raise the frequency of one or several cores, if the other cores are idle. As far as we understand, there are two turbo options: raising the frequency of several cores by one step (+133 MHz) and raising the frequency of one core by two steps (+266 MHz). It's not necessary for the other cores to be idle: Turbo Mode steps in, when the core usage allows to raise frequency in some cores without exceeding the maximum TDP level. There is an extra bonus here -- Turbo Boost, just like PCU, has nothing to do with external support. That is it works inside a processor and does not require any additional software or hardware tools.
Getting back to what we wrote in the very beginning, we can conclude our brief analysis of the new architecture from Intel with the following words: when NetBurst was stagnating, Intel needed a hero architecture to return faith that this company could still manufacture good and fast processors -- now the time of hero need is gone, priorities have changed: now the company employs the architecture-accountant -- predictable, careful, systematic, prone to optimizing everything. Being a successor to the previous architecture in many key features, the new architecture looks much more polished. It's all orderly, logical, and systematic. Everything is arranged in perfect order. Probably the key attraction of the new architecture consists in easy design and fabrication of such processors -- we can say that this architecture is "production-oriented". By the way, as any good accountant, the new architecture carefully picked up some handy features of its unpromising predecessors -- Hyper-Threading and Trace Cache from Pentium 4 / NetBurst. Why not?
If we take a look at Nehalem from this point of view, all questions about its speed versus Core 2 fall away. Yep, it will probably be a tad faster. It cannot possibly be slower, can it? But that's not the point. In fact, Intel now stands at the beginning of its road to an ambitious destination: streamline CPU design, not just manufacturing. It's all very simple actually: split a processor into units, find out which unit acts as a bottleneck, and then deal with this unit. Bottleneck eliminated, the overhauled unit is returned to the old design, and here we go -- a new improved processor. It can be done endlessly. It will be even possible to update the instruction set, if necessary -- as it's a responsibility of another standard unit. I don't know why, but I think that Intel will succeed this time. Well, in fact I know the reason exactly: because this state of affairs suits everyone in the longer term -- Intel, system integrators, and end users. It will be boring, of course. But we'll manage…
So, are there more differences or similarities between Core 2 and Core i7? In our opinion, their similarities and differences lie in different planes. Nehalem inherited an important part of the microarchitecture from Conroe/Penryn (the execution core), but it differs much from Core 2 in architecture: this processor was apparently designed to solve different tasks and reach much more distant objectives than just to win the performance race. It may sound even impudent to the Core 2 series, but nevertheless, we'll dare to assume that its strategic objective was to give Intel an opportunity to ignore various minor issues (like the performance competition) and focus on the new long-term development concept of its processors. And Core i7 processors have become the first implementation of this project.
Write a comment below. No registration needed!