Intel Core i7 Architecture Decoded and Explained

Intel Core i7 Architecture Decoded and Explained

With the release of the Core i7 just recent, many people are wondering just how these new chips work.  Before the NDA was lifted, information was scarce to come by.  With the NDA lifted, information about these new chips has been flooding out.  This article will help explain the new features and how this chip operates.

New Features

The Core i7 processors have some great features that the previous Core 2 processors did not have.

Modular Design - The Core i7 processors have been modularized to help Intel create different versions without re-working the whole processor.  This means 8-core processors, 6-channel memory, and larger cache processors are all possibilities.

HyperThreading Returns - Intel brought back HyperThreading to allow these processors to utilize up to 8 threads simultaneously.  HyperThreading is the technology that allows a single core to emulate two cores by using unused core hardware to run a separate thread. Intel claims that adding HT allows for more performance without adding almost no extra hardware.

Integrated Memory Controller - Intel has placed all memory controlling hardware directly into the processor.  This means more bandwidth and lower latencies by bypassing the FSB.  Also, much more memory is supported due to the triple-channel.

Quick Path Interconnect - The QPI is the replacement for the FSB.  With much higher bandwidth and point to point connection, it scales much better with multiple processors then previous generations of hardware.  Also, look to see greater performance with dual and triple SLI/CFX setups due to increased bandwidth.

Low-Power Modes - If all the cores are not active with a thread, the core itself will be shut down and operate in a sleep-like state.  In this state, the core operates at almost 0 watts.  There are no perceived delays as the core is able to instantly wake.

Turbo-Mode - The processor is able to self-overclock by changing the multiplier by two speed bins.  It will self-overclock if the processor senses that there is enough thermal and power headroom to overclock without straining itself.  Usually this happens when there are cores in sleep state due to lack of multiple threads.  The PCU (Power Control Unit) is advanced enough to know all this and will be very self aware.  Liquid cooling setups may even see higher self-overclocks due to this.

The Architecture

As stated before, the Core i7 has been completely modularized.  The two main modules are the Core and Uncore.  The Core controls all the cores while the Uncore controls the L3 cache, memory controller, and QPI links.  However, each module receives its own clock so that all the different modules can run independently of each other.

Core - The core portion is the location of all the cores of the processor.  This is the where the CPU frequency comes into play.

Uncore - While the Uncore technically includes the QPI, the QPI is able to run independent of the Uncore settings.  The Uncore includes the L3 cache and memory controller.  The Uncore also controls what memory speeds are available to use with the processor.  The Uncore must be at least double (2x) of the memory frequency due to memory being DDR.

QPI -  The QPI is the connection interface between processor and the rest of the system.  It runs independent of the other modules in the processor.  It also transmits and receives per clock, so hence this module is rated in transfers per second instead of frequency.

To let everything run independently, there is a single base clock used throughout the processor.  From now on, the base clock will be referred as Bclk.  The stock Bclk is 133 MHz, and all the components have a separate multiplier to achieve their desired operating frequency.

CPU Frequency (3.2 GHz) = Bclk (133 MHz) * Core (24x)

The Processors

So far, all the different processors have different multipliers.  The 920/940 processors are locked down while the 965 Extreme is able to change its multipliers.  Also shown are the frequencies able to achieve with the Bclk 133 MHz stock.

 

While these are the default settings set by Intel, some boards may be able to set higher values to the 920/940 processors by bypassing some of Intel's locked settings.  However, the settings below have been approved by Intel and should not be exceeded.  Certain boards stick to these recommendations while others bypass.

 

Core i7 965 Extreme

Core:	12 - 256	1.6 GHz - 34 GHz
QPI:	18, 20, 24	4.8 GT/s, 5.32 GT/s, 6.4 GT/s
Uncore:	2 - 30	266 MHz - 4.0 GHz
Memory:	2 - 30*	266 MHz - 4.0 GHz

Maximum is restricted to 14x and below because Uncore must be 2x of memory.

Core i7 940

Core:	22	2.93 GHz
QPI:	18	4.8 GT/s
Uncore:	16	2.12 GHz
Memory:	6, 8	800 MHz, 1066 MHz

Core i7 920

Core:	20	2.66 GHz
QPI:	18	4.8 GT/s
Uncore:	16	2.12 GHz
Memory:	6, 8	800 MHz, 1066 MHz

Conclusion

As one can see, Intel has created a radically new processor which will greatly increase multi-threaded performance.  While games will not take advantage of this yet, more and more games are starting to take advantage of more then a single core.  Other applications such as music and video encoding/decoding software will take advantage of the new SSE 4.2 instructions.

 

Once you understand how the processor works and depends upon the other modules does it make sense what Intel did with the processor.  Overclocking is easier once the fundamentals of the processor are understood.  You are now able to extract the full potential of the chip without stressing any single part to failure like the Core 2 processors.

 

With a 10% - 30% increase in performance clock for clock against Core 2 Quads and a better chipset, the X58, to operate under, this is the new crown in CPU power and efficiency.