[clue-talk] new processors

Dan Kulinski daniel at kulinski.net
Sun Jan 3 12:33:27 MST 2010


Well, not 16 cores, but 8.  I have an i7-860 in my workstation now.

There are a lot of things that the Core architecture has over the NetBurst
(P4) architecture.

NetBurst used a long pipeline to boost speeds above the 2GHz mark.  What
this means is that every operation in the pipeline is there due to a
predictive component in the CPU.  The NetBurst chips had pipe lines up to 20
operations.  If an operation was wrong, the pipeline had to be emptied and
refilled.  This was slow and ultimately detrimental to the CPU.  The
NetBurst architecture also employed HyperThreading on their CPUs.  For a 5%
increase in die size the HyperThreaded core is very limited but does allow
for some operations to be processed.

The Core series of CPUs is actually an evolution from the P3 architecture,
which itself was an evolution of the Pentium Pro.  It is able to handle more
operations per clock than the NetBurst architecture but was ultimately
limited in speed at the time.  Intel's Israeli team was able to come up with
modifications that allowed this chip to scale extremely well.  This is why,
even though slower clock wise, you get more performance.  The Core and Core
2 stay with this architecture and Intel stayed with this chip for quite some
time.

The Core i7 is based off of Intel's Nehalem design.  Once again, operations
per clock have been increased and HyperThreading make a return.  You will
also see the Core i5 which is a more mass market chip and does not support
HyperThreading.  Other advantages with the Core i7/i5 design are the
extremely intelligent power management built into these chips.  Each core is
able to separately change its clock and in fact shut down some of the cores
completely.  If you are running a heavily single threaded workload you will
see one core accelerate beyond it's specified clock rate using the Intel
turbo feature.  Recent Core 2 chips include some of this power management
capability, such as the ability to clock cores separately, but it is not
nearly in-depth as the Core i7/i5 series of CPUs.

Core i7/i5 also have an on die memory controller.  This means that the
memory is controlled directly by the CPU and brings down memory latencies.
With these technologies on the chip, the i7/i5 is about 20% faster clock for
clock than then Core 2 series, at least in theory.  In the field they are
running between 10% and 15%  faster than the Core 2.

Ultimately there are other versions of the i7/i5 chips that I won't delve
into here and that is not even to mention the Xeon chips, nor any of AMDs
chips.

Dan Kulinski
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cluedenver.org/pipermail/clue-talk/attachments/20100103/f1a29511/attachment.html


More information about the clue-talk mailing list