Core i7 has received a few upgrades over its Core 2 predecessors
NEW PROCESSOR DESIGN
Confusingly, although Intel still refers to the Core i7
architecture as Core, the same name as that of earlier Core 2 CPUs, the
architecture has undergone a massive overhaul. The primary objective of the
architects was to increase performance without increasing power consumption.
This entailed devising ways to boost performance other than simply adding more
cores, extending the pipeline or increasing the clock frequency.
This objective was achieved using two main methods: firstly,
improving existing features from the Core architecture; and secondly, by adding
new features. For example, Intel has added massively improved the amount of
micro-ops that can be processed per clock, as well is increasing their
potential complexity. A Core i7 processing core can process up to 128 micro-ops
simultaneously while a Core 2 processing core can only handle up to 96. This makes
a Core i7 more efficient at handling a range of tasks than previous designs.
As well as adding SSE4.2 instructions to the x86 instruction
set of Core i7, Intel has also revised and updated its Penryn 45nm
manufacturing process to improve power efficiency. These tweaks, revisions and
redesigns underpin the huge performance leap of Core i7 over previous
processors, while keeping its power consumption similar to a Core 2 Quad.
BAGS OF CACHE
As each processor core is now fully independent, the cache
architecture has had to be completely redesigned. While Core i7 processors
retain the same basic Level 1 cache architecture of previous Core 2 designs,
each processing core has its own 256KB of Level 2 cache.
This is much less than the 2-6MB that pairs of Core 2
processing cores share. However, Core i7 processors have 8MB of Level 3 cache
for all four processing cores to share, something previously only seen on very
high-end Xeon MP CPUs.
The Level 3 cache of Core i7 is described as inclusive,
which means that it holds the data of all the Level 2 caches. This means that if
a core needs to fetch data, and it doesn’t find it in the Level 3 cache, it can
be confident that the most up-to-date version of that data is held in the
system memory and collect it straight from there.
If the Level 3 cache weren’t inclusive, the core would first
have to look at the cache of the other three cores to ensure that none of these
had a more recent version of the data it needs. Snooping in these caches could
mean stalling these processing cores.
An inclusive cache is therefore said by
Intel to be more efficient than an exclusive cache design, even if means that
1MB of the Level 3 cache is always used by storing a copy of the 256KB Level 2
caches inside each of the four processing cores.
NEW HYPER-THREADING
Core i7 also sees the return of Hyper-Threading, last seen
in the dual-core Pentium Extreme Edition 965 in late 2006. This technology
works in the same way as it did with the Pentium 4, using the spare resources
of an execution unit to run a second thread in parallel.