Way back
at the Intel Developers Forum 2009, Intel Corp. was talking about Sandy Bridge,
the code-named architecture successor to Nehalem. Nehalem had
already seen a die shrink from 45 nm to 32 nm; shrunken Nehalem chips
(which featured lower power usage and a few other tweaks) were known as Westmere. Sandy
Bridge, like Westmere is produced at the 32 nm node.
Nehalem and Sandy Bridge are "Tock"
processors -- a new architecture -- by Intel's terminology. Penryn and Westmere are
the "Tick" -- die shrink -- that preceded them (at the 45 nm and 32
nm process nodes, respectively).
Sandy Bridge was talked about more at recent Intel earnings reports, and in recent
weeks Sandy
Bridge-based laptops were
leaked in various foreign tech publications.
On Monday, Intel kicked off what should be a week of exciting announcements
from various players in the electronics industry by finally revealing Sandy
Bridge in all its glory.
We have a roundup of various Sandy Bridge reviews from other
sites here.
I. A CPU By Any Other Name
i. General Branding "Core 2011"
Rather than ditching the "Core ix" brand name, Intel is sticking with
it. You can call Sandy Bridge Core series CPUs
"Core generation 2" or "Core 2011" or some other
catchy moniker. But the point is the name is staying the same.
The good here is that there's comfort in familiarity and customers won't be
taken aback by strange new names. The decidedly bad thing about Intel
picking this branding is that it fails to convey that Core 2011 CPUs are
decidedly different than Westmere or 45 nm Nehalem designs.
But for better or worse, that's what Intel decided and we have to live with it.
ii. How Many Cores are in a Core ix?
Likewise customers will have to live with a rather confusing mix of
"ix" brand modifiers on the desktop and laptop side. What is an
i3, an i5, or an i7? Well if you guessed a 2-core, 4-core, and 8-core
CPU, respectively you may be part right, part wrong or completely wrong --
depending on whether you're talking about desktop or laptop CPUs.
The bewildering distinction of what makes a Core i7 or a Core i3 is summed
up below.
The 2 core, 4 thread CPU can be referred to as a Core i3 or Core
i5, depending on its part number, on the desktop side; or a Core i5
or Core i7, on the notebook side.
Four core, four thread processors fall under the Core i5
distinction on the desktop side, but aren't announced for the notebook
side.
Four core, eight thread processors are referred to as either Core
i7 or Core i7 Extreme on the notebook side, but are referred to as Core
i7 only on the desktop side.
There also are expected to be 6 core, 12 thread; and 8 core, 16
thread desktop CPUs in the Core i7 and Core i7 extreme desktop brands.
Again, we
think Intel's naming scheme leaves much to be desired, but
we're not marketing gurus -- perhaps they know something we don't.
Now we get to the actual CPU numbers. If you can figure out by now what a
Core i7 desktop is versus a Core i7 laptop, you are ready for this next step.
iii. Model Lettering/Numbering
Diving into lettering and numbering, you'll find more of the same -- a
relatively confusing slew of numbers and letters to digest.
On the desktop side, the Pentium (2 thread, 2 core) entry-level processors are
named "G6xx" and "G8xx" with letters possibly being tacked
on the end (more on this later).
Core i3s are "21xx", Core i5s are
"23xx"/"24xx"/"25xx", and Core i7s are
"26xx". To our knowledge the precise name of the high-end Core
i7 Extreme hasn't yet been announced.
On the desktop side you might see some letters tacked on to the end of your CPU
name. "K" means that you have an unlocked core multiplier,
perfect for overclocking (goody!). "S" means you have a more energy
efficient processor for the same clock speed. And "T" means you
have an even more energy efficient processor for the same
clock speed.
When it comes to laptops, the entry level Celeron (currently only one has been
announced) bears the "B8xx". Core i5 models are
"25xx", Core i7s are "26xx", "27xx", and
"28xx". The Core i7 Extreme will be "29xx".
Again, the model name doesn't necessarily correspond to any certain core/thread
count. For example a "26xx" mobile CPU can have two or four
cores.
The mobile processors also have letters of significance on the end. The
most common is M, which simply means (dual-core) mobile processors, and "QM"
means quad-core mobile processors. "XM" means you have an
unlocked multiplier. "LM" means a low-voltage mobile
processor. And "E", "QE", and "LE" stand
for an embedded (dual-core) mobile CPU, a quad-core embedded mobile CPU, and a
low-voltage (dual-core) embedded mobile CPU, respectively.
iv. Chipsets
To make matters even more confusing, Intel will be releasing a total of 10
chipsets. On the laptop side there's QS67, QM67, HM67, HM65, and
UM67. On the desktop side there's P67, H67, Q65, Q67, B65.
Now you know the Core 2011 series in its full naming glory.
II. What's in a "Sandy Bridge" CPU?
Westmere brought a PCIe/memory controller and integrated
graphics processor in to the chip package; essentially moving two key
components of the chipset inside the chip package. However, the I/O
controller and the iGPU were bundled together on a separate 45 nm chip, while
the CPU was a separate 32 nm die.
i. One Die to Rule Them All
With Sandy Bridge we finally see a true one-die
solution. An iGPU is onboard, as is a multi-purpose I/O controller,
responsible for memory control, PCIe communications, etc. Best of all,
these components share a common L3 cache, which is 3, 4, 6, or 8 MB on the
laptop processor models (up from Nehalem's max of 4 MB) and 3, 6 or 8 MB on the
desktop processors (interestingly, down slightly from the 12 MB found in most
Nehalem processors). To pull off the feat of a shared common cache, Intel
uses a ring-bus design.
The new iGPU can easily handle basic graphics tasks like play videos. In
doing so it can save the power required to run a full GPU, which is especially
important in the laptop space. It can also, for the first time, play some
modern game titles (set to the lowest resolution, of course) with acceptable
frame rates (e.g. Batman: Arkham Asylum and Dragon
Age: Origins).
The iGPU is a brand new designed dubbed 2000 or 3000 HD. In the desktop
models it runs at 650-850 MHz and can be turbo clocked up to 1100-1350
MHz. While that makes it competitive in clock-speeds with traditional
graphics chips from AMD and NVIDIA, it still has far less shaders than a
traditional GPU and hence can't draw as pretty graphics as fast.
Early benchmarks show them pulling off twice the performance of
Intel's previous gen offerings. The performance bump appears to be
largely from the clock speed increase, as the number of execution units (12)
has stayed the same from the previous generation.
To put things in context, a Sandy Bridge chip's iGPU, according to Anandtech can
outperform the Radeon HD 5450, an entry-level graphics card from AMD's previous
generation.
The GPU is only capable of handling DirectX 10.1 graphics in Windows and can't
run DirectX 11 in its current form.
ii. Performance, Features Get Bumped
The chip also packs other power, processing, and multi-threading
improvements. According to AnandTech, these improvements
amount to a 20 percent performance improvement on average over identically
priced previous generation offerings for 4 and 8 core varieties.
The performance bump in the dual-core doesn't seem as noticeable -- AnandTech reports
it to be around 5 percent on average bump. Intel claims its multi-core
performance improvements can reach 60 percent, but those kinds of numbers are
yet to be seen in real-life benchmarks (but this is pretty typical -- CPU
makers often exaggerate performance increases and/or cherry-pick best-case
scenarios).
One significant improvement in the I/O end is that idle optical drives can be
turned off to save power. After the LCD screen, CPU, and chipset, this is
one of the more power-hungry components on the laptop, so this should
definitely be a boost in battery life (even when idle, DVD drives on laptops
reportedly pull down several watts).
Overclocking is expected to be great on the new chips which feature unlocked
multipliers. AnandTech was able to overclock quad and eight
core desktop Sandy Bridge chips to 4.4 GHz on only the stock
air-cooling solution. Where as in the past overclockers had to muck with
both the BLCK and the multiplier to get optimal results, you now generally
leave the BCLK alone and only play with the unlocked multiplier.
iii. Socket and Chipset
To use a Sandy Bridge CPU on the desktop end, you will have to use a Socket
1155-capable motherboard. The old Socket-1156 design from the Nehalem era
has been ditched. It can be assumed that the laptop will also use an
1155-pin design, meaning that the 1155-pin Sandy Bridge mobile
chips won't be available as an upgrade option for previous Core-based notebooks
(which had 1156-pin G1 model CPU sockets).
The one good thing here is that the majority of CPUs on the desktop side will
use the same socket (Socket-1155) versus the previous generation, which saw
multiple different sockets. The only chips to use a different socket on
the desktop side will be the highest end Core i7/Core i7 Extremes, which use
Socket 1356 (not to be confused with Intel's previous gen. Socket 1366).
The net message, though, is you will be able to use more CPUs with your socket
Intel promises its new chipsets can handle 1080p resolution 3D (just be sure
you wear some goofy glasses), which should be handy for the whole 3 or 4
worthwhile games and movies that currently support this technology.
Slightly more utilitarian is the Sandy Bridge-based chipset's
improved ability to connect to HDTVs and display your screen wirelessly.
Dubbed "Wi-Fi Display", WiDi 2.0 expects to bring improvements (the
ability to stream HD video), but won't eliminate a bit of lag or the need for a
Wi-Fi receiver for your HDTV.
Precise details on the various chipsets are scarce at this point, but a few
sites have published some information. For example, LAN OC reveals that
P67 won't support HDMI/DVI out and may cut down the 16 PCIe lanes down to 8
GB/s. By contrast the H67 features 16 PCIe lanes with 16 GB/s throughput
and the ability to out HDMI/DVI/etc. (which Intel calls "Digital
Display" technology).
III. Conclusion and the Outlook Ahead
i. Buyer's Outlook, AMD's Answer, and
Ivy Bridge
If you're in the market for a new computer, you'll probably want to hold off
about a month. Computers -- both laptops and desktops -- based on Sandy
Bridge chips shouldn't take long to arrive. Many should be
announced this week at CES 2011.
To recap, quad-core or higher Sandy Bridge CPUs should give
you about a 20 percent performance increase and perhaps an even nicer battery
life boost in the laptop space. And they're about the same price as the
previous generation models, so Sandy Bridge notebooks seem a
no-brainer, if you're picking an Intel-based laptop.
The only hesitation in the desktop space will likely come from the fact that
you'll need to buy a new motherboard to use a Sandy Bridge CPU.
But if you value performance or are looking to build a brand new system, it
makes sense to take the dive.
Of course if you can wait a bit more, AMD should be releasing it's own next
generation architecture Bulldozer (desktop primarily) and Bobcat (laptop)
cores. Performance on these chips is unknown, but Bobcat holds
at least one key advantage over mobile Sandy Bridge chips, by
offering full on-die DirectX 11 graphics, something Intel won't offer till next
year.
The introduction of AMD's competitors should help to force Intel to possibly
drop the price of its Sandy Bridge offerings, so that's another
reason to possibly wait for their release.
Looking ahead, Intel's next move will be to shrink Sandy Bridge to
22 nm, code-named Ivy-Bridge. These chips will likely be
introduced in about a year -- January 2012 seems most probable. They will
feature a significantly improved integrated GPU (thanks to the doubling of the
execution units onboard). This will allow the new iGPUs to support DirectX
11. A cache increase seems likely too, given the addition of more hungry
iGPU components.
Ivy-Bridge should also feature power improvements, given the
smaller transistor size and new processes to reduce current leakage.
ii. Far Down the Road
Looking even farther ahead, in 2013 (again, likely in January), Intel will introduce Haswell CPUs, the next major architecture. One big
boost with Haswell will be the inclusion of 8-cores as the
default for Intel's chips.
And in 2014 (assuming all goes well with processes), Intel will shrink Haswell to
16 nm, a shrink that is code-named Rockwell. This may be a
significant release, as for current die technologies 15-16 nm is expected to be
the physical limit of optical (X-Ray) lithography.
However, it is possible that probe-based lithographic techniques maybe be able
to position individual atoms on a chip (a silicon atom takes up around 0.5 nm
in a typical crystal lattice), allowing even smaller silicon-based
transistors. Another much discussed possibility is 3D chip stacking.