Sign in with
Sign up | Sign in

Modern Quad-Core CPUs: AMD Deneb And Intel Sandy Bridge

Tom's CPU Architecture Shootout: 16 CPUs, One Core Each, And 3 GHz
By

AMD Deneb, 45 nm (Phenom II X4 980, Rev. C3)

AMD's quad-core Deneb design shares its basic architecture (like the Socket AM3 interface, dual-channel DDR3-1333 controller, and 45 nm manufacturing process) with the six-core Thuban configuration. Because it includes fewer cores, most models enjoy a lower 95 W TDP. However, the Phenom II X4 980's lofty 3.7 GHz default clock rate pushes it up to 125 W.

Intel Sandy Bridge, 32 nm (Core i5-2500K, Core i7-2600K, Rev. D2)

We decided to test two Sandy Bridge-based quad-core processors for the LGA 1155 interface. The Core i7-2600K has 8 MB of shared L3 cache, while the Core i5-2500K does its job with 6 MB of shared L3 cache. They all have 256 KB of L2 cache per core.

Compared to the 32 nm hexa-core Gulftown design, Sandy Bridge appears very similar. However, it represents the second generation of Intel's Core architecture, as many aspects of it were redesigned. It employs an internal ring bus to shuttle data around faster, and it adds support for the company's AVX extensions (not to mention the built-in HD Graphics engine and Quick Sync functionality).

This architecture is expected to be the basis for all upcoming desktop, server, and notebook processors in 2011, and they'll continue existing once Ivy Bridge-based chips start to emerge. As you will see in the benchmarks, Sandy Bridge delivers the best performance per clock.

Intel Lynnfield, 45 nm (Core i7-875K, Rev. B1)

The 45 nm Lynnfield design is available commercially as Intel’s first-gen Core i7 and Core i5 for the LGA 1156 interface.

It's important to pay attention to the model numbers, as Intel's Core i7-900-series requires an LGA 1366 interface, while the Core i7-800-series is designed to drop into the LGA 1156 interface. Confusing the two would naturally create an interoperability issue.

Lynnfield has 8 MB of L3 cache and 256 KB of L2 cache per core, a 95 W TDP, and an integrated DDR3 controller that supports two channels of 1333 MT/s memory.

Intel Bloomfield, 45 nm (Core i7-975, Rev. D0)

This was the first Nehalem-based quad-core processor designed for high-end desktops and workstations. Despite their age, Core i7 chips centering on the Bloomfield layout are still very modern. But they lack some of the company's newer features like AES-NI and AVX. 

Display all 133 comments.
This thread is closed for comments
Top Comments
  • 31 Hide
    Darkerson , July 26, 2011 5:43 AM
    This was a pretty nifty article. Hope you guys revisit it after Bulldozer and Ivy Bridge drop. Thanks!
  • 22 Hide
    jcesmi , July 26, 2011 8:05 AM
    Where are those bots i buy all my cloths from?
  • 21 Hide
    fstrthnu , July 26, 2011 5:51 AM
    Thing is, Intel will already have their updated Sandy Bridge processors by the time Bulldozer comes out - which will probably maintain the gap. AMD would need a MASSIVE effort to catch up or even pass Intel at this point. Bulldozer's going to have to try really hard to win back the high-end enthusiast market. Intel clearly has had much better direction from Intel Core up to now.
Other Comments
  • 31 Hide
    Darkerson , July 26, 2011 5:43 AM
    This was a pretty nifty article. Hope you guys revisit it after Bulldozer and Ivy Bridge drop. Thanks!
  • 2 Hide
    fstrthnu , July 26, 2011 5:45 AM
    Wow, this has got to be one of your biggest comparos EVER. Didn't finish reading it yet, but it looks like this one will be quite a doozy
  • 4 Hide
    wintermint , July 26, 2011 5:47 AM
    Well AMD hasn't released a new architecture in a long time.. what you expect?
  • 21 Hide
    fstrthnu , July 26, 2011 5:51 AM
    Thing is, Intel will already have their updated Sandy Bridge processors by the time Bulldozer comes out - which will probably maintain the gap. AMD would need a MASSIVE effort to catch up or even pass Intel at this point. Bulldozer's going to have to try really hard to win back the high-end enthusiast market. Intel clearly has had much better direction from Intel Core up to now.
  • 11 Hide
    dragonsqrrl , July 26, 2011 5:55 AM
    Every single comment posted before Darkerson (there were quite a few) seem to have mysteriously vanished... strange.
  • 8 Hide
    cangelini , July 26, 2011 6:12 AM
    dragonsqrrlEvery single comment posted before Darkerson (there were quite a few) seem to have mysteriously vanished... strange.


    How many more were there? That's not something I've seen happen before.

    Best,
    Chris
  • 5 Hide
    Lewis57 , July 26, 2011 6:14 AM
    One of the most interesting articles I've seen from toms. I already had an idea of the standings between intel and AMD per core per clock by how game minimum specs usually say something like "Minimum Intel Core 2 Duo 1.8Ghz or AMD Athlon x2 2.4Ghz", but it was nice seeing it put into perspective.

    Hope you revist it with bulldozer/ivy bridge.
  • 2 Hide
    crisan_tiberiu , July 26, 2011 6:18 AM
    Great review, long live my i7 2600k :) 
  • 4 Hide
    yyk71200 , July 26, 2011 6:39 AM
    Looks like Bulldozer is going to combat SB by moderate increasing of IPC AND increase in frequencies. Purely by IPC it will lose to SB but if it can deliver high frequencies at the same time, we may have a decent competition.
  • 1 Hide
    dragonsqrrl , July 26, 2011 6:44 AM
    cangeliniHow many more were there? That's not something I've seen happen before.Best,Chris

    I don't know for certain, maybe around 6 or 8. I'm only aware of this because I was one of the people who posted a comment.
  • 4 Hide
    Anonymous , July 26, 2011 6:48 AM
    Now this was a great article! Good work guys!
  • 9 Hide
    yyk71200 , July 26, 2011 7:00 AM
    DjEaZy... you couldn't wait until the bulldozer comes out?

    I think it may be a bit difficult to test bulldozer this way because configuration of cores within modules is rather funky. That is they are not full fledged cores within a module.
  • 14 Hide
    clonazepam , July 26, 2011 7:25 AM
    Pretty cool article.

    Thanks.

    That must have been one tedious sob to pull off. I wouldn't have had the discipline to even finish the P4 tests lol...

    I'm gonna go stroke my kentsfield... I think it knows when the new system's up, it's going to be subjected to 1.5+ volts ala Frankenstein's Monster... bwuahahahaha
  • 4 Hide
    PreferLinux , July 26, 2011 7:44 AM
    yyk71200Also, clock for clock testing is not always fair. For example, in P4 vs. Athlon 64 days clock for clock testing would not be fair because while P4 was much worse in IPC, it was designed for higher frequencies. Sure, Athlon still bit it in most cases, but gap was much narrower that clock for clock tests would suggest. We may have a similar situation (in reverse) in SB vs. Bulldozer.

    You talking stock or overclocked? And if overclocked, then is it a 24/7 clock with what cooler? My point is that we don't really have any idea how Bulldozer will compare in clock speeds – it could be very well at stock but very poorly when overclocked (24/7 clock, but once again I'm not specifying with what cooling (could be low-end air to high end water or even LN2!) – but same on each), for all we know.
  • 22 Hide
    jcesmi , July 26, 2011 8:05 AM
    Where are those bots i buy all my cloths from?
  • 9 Hide
    outlw6669 , July 26, 2011 8:09 AM
    Excellent work reviewers!
    It is great to see a solid, in depth, tech article again =D

    @yyk7120, the entire point of this article was to show the relative IPC performance between the different architectures.
    In a battle of pure IPC performance, nominalizing all variables is not being unfair, it is entirely necessary.
    Think of it more as an in depth look into the underlying architectures rather than another AMD vs Intel @ price point that most are so used to.
  • 14 Hide
    memadmax , July 26, 2011 8:16 AM
    This is about as close as you can get to real raw core efficiency. I love this article and was wondering the same thing as well.
  • 20 Hide
    Anonymous , July 26, 2011 8:25 AM
    I would LOVE to see one additional page - 'performance per transistor'

    Basically - take the performance data, and divide the results by the amount of transistors per core (eg 1 billion transistors for 8 core -> 125 mil transistors for core). There would be a bias towards CPUs with huge L3 cache (as they can use all of it), but it would still be very interesting thing to see

    I would really want to see how the efficiency of the processors increased compared to the amount of transistors they use
Display more comments