Intel's 12-Core Xeon With 30 MB Of L3: The New Mac Pro's CPU?

Intel's Xeon E5-2697 V2 And Leaked Benchmarks

Pre-Production Benchmark Data Isn’t Always What It Seems

There were probably some bummed-out professionals, then, when a Geekbench result was recently uploaded to Primate Labs’ online browser showing that 12-core Xeon E5 around nine percent faster than last generation’s dual six-core Mac Pro, based on Westmere-EP.

Source: Geekbench Browser, Primate Labs

Those scores require a bit of context, though. The 32-bit build of Geekbench uses x87 code, for starters, so it isn’t optimized for any of the other instruction set extensions that Westmere-EP or Ivy Bridge-EP support. Getting close to Apple’s claim of doubled floating-point performance requires software compiled with the AVX flag. John Poole, the founder of Geekbench, posted several other reasons why the next-gen and previous-gen Mac Pros might be separated by such a narrow margin.

The leaked result was run using the free 32-bit build of Geekbench on a pre-release build of OS X Mavericks. Switching over to the paid 64-bit build of the benchmark adds SSE support, though that’s still a pre-Pentium 4 extension. Tab between the 32- and 64-bit runs on Xeon X5675-based systems and you’ll find that the SSE-capable build averages 14%-better performance.

Curious as to how the very same 12-core Xeon E5-2687 V2 compared in Windows, I ran my own test on a 64-bit build of Geekbench and scored in excess of 30,000 points—more than 25% faster than the leaked number. The individual sub-tests showed both Xeon E5-based platforms trading blows in the integer and floating-point components, but clearly a more real-world comparison was needed in order to establish the new Xeon’s performance in a workstation environment. Fortunately, I have the upcoming Xeon E5-2697 V2, the upcoming Core i7-4960X, an existing eight-core Xeon E5-2687W, and a Core i7-3970X.

Meet The Xeon E5-2697 V2

Intel's original Xeon E5 family was based on the Sandy Bridge architecture. Launched more than a year ago, they spanned thermal ceilings from 50 to 150 W, core counts between two and eight, and price points between $188 and $3620 (for the highest-end quad-socket-capable models). Truly, this was a processor line-up with something for everyone. Entry-level small business servers, powerful workstations, rack-mounted virtualization and cloud boxes, and storage appliances are all driven by carefully picked Xeon E5s. 

A transition to the Ivy Bridge architecture sees Intel increment its nomenclature to the Xeon E5-16xx/24xx/26xx/46xx V2 series. A corresponding shrink to 22 nm manufacturing lets the company add more cores and L3 cache without violating the thermal limits imposed last generation. Expect to see 10-core models rated for as little as 70 W, 12-core models that hit 130 W, and a number of TDPs in between. All the while, Intel maintains 256 KB of L2 cache and a 2.5 MB slice of shared L3 per core.

Most of the Xeon E5-2697 V2's specifications are consequently pretty easy to guess. It's a 12-core processor with 3 MB total on-die L2 cache and 30 MB of shared L3. A base clock rate of 2.7 GHz jumps as high as 3.5 GHz with Turbo Boost enabled and just one core active. With two utilized, the clock rate tops out at 3.4 GHz. Frequency drops to 3.3 GHz with three cores active and 3.2 GHz with four. So long as thermals allow, spinning five cores up drops you to 3.1 GHz. From six to 12 active cores, the Xeon E5-2697 V2 tops out at 3 GHz.

As with the Ivy Bridge-E-based Core i7-4960X we previewed, Intel facilitates DDR3 data rates of up to 1866 MT/s. Basic RAS modes and ECC are enabled. You'll see QPI transfer rates of 6.4, 7.2, and 8 GT/s up and down the V2 line-up, but the -2697 V2 specifically sports a pair of links at that top specification.

Chris Angelini
Chris Angelini is an Editor Emeritus at Tom's Hardware US. He edits hardware reviews and covers high-profile CPU and GPU launches.
  • Someone Somewhere
    The 32-bit build of Geekbench uses x87 code

    Typo, top of page two.
    Reply
  • jimmysmitty
    11342601 said:
    The 32-bit build of Geekbench uses x87 code

    Typo, top of page two.

    Where is the typo? Do you mean the x87? That's not a typo.

    This is interesting but not uncommon. The server market needs the boosts while most consumer desktop CPUs are already faster than most software can go.

    Of course in 5 years a SB i5 will be no longer relevant but until then it will serve just fine. Even a x58 i7 is still a viable option for a CPU and its been out for at least 4 years.
    Reply
  • vmem
    Interesting article. Must admit though, while the Mac Pro's performance is certainly impressive, the overall pace of development in the high-end has been rather boring for the past 2 years. can't wait to see what Haswell-E can do late next year.

    "Regardless of whether you love or hate the “wastebasket” design, the system’s specs are very impressive for the volume of space it occupies."

    And this remark touches on the core of the problem. these are a specialized, niche market of professionals who're buying this uber-expensive desktop for PRODUCTIVITY. sure it should look nice, especially in the office of a professional designer. but must it be SMALL? honestly, build a giant aluminum bookshelf if you have to. make it look elegant and artistic, maybe give people some power to customize it's looks, but ultimately give people the ability to customize the machine and buy the level of productivity they need. Apple, you've done some great things, as well as some things that I don't particularly like. but watching you kill the freedom of the small group of designers who love your products is rather sad...
    Reply
  • Someone Somewhere
    11342721 said:
    11342601 said:
    The 32-bit build of Geekbench uses x87 code

    Typo, top of page two.

    Where is the typo? Do you mean the x87? That's not a typo.

    Hmm, on a quick Wikipedia read, x87 was the instruction set used for the floating point instruction sets in the 8087 and later FP co-processors. Interesting.

    Oops... sorry.
    Reply
  • natoco
    In a years time with the haswell refresh and series 9 chipset it will still make everyone yawn even if it was this year. Everything has been going into mobile since Nehalem. On the bright side, phones and tablets will start slowing down very soon once they too reach the same manufacturing node as enthusiast pc's, since the node determines the power envelope achievable, thus mobile is about to hit the same wall.
    Reply
  • CommentariesAnd More
    What I expected for the Mac Pro's CPU was a different CPU optimized for the Mac Pro. Would be surprised if the temps of made by this 12Core beast keep things cool. But hey , this isn't final , right ? Lets hope for the best ( and an affordable Mac Pro :) )
    Reply
  • PreferLinux
    http://www.mouser.com/ProductDetail/Intel/CM8063501288843S-R171/?qs=sGAEpiMZZMvqxsBVy5ZiuowErqth9imUwPY6%2fY0Um1w%3d
    Guess what?
    "Description: CPU - Central Processing Units Xeon E5-2697v2 12 CR 2.7GHz FCLGA2011"

    "Pricing (USD)
    1: $3,249.19
    2: $3,127.04"
    Reply
  • Someone Somewhere
    11343328 said:
    What I expected for the Mac Pro's CPU was a different CPU optimized for the Mac Pro. Would be surprised if the temps of made by this 12Core beast keep things cool. But hey , this isn't final , right ? Lets hope for the best ( and an affordable Mac Pro :) )

    Nobody optimizes CPUs for anything. The set up costs are ridiculous. The closest you'll get is a custom config, like a chip with (for example) both multi-socket support and overclocking or something, but you'd have to show up to intel with a truck full of cash.
    Reply
  • Duckhunt
    as usual the folk running intel have become lazy and stupid and the developments in the desktop have gone down the hole. They just add some extra cache and extra threads and then act like they did something.Wow. ( at the stupidity).

    Instead of pushing out code or getting the rest of the industry to use more threading applications and develop it to make it more stable and useable. Nothing.
    I guess when we have a third world america. You might as well go back to a decade 1368x738 with it being the most popular in 2006. Who can afford it? It the retro push backward.
    Reply
  • Someone Somewhere
    Intel can't really do much about forcing the industry to use more threaded apps.

    Not their job to write code, other than drivers. They do make x86 Android though, because the drivers are pretty much hardcoded.

    Do agree on the 1366x768 though. It's the same number of lines as XGA, just with a few pixels on the side. Maybe Intel should have forced a PPI measurement on Ultrabooks - that might have helped.
    Reply