AMD Bulldozer Benchmarks Leaked

February 14, 2011 Sam (Insideris) AMD, Intel 37

We have just received a photo of what sender claims to be an engineering sample of AMD Bulldozer (Zambezi) Quad Core processor benchmark score table.

Unfortunately, he/she asked not to publish the original photo (it also has other test score, but for some reason we cant show it (don’t ask why)).

Anyway, here is 3D Mark Vantage CPU score chart (Sandy Bridge vs. Bulldozer)

3.5 GHz Bulldozer matches 4.0 GHz Sandy Bridge.

We don’t really care whether you believe benchmark results or not, as we just publish what we get our hands on. As always, a cat will die for every chart repost in forums, chats, etc.

Intel is screwed up until Ivy Bridge arrives.

As for other score, it’s a power consumption test.

  • Tquanda Shinequa

    ohhhhhhhh lordy, theys doned it nows!

    • Kage

      DO WANT TO BELIEVE. OMGGGGGGGGGGGGGGG

  • Quad core or Quad module? If quad core – ok, that’s great. If quad module then we’re talking ’bout EIGHT cores. And if it is really a Zambezi sample – then it’s quad module & eight core version.

    So? Which one?

    • B A.Ware

      “Quad core or Quad module?”
      Not sure it’s relevant. The Bulldozer modules (correct me if wrong) work in a similar fashion to Hyperthreading. The difference being that they use real cores as opposed to simulated ones.
      The 2600k has hyperthreading, so for these benchmarks, it’s probably safe to assume they are comparing a 2600k (hyperthreading enabled) with a Bulldozer 4 module.

      • fatty2nd

        It is relevant, unlike Hyperthreading, each module consists of 2 cores, not a single core that can run 2 simultaneous threads. It’s completely different to Hyperthreading.

        • zd

          Not as much as you would think. They are similar regarding the shared frontend (L1I, prefetch, instruction decoding and dispatch), core interface unit and L2. The difference is that with AMD’s solution the regular integer thread execution is more separated (has a separated scheduler, retirement and L1D).

          I don’t know why it’s better to them to refer to their chip as 8-core, instead of 4-core/module.

          • Jörg

            They (AMD [in form of JF-AMD]) stated that they _only_ use the name module to refer to the architecture but dont want it to become mainstream also a module is perse a dual core so it would be stupid to label it as anything different like hyperthreading, as a module has two integer units and by definition one integer unit is refered to as one core so if they say 4 core then they definitly mean 4 cores and not 4 modules.

    • Omar

      There will be Quad Cores(FX-4110),Six Cores(FX-6110) and Eight Cores(FX-8110 and FX-8130P).The TDP of X110 processor will be 95W.The TDP of FX-8130P processor will be 125W

  • Also, for god sake, Zacate is the name of BOBCAT based chip, not BULLDOZER.

  • Klepteaux

    It doesn’t say “Zacate” anywhere, so what the hell are you going on about, now?

  • Bloodcore

    Vithren, what are you on about?
    Did I miss something?

    Because I can’t see Zacate mentioned anywhere at this page. :|

  • Zacate was a typo and I apologize for that, it is now fixed.

    As for modules/cores, this is all what we have received.

  • inf64

    AMD would never market modules. They market cores.So if slide mentioned modules it’s not real. If it mentioned Quad Core Zambezi,the chances of it being real are higher. Mind you I’m not speaking about the very scores this source is “revealing” .That’s a whole other issue.
    If it’s real then X8 will be challenging 8 core SB-EN.

  • Why are there 2 measurements for Zambezi (w/ and w/ Turbo)?
    How could HW-based Turbo actively be disabled to make measurements?

    And why is the higher Zambezi score _exactly_ 17,001% higher than the lower one (while Turbo freq. is 20% higher). That’s as if it has been _calculated_ by multiplying the lower number with 1,17 and rounding (try out).

    And why does the benchmark scale at 90% (SB) resp. 85% (Z) while in other tests it scales at factors around 80%?

    ;)

    P.S. why are 3 of 4 letters in “leak” the same as in “fake”?

    • If you dig deep enough, I am sure there are tons of “evidence” to support your claim, just like flat earth or any other argument.

      Just for the record, we are not claiming that those benchmarks are real and just publish what we receive.

      However, when we get our sources, we clearly say so put trust/weight on those “claims”.

      Only time will tell.

      • shadowmaster625

        Dresdenboy is right. There aint no way two benchmarks are ever going to be exactly X integer percent apart than each other. Bogus numbers.

  • Ignacio

    If this is real, then octocore Bulldozer is 69% faster than core i7 980x not 65% and quad core Bulldozer at the same clock speed is 12,4% faster than Sandy Bridge 2600k.
    core i7 980x score is 32438

    • wICE_man

      Obviously we are talking about 8 core / 4 module Zambezi and that should be obvious to anyone who has been watching the news on this architecture. First of all, Zambezi will launch in 8 Core / 4 module versions and then add 6 Core / 3 module versions. This is a well established fact by now.

      Thus there would be no 4 core / 2 module engineering samples out there. People that are whining over 8 Core AMD vs. 4 Core Intel should get a life and stop looking at hardware sites, they simply are not getting the point. Architecture is about squeezing out as much performance out of as little die area as possible in a convenient fashion.

      Sandy Bridge 4C is 903M transistors and weighs in at around 216mm² @ 32nm and we know directly from AMD that each 2 core module is 213M transistors or about 30,9mm² @ 32nm. That times four plus 8MB of L3 cache should be around 1.2B transistors or about the same as 6 core Gulftown and the die area should be about 240mm². To contrast a Phenom II X4 is about 258mm² and a Core i7 Lynnfield is about 296mm². So the 32nm contenders will be Sandy Bridge and Gulftown vs. Zambezi at least until the arrival of Ivy Bridge.

      These numbers sure look promising :)

      • Hal

        well I remember AMD saying that Zambezi will be smaller than 6 core Phenom II so you are right on that aspect, but Ivy Bridge comes after 8-core Sandy Bridge which is sheduled for a Q4 2011 Release Ivy Bridge is atleast one Quartal after that. Also I dont think I can follow you logic you are saying “Screw the Quad cores bring the Octocores” even though almost NOTHING profits from more than 4 cores, especially games will take long until they find a way to utilisize all 8 cores, the Only thing that currently needs more cores than 4 is video encoding and I seriously dont believe that people will jump on Bulldozer just because you can encode Videos faster than what Sandy Bridge can offer (also note Quick Sync s-cks), Consider this : Buldozer gets relseased but has a lower clock per clock performance than Sandy Bridge, AMD would be freaking screwed as Intel would release Sandy Bridge Octocores later this year anyway.

        • wICE_man

          Hal, i never said anything resembling what you claim and you are not taking into consideration a very important aspect of this particular design.

          In fact the beauty about Zambezi is the fact that these 8 cores are distributed over 4 modules. Each module has its separate L2 cache, floating-point unit and front-end which are shared for two cores. When we are using just 4 cores (or less) the Zambezi can opt to use only one core out of each module, giving him all the L2 cache, floating-point capability and instruction issuing, making said core much more efficient plus it can also clock said core higher to increase the performance even more which is exactly what you are seeing in this particular benchmark.

          Thus a 8 core / 4 module Zambezi will always work well, even on a thread limited workload. This is a benefit to their approach. An 8 core Sandy Bridge will be an enormous beast at about 1.8B transistors (if they up the L3 cache to 16MB as rumors suggest) and with a die area of about 350mm². So it will be more expensive to make and the added cores will not affect single-threaded performance at all while the doubling of the L3 cache will help only slightly.

          Chances are Zambezi will best Sandy Bridge in single threaded workloads. They have equal branch prediction capabilities, equal instruction fetch/dispatch capabilities, twice the L1 cache with about 33% more bandwidth, four times the L2 cache per core and an arguably more flexible Floating Point unit (although it’s theoretically only capable of half the number of 256-bit Floating Point calculations). Even with a massive 16MB L3 cache, an 8 core Sandy Bridge will only have 18MB of total cache vs 16MB of the Zambezi, thereof 8MB L2 cache vs 2MB L2 cache of the Sandy Bridge.

          If so then we will have exactly opposite landscape to the one most people have been predicting, i.e. Zambezi will excel at less threaded benchmarks while Sandy Bridge may be more powerful in heavily threaded applications and benchmarks.

          But of course these are just speculations at this point although they seem to fit these leaked benchmarks which i am inclined to believe.

          • Hal

            well I actually agree with most of what you said it seems like I really misinterpreted your comment.

        • Anonymous

          BF3 will fully utilize 8 cores. After 8, the scaling will drop, but it will still benefit for a few more. BF3 comes out this fall.

          BF3 will use extra cores to help with simpler physics and also pre-calculate information about certain types of light sources as to optimize the lighting data before it gets to the GPU. I guess you can waste some CPU horse power to dramatically improve the efficiency of the GPU.

          Unreal, Battlefield and Tech5 gaming engines will all make use of 8+ cores and all will be out in the next 6-12 months.

          If you think more than 4 cores is a waste, then you can’t see past your nose.

          • Hal

            basically what I am saying is that it is a waste to invest in _the first_ Gen. Octocores, as no game right now is using Octocores and when these Octocores here are out only a few Games will support Octocores If at all, and if the support is there new Octocores will be to, sure if you have the money to change your rig like your underpants go ahead, but not everyone can afford a new rig every new Gen. . Also are you sure a games using the new unreal engine will be out in the next 6-12 months? thats just laughable, I dont think thats possible…

          • Anonymous

            You make it sound as if computers are expensive. I paid $800(plus tax. was from Best Buy) for my i7-920/4gb/ati4850/lcd(1080p), with 8 threads, 2 years ago.

            Also, you said not to purchase an 8 core as if someone purchasing a new computer wouldn’t benefit from it. Yes, if you have a quad core, don’t waste you money right now, but if you need to upgrade anyway, get an 8core/thread cpu, they only cost and extra ~$50.

            Also “…especially games will take long until they find a way to utilisize all 8 cores…” and ” Also are you sure a games using the new unreal engine will be out in the next 6-12 months?”

            Like I said, Rage and BF3 will be out this fall even if UE3 isn’t.

          • Hal

            I highly agree with what you said in the second paragraph of course Octocores are only a waste of money if you already have a Quadcore, but anyone with a Dualcore will profit from investing in Bulldozer Octocores, and maybe even Hexacores for that matter.

    • Ignacio

      Quad core Zambezi with 4mb l3 and turbo core are real.
      http://blog.pccomponentes.com/wp-content/uploads/2010/11/133a-1.jpg

  • Pingback: Playmagazin.eu()

  • Praesepe09

    If this is real and if i were AMD , ill show everything about bulldozer

    Cuz There are Many people to buy sandy after mainboard problem fixed

    Maybe this bench is similar thuban vs i5 under specific situation

  • Guest++

    For get about this. More cores does not mean faster multi-core processing on single application ;)

    Its nearly not possible to achieve 100% scalability. The more cores you use, the lower scalability gets.

    Just an example
    cores:
    1: 100%
    2: 195%
    3: 260%
    4: 300%

    etc

    I’m experienced with multi-thread coding myself

    • Exgaves

      what you dont understand is bulldozer is able to apply 2 of the ‘cores’ in the same module into one thread. Hows that for scalability on single threading.

      • Mike98

        are you sure? never heard about that, can you give a source?
        also didnt JF-AMD himself say that bulldozer has a lower overall bandwidth than a real Octocore? i.e. on bulldozer you get 720% bandwidth with 8 Threads as opposed to 760% you would get on a non-module arch like sandy bridge S2011 Octocore.

        • The point is for these modules to get so cheap, that when intel is releasing x-core CPUs and AMD can be releasing x-module CPUs

      • Bruno

        what you don’t understand is
        i’d catch a grenade for ya
        throw my hand on the blade for ya
        sdofj0padhg-jsdagadg

      • Helbrecht

        This statement is not true. JF-AMD has clearly stated that there is no such feature on bulldozer. One core will only execute ONE thread and each ONE thread will be executed on one dedicated core.

        • That is true but there was a caveat which was the fact that it would “disable” or “not use” the shared core so one core could utilize all resources in its module which significantly improves single core performance.

  • Helbrecht

    These numbers might very well be true for a dualmodule zambezi. I’m very certain that there are engineering samples of both dual and tri module zambezis out there.

    Even though the buldozer features shared resources, it still has more beefy resources per thread and core than sandy bridge. Well yes, all bandwidth is shared within the module, but with the separated branch predict/fetch stages and 4 x86-64 decoders, the fetch que will theoretically never stall. And a bulldozer integer core also has one extra pipeline vs sandy bridge, wich should in theory beef up the integer performacne quiet a bit.

    So in terms of integer performance, bulldozer will allmoast certainly outperform sandy bridge in single thread performance and an 8 core bulldozer might even be on par with an 8 core sandy bridge with hyperthreading turned off, and still be use less power.