Sign in with
Sign up | Sign in

AMD ''Interlagos'' Bulldozer Benchmarks Leaked

By - Source: Softpedia | B 66 comments

An independent AMD server partner has leaked benchmarks of the upcoming Bulldozer-based Interlagos chip in a dual-GPU setting.

Various reports have indicated that benchmark results for AMD's Interlagos Bulldozer-based server processor in a dual-CPU setup have appeared online. The testing was done using Arch Linux on the Linux 2.3.67 kernel, the Phoronix Test Suite loaded on a 2 TB SATA drive, and 64 GB of DDR3 memory shoved in a dual-socket Supermicro H8DGU server motherboard. Each CPU packed 16 processing cores and a clock speed of 1.8 GHz.

The testing was supposedly submitted by an independent AMD server partner, listing the new 16-core Interlagos Opteron chip as "AMD Eng Sample ZS182045TGG43_28." This particular processor has been benchmarked before with SuSE Linux 11, Ubuntu 10.04, CentOS 5.4 and Arch Linux, but not in a dual-CPU setup as seen here.

Only a few select benchmarks from the suite were actually performed in this particular testing including C-Ray, Himeno, SciMark 2, Parallel BZIP2 Compression and the Stream memory tests. For starters, the system scored a C-Ray time of 25 seconds, showing to be quite faster than the Intel Core i5 2500K (quad-core + Hyper Threading; 3.3 GHz + 3.7 GHz Intel Turbo Boost) at 61 seconds or the Intel Core i7 970 (six cores + Hyper Threading; 3.2 GHz Base Frequency + 3.46 GHz Turbo Boost) at about 61 seconds.

The system didn't do quite as well in the Himeno test (which is explained to be a linear solver of pressure Poisson using a point-Jacobi method and is not too multi-thread friendly). According to OpenBenchmarking.org's testing, it scored just 88 MFLOPS due to its lower clock speed. The system did better in the Parallel BZIP2 Compression test (v1.0.5) however, packing a 256 MB file in 6.27 seconds (with a standard error of 0.10 and a standard deviation of 7.49-percent).

"There is also other result uploads from the 'AMD Eng Sample ZS182045TGG43_28' for SciMark 2 computations and Stream memory tests," the site claims. "More may also come, but these are just the results so far we have been able to verify as valid/authentic."

When it finally hits the market, AMD's Interlagos will come packed with 12 or 16 cores, a quad-channel memory controller, up to 16 MB of level 3 cache memory, and use the same motherboard socket (G34) as AMD's current Opteron 6000-series processors. Interlagos will also be built by Globalfoundaries using the 32-nm SOI technology.

AMD is expected to release its Bulldozer-based server chips by Q3 2011.

Display 66 Comments.
This thread is closed for comments
  • 3 Hide
    jprahman , March 24, 2011 9:59 PM
    If these benchmarks are true then it sounds like Bulldozer is a lot what I'd expect it to be, great multi-threaded performance for server and content creation applications, but somewhat weak IPC, which doesn't bode well for the type of gaming performance we could see.
  • 1 Hide
    Anonymous , March 24, 2011 9:59 PM
    WOW ? , bad news for intel .
  • -7 Hide
    Anonymous , March 24, 2011 10:06 PM
    "32 cores" 1.8ghz ...... to be about 70% faster than 3ghz+ i5 4core or i7 6core..... in a threaded test
    FAIL!?
    i always had amd... might be time to change....!
  • 9 Hide
    mikem_90 , March 24, 2011 10:13 PM
    Not sure why they compared it to an i5 and i7...and not a proper Xeon, which is what this CPU here would be facing off against in the same tasks.

    If you dig through the pages a bit, they show that they did some testing on the 4x Xeon 6 Core CPUs, which does perform better (13.47sec vs 25.97sec)

    This will still be interesting to see how it turns out though. Bulldozer might not be the end all CPU against Intel right now, but it might give a good run for the money I wager.
  • 0 Hide
    bison88 , March 24, 2011 10:19 PM
    That is kind of sad. AMD has the mindset of "throwing more cores" to try and out perform Intel's engineers it seems. While it is kind of awkward to see a handful of 6 core chips from Intel and far more dual and quads, they perform extremely well with the lower numbers as they maximize each core to its extent every time they release a new generation. The reason Intel doesn't seem to be pushing 6+ cores in both desktop and server environments seems to be they are quite happy with the dual and quad core setups they already have, and having used a Xeon Quad and an AMD 12-core setup I can say Intel performed better, at least for me.
  • 5 Hide
    alikum , March 24, 2011 10:31 PM
    @Bison88, you seem to be a little confused with the term cores. AMD's cores does not equate Intel's cores. Desktop version of BD is similar to i7, instead of giving it the term "cores", they call it "modules". They are both engineered differently, how can you compare core to core? What about i7? It has HyperThreading. How are you ever going to have a fair comparison? The only legit way to compare is how they price their product and the performance they give.
  • 2 Hide
    sabot00 , March 24, 2011 10:36 PM
    doctorpink"32 cores" 1.8ghz ...... to be about 70% faster than 3ghz+ i5 4core or i7 6core..... in a threaded testFAIL!?i always had amd... might be time to change....!

    Probably the test doesn't scale all that well, if the cores are as powerful as Sandy Brige's then that means
    the 4 3.7GHz Sandy Bridge cores = 8 Bulldozer cores at 1.8Ghz, throw in immature hardware, 12 cores.
    Don't forget the Bulldozer's time of 21 seconds is only 34% of the SB.
  • 6 Hide
    Anonymous , March 24, 2011 10:58 PM
    Yep, lets all jump to conclusions, because hey, 1.8Ghz is the fastest AMD will go - it's not like it has chips twice as fast (Hz) on the market already...

    We won't know until closer to release, this is just random data that may or may not be indicative.
  • 0 Hide
    dragonsqrrl , March 24, 2011 11:06 PM
    alikum@Bison88, you seem to be a little confused with the term cores. AMD's cores does not equate Intel's cores. Desktop version of BD is similar to i7, instead of giving it the term "cores", they call it "modules".

    Are you sure you aren't the one who's confused?... the 16 "core" interlagos is made up of 8 "modules", just as the desktop variants of Bulldozer referred to as 8 core processors have 4 modules. AMD has never equated the definition of a "core" with what it refers to as a "module". AMD refers to the single module version of Bulldozer as their dual core part. This is quite a bit different from any other x86 based design, and works in a very different way than Intel's hyperthreading.

    And in any case, bison88 wasn't referring to bulldozer when he compared a 12-core AMD processor to a 4 core Intel. In comparison to Bulldozer, AMD's current definition of a core, is very similar to Intel's. Intel simply implements threading within it's cores, enabling each core to run two threads in parallel. So correct me if I misunderstood you, but at least from what I've read bison88 seems to have a better understanding of the term "core" then you do.
  • 0 Hide
    Repelsteeltje , March 24, 2011 11:11 PM
    2.3.67 kernel? Hehe... I guess that would be the 2.6.37 kernel...
  • 2 Hide
    greenrider02 , March 24, 2011 11:25 PM
    dragonsqrrlAre you sure you aren't the one who's confused?... the 16 "core" interlagos is made up of 8 "modules", just as the desktop variants of Bulldozer referred to as 8 core processors have 4 modules. AMD has never equated the definition of a "core" with what it refers to as a "module". AMD refers to the single module version of Bulldozer as their dual core part. This is quite a bit different from any other x86 based design, and works in a very different way than Intel's hyperthreading.


    I think you are holding a few misconceptions as well, as it seems you're trying to compare one of Intel's core's to one of AMD's modules, and while the idea of enabling parallelism by having multiples is the same, the actual hardware and design of the new AMD modules is VERY different from what we've come to see as a core in Intel and previous AMD microprocessors in that the module is divided into "cores" that have different dedicated functions. Making these cores pairs in modules is an organizational necessity for the design of processor datapaths and parallelism. Intel's QuickSync module is a core in the same way with its dedicated function.

    TL;DR The only point is that the definition of a core is getting convoluted and mixed up because of the advancements of the various processor designs of the competing companies.
  • 1 Hide
    Anonymous , March 24, 2011 11:36 PM
    sabot00Probably the test doesn't scale all that well, if the cores are as powerful as Sandy Brige's then that means the 4 3.7GHz Sandy Bridge cores = 8 Bulldozer cores at 1.8Ghz, throw in immature hardware, 12 cores.Don't forget the Bulldozer's time of 21 seconds is only 34% of the SB.


    I think you misread or miscount. dual SOCKET = 2 cpu
    EACH amd cpu have 16 cores
    =32 cores


    70% "FASTER"(very approximate) is YOUR 34% of SB

  • 0 Hide
    mapesdhs , March 24, 2011 11:38 PM

    Kevin, which C-Ray test are you referring to??

    Ian.

  • 0 Hide
    PreferLinux , March 24, 2011 11:43 PM
    Slow! Been out for ~3 days!
  • -2 Hide
    K2N hater , March 24, 2011 11:53 PM
    We just need decent overclocking boards and unlocked multipliers to make it fry eggs and leave SB far behind. I'm expecting it to go past 3.6GHz on air.
  • 0 Hide
    dragonsqrrl , March 24, 2011 11:56 PM
    greenrider02I think you are holding a few misconceptions as well, as it seems you're trying to compare one of Intel's core's to one of AMD's modules, and while the idea of enabling parallelism by having multiples is the same, the actual hardware and design of the new AMD modules is VERY different from what we've come to see as a core in Intel and previous AMD microprocessors in that the module is divided into "cores" that have different dedicated functions.

    umm... I'm a bit at a loss for words. I'm sorry if i didn't state it clearly, but this is exactly what I was trying to get at.

    "I think you are holding a few misconceptions as well, as it seems you're trying to compare one of Intel's core's to one of AMD's modules"
    No... I don't believe I was trying to do this anywhere... In the second paragraph I was contrasting AMD's Bulldozer definition of a core with Intel/AMD's current definition.

    "the actual hardware and design of the new AMD modules is VERY different from what we've come to see as a core in Intel and previous AMD microprocessors"
    ... I did say, "This is quite a bit different from any other x86 based design, and works in a very different way than Intel's hyperthreading"...lol

    All I'm trying to explain is that AMD's definition of a "core" in Bulldozer is different from current uses of the term, which isn't the same as its definition of a module. I think this was pretty much exactly the point you were trying to make, was it not?
  • -2 Hide
    PreferLinux , March 24, 2011 11:58 PM
    K2N haterWe just need decent overclocking boards and unlocked multipliers to make it fry eggs and leave SB far behind. I'm expecting it to go past 3.6GHz on air.

    SB can overclock, and very well (5 GHz on air, for example), remember. Also, don't forget SB-E is coming out Q3-Q4.

    As said by others too, 1.8 is no indication of final speeds, either.
  • 0 Hide
    Anonymous , March 25, 2011 12:05 AM
    actually if i recall bulldozer may not be ideal for heavy FPU intensive applications (as employed by grid computing) each module (2 cores) use a shared FPU which can schedule 2 x 128bit operation per cycle (plenty big for desktop and gaming needs) but only one 256bit operation, any kind of solver will employ full 256bit FPU operations to obtain maximum accuracy, therefore in reality reducing 16 cores to 8

    bulldozer architecture actually suits desktop application better then server with this regards whereby a single core can happily schedule two FPU operations in one cycle (especially if the apps can not utilize both cores of a module), moving away from interlargos to desktop bulldozer will yield faster speeds which would really compound the advantage of this setup, IMHO bulldozer was not a server architecture
  • 0 Hide
    JonnyDough , March 25, 2011 12:41 AM
    This is an engineering sample. It's also clocked at the lowest of most processors these days. With final production it should be able to get over 3.0ghz and then we can compare oranges to oranges.
  • 0 Hide
    jimmysmitty , March 25, 2011 12:49 AM
    mikem_90Not sure why they compared it to an i5 and i7...and not a proper Xeon, which is what this CPU here would be facing off against in the same tasks.If you dig through the pages a bit, they show that they did some testing on the 4x Xeon 6 Core CPUs, which does perform better (13.47sec vs 25.97sec)This will still be interesting to see how it turns out though. Bulldozer might not be the end all CPU against Intel right now, but it might give a good run for the money I wager.


    Thats my question. Why are they comparing, and I quote:

    Quote:
    AMD's Interlagos Bulldozer-based server processor in a dual-CPU setup


    A dual CPU setup to a single CPU setup?

    And most of all its comparing 4 cores 8 threads (we all know SMT is nice but only gives around 20% performance boost) or 6 cores 12 threads to 32 real cores.

    Talk about lop-sided.

    As for clock speed, its no indication but AMD has always had lower clocked server parts for some reason. A ES sample tends to be the best of the pre final chips so it should be able to overclock better than final ones.

    Still, comparing server chips to desktop CPUs. Thats just stupid.
Display more comments