AMD Bulldozer Zambezi FX-8100 benchmarks from SiSoftware

On September 3rd, SiSoftware published a review of AMD’s new Zambezi processor, based on the new Bulldozer architecture.

After only about 4-5 hours, SiSoftware had to take it down again, but what goes on the internet, stays on the internet, it was already cached by Google: http://webcache.googleusercontent.com/search?q=cache:QLn9rR2rn5oJ:www.sisoftware.net/?d=qa&f=cpu_amd_bulldozer&l=en&a=+sisoftware+bulldozer&cd=2&hl=en&ct=clnk&gl=en

AMD, in the person of John Fruehe, was quick to dismiss the benchmarks as fake… But how fake are they?

  • AMD started shipping Bulldozer CPUs on the 7th of September, only a few days after these benchmarks were published. So while AMD claims that all benchmarks are done with pre-production silicon, BIOS, microcode and more… how pre-production are they? (this late in the process, I would guess that pre-production systems would be within about 5% of the final product).
  • SiSoftware is not just any random site that leaks some random benchmarks. SiSoftware has been developing benchmarks for many years, and has been a popular benchmarking tool for many high-profile review sites for a long time. What reason would SiSoftware have to leak false benchmark figures? It would only hurt their core business (which is selling benchmark software, not getting website hits).
  • It seems too elaborate to be a total fake… If it is not actually SiSoftware who published this page, then it would be a pretty impressive hack to get such a page online, and in the Google cache. Also, SiSoftware has not denied that this page is real.
  • The figures, as well as the accompanying analysis seem to make a lot of sense. This is not just some random guessing. It is written by someone who at least understands the Bulldozer architecture and the performance characteristics. It makes perfect sense that it is slower than Sandy Bridge in most tasks, and peaks at the crypto benchmark with ALU-only. After all, the Bulldozer has 8 integer cores, with 2 ALUs each, where Sandy Bridge has only 4 cores, with 3 ALUs each. So that is the one benchmark where you would expect Bulldozer to excel. Everything else seems to be exactly how you expect things to be between Intel and AMD, historically: Intel has the better memory controller, cache and SIMD performance. Those aren’t gaps that I would expect AMD to close in just a single generation.

Below is a capture of the entire page, in case the Google cache gets invalidated in the future.

About these ads
This entry was posted in Hardware news and tagged , , , , , , , , , , . Bookmark the permalink.

22 Responses to AMD Bulldozer Zambezi FX-8100 benchmarks from SiSoftware

  1. Nuts says:

    Firstly, I believe that if you wanted to, you could submit completely bogus data to SiSoftware. Intel would, I bet, have a valid motivation to try this.

    Secondly John made a general comment that you are now using to justify that this one set of benchmarks as originating from a bulldozer equipped system. The correlation of cause and effect here is not demonstrated.

    Thirdly, if these processors where really so bad, why would Cray be one of the big initial customers.

    In the end this is blog is supposition, without substance!

    • Scali says:

      “Firstly, I believe that if you wanted to, you could submit completely bogus data to SiSoftware. Intel would, I bet, have a valid motivation to try this.”

      Please read the entire page.
      This is not just some results submitted to SiSoftware. It is SiSoftware themselves testing with a Bulldozer CPU, publising the results and an analysis on their web page.

      “Secondly John made a general comment that you are now using to justify that this one set of benchmarks as originating from a bulldozer equipped system. The correlation of cause and effect here is not demonstrated.”

      No idea what that is even supposed to mean.

      “Thirdly, if these processors where really so bad, why would Cray be one of the big initial customers.”

      Oh dear, do I see a lack of correlation here?
      Cray was also a big initial customer of Barcelona. And we all know how great that chip was!

      “In the end this is blog is supposition, without substance!”

      I put this blog up because I have reason to believe that SiSoftware’s page is not as fake as AMD wants us to believe. We will find out the truth when final Zambezi’s end up in the hands of reviewers. But as I said above, I think these figures will probably be within 5% of the final product.

    • Scali says:

      As you now found out, there is a reason why I specifically picked these leaked benchmarks.
      Still thnk they’re fake, and still think Cray bought Bulldozer because it’s so great?

  2. despoiler says:

    FAKE! They look to be pitting a FX-8100 that has the wrong turbo speed (it should be 3.7ghz) vs a non-existent Sandy Bridge part. There is no SB chip that runs at 3.0/3.6ghz. Also, why would you run DDR3 1333 when Bulldozer supports 1866 DDR3? This is BS and took me all of 5 minutes of research to get this info.

    • Scali says:

      There could be plenty of explanations for that.
      3.7 GHz is a rather strange speed, especially if the base frequency is 200 MHz, as SiSoftware says. You’d need to have fractional multipliers. I’ve seen other sources claiming 3.9 GHz for the FX8100 turbo speed. Equally suspicious. 3.6 GHz or 3.8 GHz would make a lot more sense.
      It could also be that the turbo is slightly different on their pre-release sample than on the final product.
      It could even be that their software just has a bug detecting the new CPU, and they didn’t catch it. Or that they just made a typo when writing this page.

      Likewise, for the Sandy Bridge CPU, there is no version at 3 GHz, you say? Well no, but you don’t HAVE to run the CPU at stock speeds obviously.
      So it could be an overclocked or underclocked CPU (they may not have had any desktop SB’s available, so they overclocked an SB to be around the same speeds as the Zambezi… slightly higher base, but slightly lower turbo, so about equal on average).

      The memory could also be at 1333 because it’s a mobile version. Or perhaps because the Zambezi pre-production sample was not yet stable at higher memory speeds.

      • looncraz says:

        AMD has long used “fractional” multipliers. A multiplier of 18.5 is not unreasonable or unexpected for an AMD product.

        Using DDR3-1333 to benchmark a system is highly suspect and unusual. Normally you run a system the way it is designed to be run when doing a benchmark – something that SiSoftware normally does…

        There is no mention of under-clocking the sandy bridge, as would have been expected if they had done so. Also, it would thus make bad comparison – and, once again, goes outside the realm of other SiSoftware benchmark articles where comparing two competing CPUs.

        If you are comparing two designs head-to-head, you normally match clocks…

        Indeed, the absence of model #s, detailed motherboard information, and a misunderstanding of the Bulldozer FPU arrangement suggest this is very much fake – or poorly done. Performance in 8-threaded 128-bit FPU should more or less equal the performance of 4-threaded AVX on Bulldozer. Here, it is cut in half because the inventor of the article doesn’t understand that the FlexFPU can do 8 threads in 128-bit mode (though they eluded to such in the beginning) and that the results are NOT in line with expectations.

        The fact that the review’s numbers reflect the 10h CPUs relative performance characteristics is highly suspect considering Bulldozer uses a different design altogether and would be expected to act very differently. It has 2 ALUs & AGUs, new predictor, different L1, 4x the L2, improved memory controller, and much more and this review does not reflect those changes.

        Even more issues arise when one considers AMD claim of a 35% performance boost from using Bulldozer. If that is to include the extra 33% more cores, then you would expect IPC to remain the same, however this “review” shows Bulldozer running 15-20% slower than Phenom II.

        However, one should also note that JF-AMD had been VERY keen on use of the word “performance.” JF-AMD has always refused to answer that question, saying performance isn’t the same as throughput. The throughput improvement is 50%. 35% better performance would suggest IPC improvements, which JF-AMD has claimed should be expected.

        If this is a real review, it is garbage! There is very little information about the system configuration, results are not in line with the chip design ( the benchmarks suggest that the FlexFPU is not working with two threads at all ), and the expectations of the author are not inline with the chip design (which oddly mate up with the actual “results ). If the review is done with an ES CPU, then it is still completely junk anyway.

        AMD prices CPUs relative to Intel’s pricing, so the 8 core Zambesi should perform somewhere on par with the i5 2500 as they are priced similarly. AMD wouldn’t even release a product that performed this poorly against its OWN offerings, they would just delay again (what is another quarter compared to 3 years??).

      • Scali says:

        “AMD has long used “fractional” multipliers. A multiplier of 18.5 is not unreasonable or unexpected for an AMD product.”

        Not saying it isn’t possible.

        “Using DDR3-1333 to benchmark a system is highly suspect and unusual.”

        I don’t see why. Even if you consider it unusual, I don’t see why it is suspect, given that both are using DDR3-1333 spec. In fact, if you’re faking anyway, why even bother to go for less?

        “Also, it would thus make bad comparison”

        Depends on what you’re trying to compare. If you want to compare architectures clock-for-clock (which is the first thing that I’d want to test when a new architecture comes out), you will have to manually adjust clockspeeds every now and then. Given that turbo modes make it difficult to do exact matching of clockspeeds, I can see why the clockspeeds are the way they are, as I’ve already argued earlier.

        “Performance in 8-threaded 128-bit FPU should more or less equal the performance of 4-threaded AVX on Bulldozer.”

        Not at all actually. AVX-256 will generally give you slightly better memory access and slightly lower latency. Aside from that, it mostly depends on how well the new FPU scheduler does its job.

        “It has 2 ALUs & AGUs, new predictor, different L1, 4x the L2, improved memory controller, and much more and this review does not reflect those changes.”

        As I’ve written earlier, effectively the Bulldozer has less ALUs per thread than other AMD and Intel architectures. So that is something they will have to compensate for. Aside from that, just because things are new/changed is no guarantee that they are faster (Pentium 4 ring a bell?)

        “Even more issues arise when one considers AMD claim of a 35% performance boost from using Bulldozer. If that is to include the extra 33% more cores, then you would expect IPC to remain the same, however this “review” shows Bulldozer running 15-20% slower than Phenom II.”

        Ouch… I’ve already covered this in detail in earlier posts… But the short version is:
        1) Don’t believe AMD’s claims. AMD also claimed Barcelona was 40% faster than any other x86 out there, including Kentsfield. Were all benchmarks fake?
        2) (As you ironically mentioned yourself, with FlexFPU above,) It is quite difficult to extract per-core IPC performance from Bulldozer performance estimates, since we don’t know how well the shared resources work.
        3) Performance is dependent on more than just IPC. Clockspeed is another factor, and then there are various other bottlenecks, such as cache synchronization among cores, memory etc… all of which will have different characteristics on Bulldozer than on other AMD architectures. We know at least that part of the ‘35% performance’ estimate comes from higher clockspeeds of Bulldozer, and not from IPC, since Bulldozer clockspeeds are higher (both stock and turbo).
        4) Considering that Bulldozer has less ALUs and AGUs per core, and has to share its FPU between two cores, it is actually expected that it will have lower IPC than Phenom II. The question is more: has AMD been able to increase performance enough to compensate for the lower resources per core?

        “the benchmarks suggest that the FlexFPU is not working with two threads at all”

        Which may well be the case. AMD was also talking about how their native quadcore design would make core-to-core communications highly efficient. But when it came down to it, benchmarks showed little to no difference between a ‘native’ quadcore, and a two-socket system with two dualcores.
        Designs don’t always work as well in practice as they were intended.
        Aside from that, the lack of a Bulldozer-aware scheduler in Windows 7 may also have an effect on this. While AMD claims that this invalidates the benchmarks, I would argue against that: there *is* no such scheduler available at this point. You have to work with what you have. Same goes for Bulldozer-optimized software: none is available. If you buy a Bulldozer today, this is the performance you’re going to get. It is up to AMD to make Microsoft release an updated scheduler for their CPUs, and to convince software vendors to release Bulldozer-optimized updates for their products.

        “AMD prices CPUs relative to Intel’s pricing, so the 8 core Zambesi should perform somewhere on par with the i5 2500 as they are priced similarly.”

        But the CPUs are not released yet, prices could still change.

        “AMD wouldn’t even release a product that performed this poorly against its OWN offerings, they would just delay again (what is another quarter compared to 3 years??).”

        The CPUs aren’t released yet… they could still delay them (that is, to the consumer market. They started shipping server/workstation chips, probably because they are bound by contracts. They ran into similar trouble with OEMs when Barcelona was delayed.. they decided to ship the TBL-bugged Barcelonas to OEMs anyway).

      • Scali says:

        “AMD wouldn’t even release a product that performed this poorly against its OWN offerings, they would just delay again (what is another quarter compared to 3 years??).”

        Guess what? It DOES perform poorly against its own offerings, and they released it anyway?
        You think I am just a moron who risks his reputation by calling out JF-AMD on his bogus claims, and referring to this leaked benchmark (but not others)?
        No, I’m not as stupid as JF-AMD. I don’t risk my credibility and reputation by spreading nonsense.

  3. Bonzai says:

    We are about to find out real soon. But if that is real? Fuxor. Crash and Burn. Unless overclocking it makes it soar like mad (Which is something to consider). Ah well, we get to see it soon! I just would like to see more processors that can get OC’ed have less heat (Ha pipedream). My i7-920 is a hottie @ 4.2. But I really don’t want it to fail. Competition is good.

    • Scali says:

      Well, i7 920 is ancient by now. The 800 series already lowered power/heat quite a bit, while maintaining similar performance levels. And Sandy Bridge gives you a lot more performance at the same clockspeed and power. Thanks to 32 nm, it should also clock even higher. So there’s plenty of improvement from an i7 920 already… and before long we will have 22 nm CPUs, should give even more improvements :)

      Some things I wonder though:
      If Bulldozer is so good, and all benchmarks are fake, why haven’t we seen any GOOD benchmarks being leaked? Or even officially released by AMD as a preview?
      Whenever a new Intel architecture is near, Intel just calls up a meeting and demonstrates the CPU to the public, a few months ahead of time. And review sites will generally be sent actual hardware a few days or even weeks before launch.
      Take this for example:
      http://www.anandtech.com/show/3871/the-sandy-bridge-preview-three-wins-in-a-row
      Months before Sandy Bridge was released, AMD was allowed to use an engineering sample to post a full preview of Sandy Bridge. Note also that even though this was months earlier, performance was already pretty much the same as the released product (and no crying from Intel about how you need the latest BIOS updates, latest microcode, latest OS updates, and specially optimized software. Because that’s all nonsense).
      Same story with Conroe a few years earlier, for example:
      http://www.anandtech.com/show/1963
      (And people were completely in denial about that… benchmarks looking too good to be true they said… but they weren’t. Even the pre-release stuff worked very well indeed).
      I’m sure AMD would do the same, if they had anything to demonstrate.

      • Bonzai says:

        I like the fact that they just put up a publicity stunt with the FX 8-core hitting a clock of 8ghz with Liquid Nitrogen. That’s great…..BUT I WANT TO SEE NUMBERS! If it rocks some socks and is quite easy to overclock (Without using Nitrogen) then now we are talking. I still hate how the CPU’s have the pins on them though.

      • Scali says:

        What’s the point of overclocking when you don’t run any benchmarks to show how fast the system is?
        We already knew that Bulldozer has longer pipelines, so it would naturally clock higher than other architectures (which generally will come with a decrease in IPC, I believe I already mentioned that earlier as one of the things AMD will have to compensate for with the ‘secret sauce’ of their new architecture).
        And if anyone should know that clockspeed is no indication of performance, it should be AMD, with the Athlon vs Pentium 4 saga, performance ratings and all that.

  4. k1net1cs says:

    http://www.extremetech.com/computing/95633-amd-bulldozer-fx-pricing-revealed-a-lot-cheaper-than-sandy-bridge

    If that article has its sources accurate then I wonder why it seems to be priced to sell.
    $266 is a far cry from $320 the 8150 was originally slated for.
    But hey, if its performance is comparable to a similarly priced Intel SB parts, then I don’t really mind, hyperbolic BD expectations be damned.

    Then again, I wonder when Intel releases i7 2700K (around the same time as BD release, I believe), the rest of SB parts would also get price cuts…

  5. NEW-IMPROVED-JDWII says:

    See this is what i’m talking about Amd had so much time to make this i swear unless the next thing Amd makes is so great i’m never buying anything from them again. Except their video cards i love those don’t hate on me for that. :)

    I can’t believe i spent 200$ on a Sabertooth AM3+ board back at the end of June i wish i can go back and buy the 2500K i would probably have it at 4.4Ghz right now. Instead i have to get the 8150 that will perform like a I7 920 in gaming and a I72600K in Multithreading. At least that’s the very best i seen when it comes to leaks. SAD i’m even hoping that is real.

    • Scali says:

      I always say that upgrading CPUs is a fantasy. All this talk about socket compatibility is nonsense really. I mean, firstly, there’s plenty of evidence to show that newer CPUs don’t necessarily work in older boards with the same socket, even though they will physically fit. Newer CPUs might require lower voltages, or draw higher currents, which the board won’t support (you might actually run the risk that your motherboard doesn’t work properly with a Bulldozer CPU, and new revisions of the motherboard will be released… wouldn’t be the first time. I once had an Athlon board from Abit, the KT7A-RAID, which was a very popular board… however, officially it only supported Athlons up to 1200 MHz. I bought a Thunderbird 1400, and it just didn’t work properly, it would have problems booting, and it would crash every now and then. Had to buy a different board, same chipset, socket and everything… but one that did work properly with the newer CPU).

      Secondly, even if it does work, do you really want to run a new CPU on an old board? You might be stuck with older and slower memory, harddisk controllers, USB ports, PCI-e slots etc, so the CPU won’t reach its full performance potential anyway.

      And your case describes the third problem: if you want to upgrade the CPU on your board, you’re stuck with whatever CPUs the board will support. Are these even the CPUs you want? Just because you don’t have to buy a new board doesn’t mean you’ll get the best price/performance.
      There are plenty examples of cases where buying a new board and CPU was a better deal than sticking with your old board and upgrading the CPU.
      I can name two that I’ve encountered recently:
      – When I bought my Core2 Duo E6600, it cost around $316 (introduction, http://www.anandtech.com/show/2045/2). Its performance was considerably better than the Athlon64’s of the same price. Sure, if you had an Am2 motherboard, you could upgrade your CPU… however, it was cheaper to get an E6600 AND a motherboard, and you’d still get better performance.
      – When the first Nehalems came out, the Core i7 920 was priced very aggressively. It was faster than the fastest Core2 Duo, and also cheaper. So if you had a Core2-system, you could upgrade to one of the fastest Core2 Quads, but for about the same price, you could get a Core i7 920 + motherboard, which had better price/performance. I suppose today it’s more or less the same story with Sandy Bridge: the CPUs are priced very aggressively, and the motherboards are cheap as well… so they have very good price/performance.

      So in general I would always advise to just get a new CPU and motherboard together. It’s usually the best deal, both in performance, and also in features.

  6. NEW-IMPROVED-JDWII says:

    I’ll have to say Intel has great Price/Performance and they also have Great Overall performance, The main reason i did not get a 2500K back in June(when i had the money and before i got my board which i do like) is because i saw it losing to A AMD Phenom II x 6 in encoding and that is most important to me. But for Pure Gamers and most other things the 2500K Great for the money and i’m surprised Intel would Price it that low. I probably had the money to get a 150$ board and a 2600K and i would be set. But i kinda think Hyper-Threading is a rip off i mean it only takes 5% more die space it’s not worth 100$ more. At least not to me.

    • Scali says:

      Hyper-Threading doesn’t take any extra die space at all. 2500 and 2600 use the same die, HT is just not enabled on 2500. But the transistors are there (it’s been that way with all previous HT processors as well, going back as far as the Pentium 4).

      But you should never make the mistake of comparing die size and price. The die size generally has little to do with the overall price, especially with CPUs (GPUs tend to have tighter margins… easy to see: GPUs tend to be quite a bit larger than CPUs, yet entire videocards are cheaper than high-end CPUs).
      Performance is all that matters, when it comes to price. The better a CPU performs, the more you can charge for it. HT improves performance, so it adds extra value to the CPU.

      The true cost of a CPU is not in manufacturing, but in the development that went into the microarchitecture (and in a sense also the investment in the manufacturing facilities, going to smaller process nodes every few years).
      These costs are spread over all CPUs in an exponential manner. The fastest/most feature-rich CPU bins will get the largest profit margin, meaning that they get the biggest return-on-investment per unit sold. Then again, the cheaper CPUs will be sold in larger volumes, so even with a smaller profit margin, the total return-on-investment is significant.

  7. NEW-IMPROVED-JDWII says:

    “But you should never make the mistake of comparing die size and price.”

    I understand Performance determines price but it just seems kinda low for Intel to do that and now since you told me the 2500K has HT as well but just disabled so people by a higher-end product that just pisses me off more i wish their was a Good 3rd company that made Processors. NOT VIA. :)…..
    Also
    I Hate How Intel does not allow free unlocking(50$ more for a card to unlock their all ready made silicon is lame) with their processors and i hate how their not all overclockable. Intel only did this so people buy their higher-end products.

    • Scali says:

      That’s just how it works… All dies of a given CPU family are the same. Some just work slightly better than others, so you will find CPUs with lower clockspeeds, less cache, cores disabled etc. HT is just one extra thing that is only found on the more high-end CPUs.

      And yes, it’s a shame that CPUs have locked multipliers and such these days… then again, can you blame Intel? I bet that over 90% of their CPUs will never be overclocked anyway (most OEM systems don’t even support overclocking anyway, the options are simply not exposed in the BIOS), so the majority is not affected. It just makes sense for Intel to launch a separate line of enthusiast/overclocker products. They’re not THAT more expensive anyway (not like the Extreme Editions or Athlon FX CPUs of old… now THOSE were priced over-the-top).
      Besides, Intel is just responding to AMD with their Black Edition line.

  8. NEW-IMPROVED-JDWII says:

    “Intel is just responding to AMD with their Black Edition line.”

    The difference is even a Sempron can overclock to around 3.5Ghz and it can some times unlock to a dual-core. And their is a rumor that all the FX processors are going to be made off the same die so that means their is a possibility of a 4 core bulldozer to unlock to a 6 core or 8 core. Intel should of never locked the bus speed, Their cheaper processors used to overclock very well. Now they locked it to force customers to spend more money and what stops those customers from buying a Amd FX 4 core(when it comes out). That is why i have stuff that i hate about Intel and I also hate their Lower-end products like the Atom my E-350 is total beast for the money, it’s like the best thing Amd made all year. I kinda don’t like both companies anymore. And i have reasons why. The only company’s i do like a lot are Nvidia and Arm when it comes to tech.

    I don’t like how Intel does their business when they make so much money they still need to cheat their customers and i HATE how Amd did this bulldozer launch staying quiet to their fans. JF-AMD keeps saying to keep their current products flowing but then why did Intel allow for a sneak peak at their Sandy-E months before it comes out. I don’t get what Amd is thinking, Why not show bulldozer’s performance if it sucks on launch most people will just buy Intel and if it’s a Winner most people will wait.

    • Scali says:

      Well, sadly that’s just how it is. The overclockers market is small enough for Intel not to worry about them going to AMD.
      I try not to like companies in general. In the end they’re all just trying to survive and be as successful as they can be. Sometimes the consumers benefit, sometimes they don’t.

      I’d like to hear Fruehe explain the logic on that one yes… Unless I’m missing something here, the strategy of keeping quiet will only work if you are currently the one with the fastest products. That is not the case here: Intel has plenty of CPUs ‘out in the open’ that outperform AMD’s current lineup, which has a much bigger effect on AMD’s sales than a preview of their upcoming architecture would have, I gather.
      In fact, one could argue that it would be in AMD’s best interest not to keep quiet, especially if the upcoming product is good, because that could stop their customers from going to Intel, and instead wait for Bulldozer.

      Then again, it seems that Fruehe has a whole lot more to explain in a few days anyway…

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s