Poll: Gauging Bulldozer’s performance prospects

If you believe the latest rumors, AMD’s Bulldozer-based Zambezi processors will debut in just a few weeks. It’s about time. More than a year has passed since AMD unveiled its next-generation CPU architecture, and we’re desperately curious to see how the first commercial implementations stack up against Intel’s Sandy Bridge CPUs.

While we wait for AMD to lift the curtain on its new FX-series CPUs, let’s engage in a little speculation about how the flagship model will perform. Will the fastest Bulldozer offering be able to keep up with the Core i7-2600K, or will it be more competitive with slower CPUs like the i5-2500K or Phenom II X6 1100T? You can cast your vote in the poll below or in the middle column on the front page.

In our last poll, we inquired about the state of your PC when it’s lying dormant. Almost half of you (49%) shut down your primary system when it’s not in use. 22% prefer to put their PCs to sleep, while nearly the same percentage is content to let their system idle without turning it off. The remaining 7% favor hibernation, which may further fall out of favor as solid-state drives become more popular.

Comments closed
    • heinsj24
    • 8 years ago

    I voted for Core i5-2400, mainly because I got tired of waiting for Bulldozer and am buying an i3-2100 and the i3 was not an option in the poll.

    • daddy0623
    • 8 years ago

    Speculate all you want. It really doesn’t matter. All AMD has to do is hit the market segment with the proper performance with price and still have margin. No one on this form must work to much in the real world. No one ever talks about sales margin. In the real world that is about all that matters.

    I suspect thought AMD is keeping a tight lip because my guess it is close to the 2600. If that is the case they don’t want intel to steel their thunder to quickly which they will. Check out Lenzfire.com. Apparently they have bench marks that support what I say regarding the 2600k.

    • Arclight
    • 8 years ago

    So there are 188 votes at the moment for Via QuadCore, which tells me there are 188 Intel fanboys on the forum.

      • daddy0623
      • 8 years ago

      Really via.. wow flaming Intel fan boys… It is easy to be an intel fan boy because you don’t have to think about it.. All you have to know is i5 and i7.. anything more then that is asking to much…

      • axeman
      • 8 years ago

      I voted Via because I thought it was a type of cheese.

    • tikrjee
    • 8 years ago

    It’s interesting seeing how many people believe that the new 8 “core” Zambezi will only be able to outperform Intel’s quad-cored i5. AMD even brags about being able to best the 980X, though let’s be honest. Any Sandy Bridge i7s beat the 980X. This begs the question “what will offer the best performance per dollar?”
    Recent rumors and pre-order postings show the top-tier FX processers could be as low as $245. If that’s true, it could prove quite the contender. But my big question remains unanswered: how will the hexa-core Zambezis compare to the previous generation, and how much will they cost? For that matter, I can’t wait to see how a quad-core Zambezi stacks up against the i3 or even the i5.

    • chuckula
    • 8 years ago

    Ryhadar found an interesting set of slides over on Donamin HAber: [url<]http://www.donanimhaber.com/islemci/galerileri/AMD-Bulldozer-FX-resmi-test-sonuclari_1.htm[/url<] If true, this confirms the October 12 launch date, and puts the performance of the FX 8150 somewhere between the 2500 and 2600 in heavily multi-threaded applications.

      • NeelyCam
      • 8 years ago

      You can’t confirm a rumor by using its source.

        • chuckula
        • 8 years ago

        Fair enough… if that was the source….. many Bothans died to get these slides into Turkey for Donamin Haber to post for us! I think the secret plans for the Death Star — uh, I mean Bulldozer — are stashed in one of their hacked TouchPads.

    • ermo
    • 8 years ago

    [url<]http://www.zdnet.co.uk/news/processors/2011/09/07/amds-bulldozer-rumbles-off-to-server-makers-40093865/[/url<] Am I the only one who finds the "BEWARE OF LOW HEADROOM" warning in big red letters in the background strangely appropriate, given the delays and reports of 32nm process clock speed scaling issues?

    • Hattig
    • 8 years ago

    Whilst I believe that BD will actually finally perform very well – around 2600K at least – I expect it to suck in legacy x87 tasks (e.g., SuperPi) due to the new FPU arrangement and optimisation for SSEx, AVX, etc.

    BD will show its best side in multithreaded tasks, where the eight threads will get a supposed 80% scaling over four threads, which is far more than HyperThreading gets on Intel processors.

    In addition it does look like the turbo and clocks are quite high. Even if the IPC isn’t as high as SB, the clocks could make up for it. Single-threaded performance is a tough one to call however, where SB turbo is pitted against BD turbo. So far the ES benching suggests BD will be comparatively weak. Of course this is less and less of an issue going forward, as apps become multi-threaded and systems heavily multitask at all times.

    SB with QuickSync will presumably thrash BD at video encoding.

    • luisnhamue
    • 8 years ago

    Im not expecting AMD to wipe the floor, but I believe they will be able to compete against 2600K, because hsitory has shown that they need more cores to match intel cpu’s. If the same thing repeat this time, their 8core cpu will match 4core (8 thread) intel cpu. Im not seeing 2500k being able to keep up, under heavily multi threaded load. And AMD has been doing a great job on this department, the big problem they have to address is single threaded loadings, where we may well see the 2500k outshine Bulldozer.

    • ronch
    • 8 years ago

    Does this poll indicate we could apply The Secret to Bulldozer?

    • MadManOriginal
    • 8 years ago

    Unfortunately AMD is pulling some major obfuscation with their ‘core’ definition, especially to less informed who are most likely to think that ‘moar coars = bettar.’

    If Intel sold 4c/8t HT CPUs as ‘8 core CPUs’ the uproar would be deafening. When AMD does it you can hear a pin drop. (Yes, I know an 8-count of BD cores is not truly equivalent to an 8t-count of Intel cores, but the comparison still works.) The computer nerds truly have double standards when it comes to the two companies.

      • Deanjo
      • 8 years ago

      People really get to uptight about “cores” on cpus and how a manufacturer defines them. Remember intel slapping two dual cores together to make a quad? Shouldn’t that technically be then dual dual cores and not a quad core? Or how about shaders nv vs amd vs intel? Bottom line is that it is a term that never had a set defined industry standard definition of implementation and as such, nobody can really say xyz is obscuring of what a “core” is.

      • NeelyCam
      • 8 years ago

      [quote<]The computer nerds truly have double standards when it comes to the two companies.[/quote<] No - just the AMD fanboi army. You don't hear Intel proponents claim that SB CPUs are somehow octa-cores. Although, Intel Marketing probably should say that. I mean, why not? When you look at Windows Task Manager, it really looks like 8 cores.. muggles wouldn't know the difference.

        • cegras
        • 8 years ago

        trolololo, what about AMD and nvidia’s definition of GPU cores?

      • Anonymous Coward
      • 8 years ago

      [quote<]If Intel sold 4c/8t HT CPUs as '8 core CPUs' the uproar would be deafening. When AMD does it you can hear a pin drop.[/quote<] Sadly, I missed the pin dropping because certain people just won't shut up about how AMD counts cores.

        • NeelyCam
        • 8 years ago

        I’m just trying to help you make an informed decision, instead of being bulldozed by AMD’s dishonest marketing.

        I’m doing it for [i<]your[/i<] benefit.

      • Hattig
      • 8 years ago

      There is a massive difference between supposed thread scaling of 80% (Bulldozer’s second thread on a module / second core) and up-to-30% (Intel’s HyperThreading). Traditionally cores are counted as Integer Cores anyway (Sun’s Niagara chip was 8 cores despite having 1 shared FPU between all eight cores) and BD has discrete integer core logic for that function.

      Yes, BD blurs the line a bit with the module concept, but it isn’t just about filling in gaps in the execution pipelines of a single core, each thread gets its own integer pipelines to itself.

      We’ll see how well it works in practice when we read the reviews on various websites.

      • forumics
      • 8 years ago

      BD was never actually supposed to be 8 core, the intention was supposed to be 4 core with extra hardware to go against intel’s hyper threading. however marketing took advantage of the extra core to label it as 8 core
      nevertheless an 8 core BD was never supposed to go against an intel 8 core product, it was meant to compete against an intel 4 core with HT

        • ronch
        • 8 years ago

        Calling a 4-module chip an 8-core chip is a double-edged sword. You fool consumers who think they’re getting twice as many cores for the same money, but you also set yourself up to be compared to real 8-core products. Lucky Intel doesn’t have that yet for desktops, but reviews of 8-core BD Opterons will inevitably talk about how an 8-core Xeon compares to an 8-core (4-module) BD. And also, desktop BD reviews will inevitably compare an 8-core BD to a 4-core SB and see how much AMD goodness you really got for your money. AMD needs better marketing people.

        And while we’re on the topic of marketing, I seriously don’t think AMD’s John Fruehe is the right guy for the job. I believe a marketing guy needs to be a real PR person, but John looks like a constipated stiff who resorts to trash talk, like saying AMD’s 65nm outguns Intel’s 45nm. Yeah, whatever, John.

      • ronch
      • 8 years ago

      [quote<]obfuscation[/quote<] Man, you made me use my dictionary.

      • sschaem
      • 8 years ago

      No. HT offer way to increase IPC, sometime, by reducing pipeline stalls, but Bulldozer module duplicate execution units so you do get double the raw performance with 2 thread vs 1.

      The issue is that a BD core is kind of lightweight so you need more threads running to leverage all the units available.

      AMD could have added HT to BD core to squeeze out pipeline stalls (underused units). And nobody would expect a HT style Bulldozer to be called a 16 core processor.

      Case closed.

      • xeridea
      • 8 years ago

      It is in every way 8 “cores”. There are some shared resources, but there are 8 integer cores, and for FP it can function as 8 128-bit FPUs. The lost performance due to some shared components is ~7-8% so thats ~7.4 cores of performance.

      HT has ~40% performance hit, so 8 threads is ~4.8 cores of performance. Also, HT doesn’t add any cores of any kind, just allows existing cores to process more, at a huge hit to per thread performance (but more overall). They can’t say “8 core” because there are only 4 cores. Even if Intel has more IPC, that is a different matter. The classification of “cores” from AMD is valid.

        • forumics
        • 8 years ago

        well if you can’t say that a 4core + HT is 8 core, then you can’t call a BD 8 core either because every 2 of its INT cores is connected to 1 FP core, thus its a crippled core and it won’t be fair still to pit it against a single intel core!

        you also can’t say that its FP core can function as 8 128bit FPU, i can just as well say its intel’s fault for implementing 128bit FPU when they should follow BD and implement a 256bit.
        alternatively, when intel improves its FPU to 256bit, how is BD going to compare?

        bottomline is that the engineers at AMD never intended a BD to go against an 8 core intel. the design is 4 modules vs intel 4 core. the marketing might have changed the sales pitch but it doesn’t change the battle; a 4 module BD will be competitive against a 4 core intel but never against anythign more than 4 cores

          • Waco
          • 8 years ago

          I don’t understand why people are so up in arms about this.

          A BD module is 2 cores, period, whether you like the performance of that or not. There is no standing rule that resources cannot be shared between cores, so don’t pretend there is.

          Just because the FPU is shared in some cases does NOT make it a single core. That makes no sense at all.

            • NeelyCam
            • 8 years ago

            Words like “module” and “core” are meaningless. They are only useful for marketing or trolling.

            What matters is performance, period. I don’t care if it came from 8 cores with some IPC, or four modules with twice the IPC, or what.

            I think Intel is being a bit more honest here, marketing their CPUs are 4-core CPUs where each core is actually a “core+”… AMD is marketing their parts as 8-core CPUs where each core is a “core-“. I’m not worried about the enthusiasts who can see through these marketing tricks and tend to understand benchmarks and make informed decisions regardless. I’m worried about muggles who may think that an AMD 6-core is better than Intel 4-core.

        • Jason181
        • 8 years ago

        Actually, I think you mean it has an 80% performance hit, because there are 4 “hyperthreads” and 80% of 4 = 3.2 and 8 cores minus the 3.2 core “performance hit” would equal 4.8 cores of performance. So you came out with the right number, but you’re basically saying that the hyperthreaded “cores” are only worth 20% of an actual core.

        I mostly agree, except that I’d say they’re worth more like 30% of an actual core in most situations for 5.2 “actual cores” but there are some cases (media encoding specifically, but there are others) where the scaling is 80-90%. If you are buying for one of those edge cases, the extra performance makes the 2600k a monster.

        I believe for gaming, the BD will have a hard time keeping up with even an i5-2400, but it will do considerably better (probably 2500k-2600k range) in most workloads, and there will be some cases where it beats the 2600k by 30-40% (I’d expect those to be mostly server-based loads). As mentioned before though, this is all pretty uninformed speculation partially based on the asking price, information about the execution units and cache, and AMD’s past history on new releases.

        Full disclosure: I have a 2600k, have owned probably 7 other Intel processors, and have only had one AMD processor (Athlon XP 1800+). That’s mostly because I have more money than time, typically; when I was poorer I favored cpus I could overclock the crap out of, which typically were Intel.

      • evilpaul
      • 8 years ago

      Bulldozer “modules” have a shared front end, two Integer units, and a shared double-sized (for AVX support) FPU. So maybe AMD’s AVX optimized app performance will suffer, and people seem to believe the single threaded Integer performance won’t match Intel’s, but for the most part I don’t think it will matter in the real world.

      And it’s not like AMD and Nvidia have anything similar performance-wise in what they’re calling cores/processors.

    • LaChupacabra
    • 8 years ago

    I’m agreeing with the sentiment it’ll perform between a 2500k and 2600k. What will really be interesting is how these will shape up as a server chip, and how well they play with virtualization. If a single socket could give 1 “core” to 8 VM’s (or 16 if AMD does a dual chip package) that’s a pretty potent value proposition.

    • anotherengineer
    • 8 years ago

    I just got a gigabyte 990x and 955BE combo for 240 and free shipping and 2 games. Sold the games for 20 each.

    Unless the dozer can double the performance of the 955BE which I am happy with I see no reason to upgrade. (since it’s basically a $110 chip)

      • Waco
      • 8 years ago

      Which Gigabyte 990X?

    • capricorn
    • 8 years ago

    I don’t understand the logic for those who voted for AMD FX-8150 ($266) only competitive with a $216 price tag Core i5 2500K.
    Since when AMD dare to mark their CPU price 23% higher when it is only competitive with the opponent which has launched since almost a year ago?

      • Corrado
      • 8 years ago

      Agree. AMD has NEVER overpriced CPUs unless they knew they could. IE: There was no competition, or they were faster than the competition.

      • NeelyCam
      • 8 years ago

      Recent rumors indicate FX-8150 will be $245:

      [url<]https://techreport.com/discussions.x/21685[/url<] And what should those folks pick? i7-2600K? I for one won't believe that FX-8150 will be competitive against 2600K.

        • capricorn
        • 8 years ago

        Well, I personally thought that i7 2600K doesn’t match its $317 price tag when compare to i5 2500K.
        It doesn’t stop Intel to price it at over 46% higher (due to lack of competition from AMD).
        [url<]http://www.anandtech.com/bench/Product/288?vs=287[/url<] AMD in the other hand shall have expected the refresh of both i5 & i7 line-up from the Intel camp whenever Zambezi launch, they might have no choice but to set a rather competitive price in order to stay better cost performance than their opponent.

      • Lazier_Said
      • 8 years ago

      AMD’s $205 X6-1100 is not merely beaten but omfgpwned by the $184 i5-2400 at everything outside of a few niche 6 threaded HPC demonstrations.

      An 8 cored product will see a higher markup.

        • capricorn
        • 8 years ago

        So you mean the FX-6100 will pwn the $184 i5-2400?
        Or you are saying a $175 3m/6c FX-6100 is even cripple then a 6c X6 1100T?
        Allright, it’s the IPC all over again.
        If Zambezi IPC is so bad and they are forced to sell the 4m/8c at 4c price tag, than I’ll say shame on you AMD.
        Why don’t you just shrink Thuban to 32nm and add another 2 cores which can save time and resources, and probably can achieve smaller die size but in the other hand stronger than Zambezi too.

          • NeelyCam
          • 8 years ago

          [quote<]Why don't you just shrink Thuban to 32nm and add another 2 cores which can save time and resources, and probably can achieve smaller die size but in the other hand stronger than Zambezi too.[/quote<] This is what I've been asking myself.. or just keep it a 6-core, but run it faster because 32nm is faster than 45nm.

          • forumics
          • 8 years ago

          once again, BD wasn’t supposed to be 8c, the figure 8 most probably came from marketing.
          it was supposed to be 4c with extra hardware to compete against intel’s HT
          therefore going by the original strategy, a 3c FX6100 should be priced cheaper than an intel’s 4c without HT

          • Lazier_Said
          • 8 years ago

          No, I mean AMD already prices their 6 core products (and their high end 4 core products, the i5-2320 omfgpwns the X4 980 too) significantly higher than a SB product of equal performance. So concluding that BD’s $266 pricetag means i7-2600 performance is bull.

      • ronch
      • 8 years ago

      Even if BD pricing is official, they’re just list prices. AMD has a habit of pricing lower than list prices. Could end up priced not too differently from 2500K.

    • no51
    • 8 years ago

    Lots of wizards who can see the future commenting in this article. Would any of you mind giving me this week’s upcoming winning lottery numbers? Next week’s would be fine too.

      • faramir
      • 8 years ago

      3, 9, 12, 12, 28 and 33. Yes, 12 will be drawn two times so be sure to tick it correctly.

      • Chrispy_
      • 8 years ago

      Indeed.

      People seem to be taking rumour and info from a website proven to fake their results as hard fact. To date, the ONLY price/performance hint we have had is that an FX-8150 is worth “around $300” as a result of the competition prize info.

      If anyone, has any hard facts, then they’re doing better than all the hardware journalists in the world combined.

        • ronch
        • 8 years ago

        Yeah. Looks like everyone already knows how BD will turn out. No need for TR to write a review then.

    • ermo
    • 8 years ago

    There’s a bit of misinformation in the comments, I think.

    Each BD module has 1 front end.

    Each front end supports two quad-issue (NOT dual issue — that’s Bobcat!) ‘cores’.

    So if you have 4 modules, you can dispatch/retire 4*(2*4) = 32 instructions per clock in the ideal case where you have 8 threads keeping all the 8*4 pipelines busy with fast (1 cycle) instructions.

    4*(2*4) = 32 is almost *twice* the 6*3 = 18 instructions per clock of a hexa-core Phenom II in the ideal case. My guess is that 1 BD module will equal [s<]2.25-2.5[/s<] 1.6 - 2.0 Phenom II cores in terms of IPC/throughput. [i<]2.5 is probably overly optimistic as sschaem points out.[/i<] What AMD has done is quite clever, really. To compete with intel on IPC, AMD had to add an extra pipeline per 'core' (up from 3 to 4), but sharing resources at the module level is IMHO a very elegant solution -- especially for highly threaded workloads. One of the reasons the CPU has so much cache is probably because it is beneficial when it comes to keeping all execution units as busy as possible, as it is probably not trivial for one front-end to feed 8 execution units divided between two threads. Worst case, a 4 module BD will perform as a 2500K or possibly an i5-2400. Best case (highly threaded), it will come out on top compared to a similarly priced 2600K due to having more execution units available. For purely single threaded stuff, AMD will probably be anywhere from 10-20% slower than SB clock for clock and will have to rely on clock speed to gain parity -- the good news is that BD was designed with a deep pipeline to make this at least theoretically possible. My guess it that it'll be priced between a 2500K and a 2600K (after intel introduces the 2700K) and be a smash hit at that price point once AMD sorts out its process issues. Oh, and bargain hunters will be happy with the 3*2*4 = 24 issue triple-module BDs, which I'm guessing will compete favourably with the 2500K at a slightly lower price point by virtue of being able to clock the 3 working modules higher while staying within the same TDP as the 4-module parts.

      • sschaem
      • 8 years ago

      Tell the whole story and mention that those 4 units are not general units, only two can execute flow control and integer.
      The math is (4 module * 2 core * 2 general unit) for BD vs (6 core * 3 unit) for K10

      [url<]http://www.realworldtech.com/page.cfm?ArticleID=RWT082610181333&p=7[/url<] So for any workload able to leverage 1 to 6 thread we have a 50% drop in raw execution units. Bulldozer only win when their is more than 6 thread active *AND* the workload is an even mix of integer & floating point. To 'unlock' bulldozer potential you will need fine tuned software... (I expect the first to be the Radeon drivers)

        • Deanjo
        • 8 years ago

        OK if you are going to tell the WHOLE story how about mentioning that IPC will not factor into a whole hill of beans for 90+ percent of the applications out there that people run. Most PC’s out there run for most of their functional lives in their lowest pstate for crying out loud.

          • NeelyCam
          • 8 years ago

          Then why not use Atom? Most PCs are idle or off 90% of their functional lives.

            • Deanjo
            • 8 years ago

            Because Atoms struggle at times just even running the OS.

          • travbrad
          • 8 years ago

          “You don’t need the performance our new CPU gives you” Wow you should work for AMD’s marketing team. 😉

            • WaltC
            • 8 years ago

            Why would AMD’s marketing department *ever* say that?…;)

        • shank15217
        • 8 years ago

        So you are assuming that because they are not general units somehow they are worse than general units? What qualifies you to say that?

        • bwcbiz
        • 8 years ago

        Since when is 4*2*2=16 a 50% drop from 6*3 = 18? That’s the second time you’ve thrown that 50% number around in this thread and gotten it wrong. Your 8 vs. 12 comparison earlier at least could be explained by incorrect phrasing, but this is complete FUD.

        16 is a 2/18 drop from 18. 18 is a 2/16 increase from 16. Neither of those come anywhere near 50%.

        My personal opinion is that thread for thread, clock for clock AMD will be about 75% of SB performance. So a 4 module AMD unit running 8 threads would need about a 33% higher clock rate to beat a 2600K. But it would beat a 2500K in highly threaded apps due to the extra thread capacity. 4m/8T vs. 4c/4T. Under light usage, AMDs individual cores will perform worse than any individual SB core, and while a module will beat a single hyperthreaded core, the modules will make the power efficiency comparison even worse because a full AMD module will consume more power than an Intel hyperthreaded core. So overall I think Bulldozer will compete with SB on all-out multi-threaded performance up to and including the 2600K but lose on power consumption, single-threaded performance and scalability for future die shrinks. So my net rating is comparable to 2500K.

      • chuckula
      • 8 years ago

      Sschaem has a very good point.. you are painting an unrealistically optimistic picture of Bulldozer since practically no real workloads have balanced integer & floating point instructions that would bear out your performance numbers.

      AMD itself discounted the FP units in Bulldozer in favor of more integer cores! That hardly sounds like AMD expects everyone to be sending balanced integer & floating point instructions to the CPU at once.

    • link626
    • 8 years ago

    anything multithreaded, bd will keep up, just because it has more cores.

    anything single threaded, or optimized for a limited number of cores, BD will fall behind.

    • StashTheVampede
    • 8 years ago

    My completely unscientific guess:
    – Bulldozer, clock for clock, will be slower than Sandy Bridge.
    – Bulldozer fastest chip will not be as fast as the 2600k in most conditions.
    – Bulldozer, in the heaviest of threaded workloads, will perform very close to SB and will sometimes win.
    – AMD will price Bulldozer parts accordingly

    The future for Bulldozer lies in server farms, not the desktop.

    • can-a-tuna
    • 8 years ago

    Via QuadCore, you’r so funny. Phenom II,.. right. What kind of joke poll is this.

    • Aspleme
    • 8 years ago

    I believe the issue is a little more complicated than the poll suggests. We also can’t judge performance just by price since AMD has a habit of being a little cheaper for just a little bit less performance. That being said, Intel has demonstrated superior numbers in raw compute in the past, and I don’t think Bulldozer is going to change that. I don’t think AMD is expecting to change that with BD.

    Let me see if I can explain. In my family, I’m the only one who has a desktop with an Intel processor. Why? Because earlier generations of processors, AMD has been more common in the cheaper prebuilt computers. From what I remember, it was also more common to see AMD computers go on sale. Now that computing has changed, with more programs, browsers, and even windows itself using parallel processing and graphics acceleration, AMD is trying to regain the mainstream market… and maybe even offer their graphics cards an advantage with asymmetric crossfire.

    What I suspect we’ll have in BD is a CPU that is ‘good enough’ at its price point and offers better balance performance for most users. Even for gamers, there is little difference between a $250-$300 CPU and a $500-$999. Even my $200 i5-2500 only peaks out when I’m doing video encoding… and if handbrake offered quicksync and gave me a reason to switch to a Z68 board, it would be even better… at the one task that actually uses its full power. If Bulldozer offers better graphics capabilities with similar raw CPU processing of a 2500K, I can see people choosing it over Intel.

    • Krogoth
    • 8 years ago

    At mainstream tasks, about i5-2400 at best.

    At server-workstation related tasks, slightly edging out i7-2600K and 990X.

    Bulldozer will lose on the power efficiency front, probably on par with Bloomfield chips.

      • ermo
      • 8 years ago

      I’m curious to know why you think it will lose out on the power efficiency front? I was under the impression that the whole module thing with shared frontends was all about efficiency gains vs running with full featured cores with seperate front ends?

      Granted, the first shipping stepping might not be terribly good (just unconfirmed rumors so far), however the chip was designed from the outset to be power efficient and have extensive clock gating.

      I expect the Q1 BD release to give a proper idea of how it will compete in terms of power efficiency. Yeah, too bad it is Q1’12 instead of ultimo Q2’11 as originally planned …

    • ALiLPinkMonster
    • 8 years ago

    This is way to complex of an issue for a simple poll. I think that per core, BD will outperform PII but fail in comparison to SB. However, those eight cores will obviously give BD an advantage in multitasking and heavily multithreaded applications. The redeeming factor for gamers could be overclocking abilities that compare to those of SB. If the quad core BD units can be overclocked straight to valhalla, they would certainly make a great alternative to the 2500K while being a little cheaper.

    If course, I could be completely wrong. We’ll see.

    • cegras
    • 8 years ago

    At the end of the day, you have two processors, one from AMD and another from intel, at the same price point. Pick the one that performs better for the money. Deconstructing their performance into IPC is incredibly silly. There is nothing wrong with an architecture that makes up for lower IPC by increasing clock speed, if that is what gets the job done.

      • bcronce
      • 8 years ago

      Removed my own comment

        • NeelyCam
        • 8 years ago

        Funny – when I post something like this, I get -4

          • Firestarter
          • 8 years ago

          gave you -1 because you weren’t at -4 yet 😛

      • sschaem
      • 8 years ago

      Untill we have official result we can use the expect IPC to make sense of leaked benchmarks.
      But yes, the most telling is the price. AMD FX-8120 is to replace the 1100T

      And I agree it will have to be a personal choice, because its clear that the PhenomII 1100T might be faster than the FX-8120 in more then a few type of applications. And for the same price people will choose the better chip, and that can still be the 1100T.

        • cegras
        • 8 years ago

        The 1100T can be faster? What do you base this conjecture on? JFAMD has been adamant that all information released is fake.

          • forumics
          • 8 years ago

          even if its real, the performance might be running on pre-finalized bios which would not take full advantage of the new BD architecture.

    • kamikaziechameleon
    • 8 years ago

    I’m amazed you removed core 2 processors.

      • NeelyCam
      • 8 years ago

      Tegra 2 should’ve been there.

      • Theolendras
      • 8 years ago

      Core 2 is in the same playing field as Phenom II. Nehalem would have been great addition tough.

    • Arclight
    • 8 years ago

    I’m guessing the top one should be about on par with the 2500K judging from the rumoured price. I was initially hoping it would beat the 2600K, but if price is any indication….it probably won’t.

      • bwcbiz
      • 8 years ago

      Either that or they are counting on a markdown of the 2600K when the 2700K comes out.

      • Hattig
      • 8 years ago

      I expect the pricing has also been set against the expected Intel price drop of the 2600K due to the release of a 2700K or 2800K in the near future.

    • sircharles32
    • 8 years ago

    As long as it is faster than AMD’s previous generation (Phenom II X6 1100T), then I will buy it.
    This will be a rebuild for my current single core AMD Opteron 150 (socket 940) workstation (6+ years and counting).

    If it is not faster than AMD’s previous generation, then I will most likely save money, and just buy from the previous generation.

      • Corrado
      • 8 years ago

      This is how I’m viewing it. If its no good, I’ll buy a Thuban to put in my existing server/PVR.

    • ronch
    • 8 years ago

    Well, this is a bit pitiful. I don’t recall there ever being a CPU which triggered as much speculation, anticipation and guessing because it’s been so long overdue. Bulldozer must hold the Guinness world record for most anticipated CPU architecture ever.

      • Theolendras
      • 8 years ago

      If I recall correctly, first K7, K8, K10 generated almost the same level of anticipation.

      • Geistbar
      • 8 years ago

      You must not have been around during the time preceding the original Athlon64! There was an enormous about of speculation around it, and it had more than its fair share of delays. That one ended up being a lucky* break for AMD, maybe Bulldozer will be too. Maybe not.

      * Lucky not being quite the best word, since a great many skilled and talented people put years into designing it.

        • ronch
        • 8 years ago

        K8 did trigger a lot of excitement alright. What I said was, BD must be the most anticipated CPU ever. Prior to K8’s launch I don’t recall everyone having the sort of anxiety etc. that we’re having towards BD nowadays.

        • NeelyCam
        • 8 years ago

        Intel was complacent back then. I would say Intel is no longer complacent.

          • forumics
          • 8 years ago

          intel wern’t complacent, they just followed the wrong strategy
          during those days everything was about clock frequency not IPC or power efficiency and intel were always the 1st to break clockspeed records; thus they were not being complacent
          all their talents were going into increasing clockspeeds

            • NeelyCam
            • 8 years ago

            They were complacent; as a result, they followed a wrong strategy blindly without innovating or looking at what the competitor was doing.

            • Beelzebubba9
            • 8 years ago

            The Pentium 4 was incredibly innovative; it just wasn’t all that good.

            • ronch
            • 8 years ago

            [quote<]they just followed the wrong strategy[/quote<] And this is exactly what AMD is dong with BD. Wrong strategy. They just shot themselves in the foot.

            • kamikaziechameleon
            • 8 years ago

            A symetrical competition is the only way they can compete with intel in the short term. I’m still curious how they plan to resolve allot of things in the long term but BD makes sense.

    • Bensam123
    • 8 years ago

    I’m just going to go all the way with this one. I’m a optimist when it comes to big names and innovation, so I’m going to hope that it’s better then Intel’s best.

    • Duck
    • 8 years ago

    It’s going to be competitive with the 2600K for sure in heavily threaded apps.

    General purpose stuff like browsing web and gaming, I expect the 2600k will edge out a small but consistent/comfortable lead.

    So Bulldozer will be competitive with the 2600K. Just how much will depend on what you want to do with it.

      • travbrad
      • 8 years ago

      I pretty much agree with this, but I think the single-core performance of the 2600K will have more than a small edge, at least in synthetic tests. It’ll be a bit closer in gaming though, because at realistic resolutions (like 1080p) the GPU often becomes a partially limiting factor.

        • Duck
        • 8 years ago

        For general purpose stuff, I doubt anyone could feel the difference if the 2600K had a small advantage. So lightly threaded workloads might not be so important as bulldozer will be ‘fast enough’.

          • NeelyCam
          • 8 years ago

          The “fast enough” argument is weak. One could argue Atom is “fast enough”… you’ll just have to wait a few milliseconds longer than with Llano/BD/SB/IB/etc.

            • Deanjo
            • 8 years ago

            Hardly, Atom struggles at times just running the OS, let alone something that adds a bit more load like loading a dbase, cut and pasting graphics in a Word document, etc.

            • chuckula
            • 8 years ago

            Look NeelyCam you need to get with the AMD program here:

            “Fast Enough” –> Any AMD system

            “Overkill” or “Not necessary” or “Wasteful” or “Overpriced” –> Any Intel system that’s faster than an AMD system. Note that even if the Intel system is less expensive than AMD systems that it beats, as was the case when the 2500 came out, it is still overpriced because there’s probably some budget AMD chip out there that’s cheaper

            “Underpowered” “Inadequate” “Obsolete” –> Any Intel system that’s behind AMD’s chips, even though they likely are aimed at completely different markets. Since Intel’s real goal for Atom is to put it in tablets & cellphones, AMD users can crow about how it’s not good enough even though AMD doesn’t even want to compete in this market.

            • ermo
            • 8 years ago

            I +1’ed you.

            And I’m an AMD apologist — curse me for rooting for the underdog with the elegant designs.

            Then again, I’m definitely NOT a fan of Intel’s product segmentation strategy. But that’s a whole other kettle of fish.

            • NeelyCam
            • 8 years ago

            Ah, [i<]now[/i<] I understand. Thanks for calibrating the meaning of those words with respect to various processors. I much appreciated it. Just to clarify, when new AMD chips are relased with performance (possibly) going up, will the "fast enough", "overkill", "underpowered" etc. performance specifications be re-calibrated so that the definitions above (e.g., "Fast enough" --> Any AMD system) still hold true? I'd like to prepare myself for the eventual Trinity launch, so I don't embarrass myself by accidentally calling it Inadequate/Overkill when it's supposed to be called "Fast Enough"..

            • Duck
            • 8 years ago

            No. Fast enough meaning it’s close enough to not be able to feel the difference. Fast enough meaning you better be running 4x Vertex 3 max iops in RAID0 trying to make the CPU the bottleneck instead of the storage subsystem. Fast enough meaning the onus is on performance increases coming from changes at the software level (putting more cores to use) rather than the hardware level.

            Fast enough in lightly threaded apps, superior performance than the 2600K in heavily threaded workloads. (probably)

      • sschaem
      • 8 years ago

      AMD can turn this around if they done a magnificent job with the power management/usage.
      First page I will jump to is “Task energy” , because this is THE key to bulldozer long term validity.

        • ermo
        • 8 years ago

        Hear hear.

        • NeelyCam
        • 8 years ago

        Yes. It’s just unfortunate that I’ll have to go to Anandtech first, because their review will likely be up at 12:01am EST on the day the NDA expires.

      • faramir
      • 8 years ago

      I’m still inclined to believe T. Seifert’s assessment regarding Bulldozer performance, ergo: no IPC gain over Phenom II and 8 cores versus 4 cores means performance roughly on par with 4 core i5 2500K (with heavily threaded loads). Single-thread performance will likely s*ck some major bollocks.

    • chuckula
    • 8 years ago

    For massively threaded tasks BD will usually pull ahead of the non-HT 2500K and sometimes (but by no mean always) pull ahead of the hyperthreaded 2600K. At everything 4 threads and less (which is quite a lot), the 8150 will only beat a 2500K if its turbo-boost is REALLY working well.

    P.S. –> If you are OC’ing a 2500K or 2600K up to 4.4 Ghz, then you’ll probably close most of the gap in areas where Bulldozer is ahead. Even if you OC Bulldozer to 5Ghz it’ll probably not get the same performance boost.

      • sschaem
      • 8 years ago

      With a 4 thread workload, the match is between 2 execution units per core for BD and 3 unit for Sandy.
      Bulldozer start the fight with 50% less execution units (8 vs 12)… Its not going to come anywhere close to good old Sandy.
      Sadly the 8120 will have a hard time even competing with the 1100T for most workload.
      The only real edge BD got is with massively threaded SSE4/AVX optimized code that uses at least 8 thread.

        • MrBojangles
        • 8 years ago

        “Bulldozer start the fight with 50% less execution units (8 vs 12)”

        So 8 is half of 12??? I’m no mathematician or anything but i calls bs, my friend.

          • Corrado
          • 8 years ago

          50% of 8 is 4. 8 is 4 less than 12. Therefore, 12 is 50% more than 8. But 8 is not 50% less of 12. So… you’re kinda both right.

    • Ushio01
    • 8 years ago

    In a clock for clock and thread for thread comparison I think it will be between a 45nm core2 and a 45nm Nehalem at best.

      • sschaem
      • 8 years ago

      Clock for clock BD is slower then K10, so no way its going to come anywhere close to Nehalem.
      The only hope is with AVX/SSE4 code that K10 cant execute and need to fall back to slow SSE2 code.

      Its amazing llano released this year still only offer SSE2 … AMD = Antiquated & Mediocre Design

        • khands
        • 8 years ago

        I don’t believe amd would ever release a slower ipc cpu as the mainstream offering. If the reviews show that to be true it’s time to split up intel because that’s the only way we’ll have competition in this marketplace again.

          • NeelyCam
          • 8 years ago

          So, Intel needs to be split because AMD keeps f*cking up? That’ll go well with the courts…

          • Lazier_Said
          • 8 years ago

          If only Intel’s products were later, slower, hotter, and more expensive! That benefits the consumer how, exactly?

          • ronch
          • 8 years ago

          Judging by the info released so far (and this is speculative!… I’m not DonanimHaber!!), BD may well match SB’s [u<]IPC[/u<] on multi-threaded apps, but fall far behind in single-threaded ones. The BD project was started 6 years ago, and 2005 was the time when multi-core was really starting to get all hyped up. If you were AMD, what would you decide your next architecture to be? But the thing was, Intel poured more brains into making higher IPC processors with Core 2 moving forward, and AMD just couldn't scrap BD while it's already going full blast. They instead probably hoped that BD will clock high and make up for that lack of IPC, betting on improved manufacturing processes by the time BD hits the road, as well as architectural features meant to let BD achieve high clocks. Problem is, it looks like AMD isn't hitting the right clocks today. Bad timing, because BD is delayed and Intel is two generations ahead of AMD's current lineup. This is worse than the Barcelona story when Intel was just one generation ahead. Being two gens behind, AMD has to make BD close a bigger gap, and if it doesn't, it'll be in a worse position than it was when Barcy came out.

    • Vasilyfav
    • 8 years ago

    I know you guys like to include troll vote options for s**** and giggles, but half of them?

    Even AMD’s lowest performance estimates is that the top model will have at least 30% more throughput than the previous top model.
    So it’s obviously more competitive than 1100T and by extension, X4 980 and the ephemeral Via Quad.

    My bet – between 2400 and 2500k. Still going to be behind in single core performance compared to sandy.

      • sschaem
      • 8 years ago

      Clock for Clock BD is the slowest of them all (beside the Via)
      The 1100T got 6*3 execution units, BD got only 8*2. So its 18 vs 16 … BD need to make up the difference with clock speed.
      The main reason why app like Cinebench do much better is mainly because the 1100T cant run Cinebench optimized code and fall back to the SSE2 code path.
      BD only win when 8 thread are used, and those 8 thread run at least some SSE4/AVX code.
      In single threaded mode BD will need to hit ~4.2ghz to match a 3.2ghz K10

        • Vasilyfav
        • 8 years ago

        Are you saying a Bulldozer module is 33% slower than a Thuban core if the clocks are equal?

          • sweatshopking
          • 8 years ago

          that’s what the man is saying. i’m not sure the accuracy, or how the missing execution unit per core will impact performance. the discussion has changed, based on what the initial hope was. they were hoping people would look at the 8 core as a quad core, and say “wow, this one has 4 execution units per module, vs 3 on stars! that’s great!” but they didn’t. for some reason, the modules was found to be too confusing, and so they had to change the marketing to “8 core” from 4 modules, and now bulldozer doesn’t look so great. this was to be the quad core replacement, and scale up to an 8 module later, which would of course have 16 cores, but it sadly, wasn’t to be. if you compare modules to cores, BD comes ahead, core to core, it doesn’t look as good.

            • Vasilyfav
            • 8 years ago

            That’s disappointing. I think I’ll just go with a 6 core SB-E then.

            I need something to be able to encode 1080p at 25 fps (or 720p at 30-35 fps) in the background on 2 cores, stream, play Starcraft 2 at 60+ fps and still have 1-2 free cores left for multitasking and running flash and java in 2 browsers on different monitors.

            My Phenom II cannot handle the load atm, and I’m not even sure the top bulldozer model will be able to do this flawlessly without hiccups.

            • Vulk
            • 8 years ago

            I love the trolling.

            • sweatshopking
            • 8 years ago

            ermo posted disputing his point on the number of execution units. it’s worth taking a look at. I’m not sure, again, which post is correct. I am not aware of the number of execution units, but my point on the marketing is correct.

            • ermo
            • 8 years ago

            From my perspective, sschaem paints a slightly more pessimistic picture of the IPC potential of the BD approach than what is warranted.

            Yes, each BD ‘core’ has 4 pipelines divided into 2 groups of 1 AGU + 1 ALU each, so each module can dispatch 4 integer instructions per clock PLUS any additional load/stores and SIMD instructions, whereas Deneb/Thuban can dispatch at most 3 instructions per clock per core period.

            If AMDs performance scaling math works out as they predict, 1 BD module will equal at least 1.6 Deneb/Thuban core. If AMD has made efficiency gains of, say, 25% per frontend/module by tweaking other parts of the architecture, then 1 BD module might equal 2 Deneb/Cores, clock for clock.

            In any case, I (unlike sschaem) have a hard time believing that AMD would willingly make its IPC per module/frontend lower than per Deneb/Thuban core in the worst case scenario.

            But I am of course open to revising my opinion once reproducible benchmarks establish the naked facts, as I am looking to buy an 8 thread 32nm CPU come Q1’12 whether it’s branded AMD or Intel.

            • Antimatter
            • 8 years ago

            John Fruehe has said that the IPC of Bulldozer will higher than K10. The number of execution units are not the sole factor governing the performance of a CPU.

        • Vulk
        • 8 years ago

        I’m not an expert on this, and clearly you aren’t either.

        When I asked an expert this is the response I got:

        I assume you’re talking about integer performance? Please keep in mind that in microcode execution execution units aren’t all equal. According to AMD’s own performance testing it was used very infrequently, and a good scheduler would be able to alleviate the problem entirely. That’s why they ditched it and instead beefed up the prefetch logic.

        The issue I see with Bulldozer is just the sheer length of the thread execution pipe. It will make it easier to overclock, but slower per individual clock if there is a lot of context switching even though it’s an out of order design.

        That said this architecture should be a BEAST in integer thread execution. The only place where it might fall apart is in FPU calculations.

          • Prion
          • 8 years ago

          [i<]That said this architecture should be a BEAST in integer thread execution. The only place where it might fall apart is in FPU calculations.[/i<] This is why tighter cpu/gpu integration needs to get off the ground. Schedule the SSE/AVX/XOP/FMA/etc. FPU calcs to the specialized units on the "gpu" side, and let the general purpose "cpu" side do the things that it's best at.

          • evilpaul
          • 8 years ago

          Quad-core Sandy Bridge chips only have four FPUs in them despite supporting eight threads, so I think it could alright in that regard.

      • dpaus
      • 8 years ago

      [quote<]I know you guys like to include troll vote options for s**** and giggles, but half of them?[/quote<] Not just that it's half of them, but the half are entirely negative. Hey, I enjoy 's**** and giggles' as much as anyone, but why not throw a 'more powerful than a Cray' option, or a 'faster than a speeding bullet' choice in there too?

        • NeelyCam
        • 8 years ago

        X6 is a reasonable option IMO (the bottom two not so much), but SB-E could’ve been added as an option, or IB.

      • ronch
      • 8 years ago

      Judging by the price, If I were to do this poll, the choices would be

      1. Intel Core i5 2500
      2. AMD Phenom II X6 1090T

      Even Phenom II X6 debuted at a higher price than this!

Pin It on Pinterest

Share This