Comments Locked

81 Comments

Back to Article

  • wumpus - Wednesday, August 28, 2019 - link

    There's a formal definition for core? I've been into computer architecture since before Anand started this site and I've never heard one (I suspect ones exist that aren't specific to CPUs and that a Bulldozer would contain far more than 8 "cores").

    Any indication that the research for this case was done by the legal offices of IBM? Not that they have any issue with AMD or Bulldozer, just that the IBM-Power architecture seems to rely on IBM's definition of "core" for per core licensing and it is almost the opposite of AMD's definition.

    A stupid decision all around (although it may match the law, no idea about that. The law is certainly often an ass).
  • DigitalFreak - Wednesday, August 28, 2019 - link

    The layman expects 8 cores to mean 8 identical cores, not some slight of hand marketing BS.
  • DigitalFreak - Wednesday, August 28, 2019 - link

    You notice that Intel didn't call a Core i7 with Hyperthreading and "8 core" processor.
  • Xyler94 - Wednesday, August 28, 2019 - link

    I mean, the CPU did have 8 Integer execution "Cores", but only 4 Floatpoint execution cores, so each integer core has to share a floatpoint with another integer core. it's not like Hyperthreading/SMT, where each core could execute 2 threads, it was that each integer core had to share a floatpoint core with another, so how do you categorize that?

    AMD's marketing at the time called it 8 cores, but it was more like 4 cores in terms of performance. It was indeed a weird core... but it wasn't threads of SMT/Hyperthreading.
  • eek2121 - Thursday, August 29, 2019 - link

    Definitively not the first time this has happened. Likely won't be the last.
  • Techie2 - Thursday, August 29, 2019 - link

    The siren chasers collect millions off of these bogus lawsuits which are only allowed in the U.S. judicial system. Anyone qualified to even discuss this subject would know full well what a CPU core is and what a math coprocessor is. If not see Intel 286 CPU for a clue.

    By filing a meritless lawsuit unscrupulous lawyers force the hand of companies who could rack up tens of millions in litigation costs and then be subjected to the technical ignorance of a jury with the mentality to pay a woman a million dollars for pouring hot coffee on her crotch in a moving vehicle. This outrageous behavior that only exists in the U.S. judicial system is named: Jackpot Justice where paid liars dupe a jury into believing that it's always someone else's fault and that any corporation with deep pockets should pay for any perceived injustice even when as in the AMD CPU core case there was none. If you can't understand what a CPU core is then you should educate yourself instead of filing frivolous lawsuits that make bottom feeders wealthy. The real injustice in this case is that AMD is raped out of $12 million by devious players manipulating the system for profit. If $35 makes you "whole" in judicial terms then you were never "damaged" in the first place by buying an AMD CPU - which is precisely the case.
  • 0ldman79 - Tuesday, September 17, 2019 - link

    Well said.
  • Pinapple1977 - Wednesday, December 18, 2019 - link

    You should research the McDonalds hot coffee lawsuit before you go using it as an example. That lawsuit was infact NOT frivolous, the media blasted her at McDonalds behest- which goes to show the power of media. The Molten lava branded as coffee could have killed her. She was hospitalized. Infact she and her family first asked that they reduce the temperature of their coffee- which was refused- and not kindly. The lawsuit was launched there after. And rightly so. Beware of media smear campaigns, her case is an example of it, and how successful it can be.
    What you are saying about the CPU lawsuit is true. McDonalds coffee lady - Bad example for your argument.
  • mickulty - Friday, August 30, 2019 - link

    The thing is it was like 4 intel cores in 8-thread performance, but nothing at all like 4 intel cores in 4-thread performance outside of specifically 256-bit FP.

    Look back on it, the biggest issue IMO wouldn't be the FPU that people always focus on, but the front-end. The shared front-end on a module could only decode instructions for one core at a time (although it could switch every cycle). The thing that really messed it up for AMD though is their own marketing, trying to position the idea of "CMT" as some kind of equivalent to SMT and a bulldozer module as the equivalent of an SMT core.

    Of course, having said all of this, at the end of the day if Bulldozer had had the same architecture, the same limitations and the cores had just been faster then I doubt anyone would have complained in the first place. Realistically the problem is that people saw "8 cores" and expected perhaps 2x an i5-750 in performance, not that people expected some specific hardware implementation.
  • Samus - Wednesday, August 28, 2019 - link

    There's a huge difference between logical (virtual) cores and physical cores. Bulldozer still had physical cores. They were just weaker in IPC and greatly penalized with a pair of cores sharing a single floating point unit.

    Programs and operating systems still detected every core as a core, and they were silicon. So this is kind of a sketchy lawsuit and shows (unsurprisingly) people who purchased these CPU's - regardless of marketing - had virtually no understanding of what they were buying into.

    If the expectation was to have superior performance to Intel, well, that's ridiculous. AMD hadn't been IPC competitive with Intel for nearly a decade prior to Zen and anybody telling you otherwise was either trolling you or a blind fanboy. Adding more cores with incredibly weak IPC and the added FPU penalty wasn't going to magically enhance the product.
  • 0ldman79 - Thursday, August 29, 2019 - link

    This whole thing is a load of crap.

    The FPU is split anyway, it's just not a monster.

    If the software was optimized for it, it would behave more like an 8 core. As nothing was optimized for it, it behaved like an 8 core with 4 FPU.

    It's a baseless lawsuit and proves our government has no business making law regarding technology whatsoever.
  • nevcairiel - Friday, August 30, 2019 - link

    The notion of a seperate FPU hasn't been reality since like a 486.
    Regardless, the far more important part is the shared frontend, which does instruction decoding. This is very often a bottleneck in modern processors, so sharing it over 2 cores does noone any favors.

    It was basically one core with 2 integer pipelines. Not two cores with shared FPU and Frontend.
  • eek2121 - Thursday, August 29, 2019 - link

    Yeah, I'm sorry, I get the need to settle a lawsuit, but this is not something I would have settled for.
  • Xyler94 - Thursday, August 29, 2019 - link

    Sometimes, it is cheaper to settle than to try and win.

    In this case, AMD just paid 12.1Mil, which if they had fought more with Lawyers, they potentially could have paid more. They made the decision to settle for a low amount, than to fight it in court.

    In the end, laywers win
  • JoeyJoJo123 - Thursday, August 29, 2019 - link

    Lawyers are the only ones that win in litigation.
  • Operandi - Thursday, August 29, 2019 - link

    AMD's expectation was certainly to have superior performance. Bulldozer was a big, complex, and ambitious architecture it was without a doubt aimed to take on Intel on the high-end of the market. If things would have worked out Bulldozer should have crushed Intel in general purpose work loads and pretty much anything that didn't heavily rely on complex floating point math. Of course thats not exactly what we got....
  • silencer12 - Thursday, August 29, 2019 - link

    Correction "You notice that Intel didn't call a Core i7 with Hyperthreading an "8 core" processor."

    Notice I fixed An for you instead of And
  • Manch - Friday, August 30, 2019 - link

    Yeah the 386 and some 486 did not have an FPU, That was a physically different chip and extra $$. There are a lot of things included in cores now that were not included before. Now AMD has remove the memory controller back out. You want to bitch about how many cores are in Zen? Bulldozer was a poor design by virtue of noone building for it. Still those porcs are 8 cores. This lawsuit was bullshit but 12mil is probably the cheaper payout.
  • RedGreenBlue - Wednesday, August 28, 2019 - link

    Bulldozer cores do have identical cores. What? You want Intel to re-label their old chips when they double an FMAC pipe for a new architecture?
  • eek2121 - Thursday, August 29, 2019 - link

    No, not really. It blows that AMD let it get this far. However, the last CPU without an FPU was the 486 IIRC (sorry, a bit rusty in my old age). Anything Pentium and up had an FPU.
  • RedGreenBlue - Monday, September 2, 2019 - link

    The bulldozer cores had FPUs. It is no less accurate to say that each core had a 128 bit FPU than it is to say they shared a 256 bit FPU. The only thing you could argue over is whether the scheduler had the capability for two FPUs, and if I’m not mistaken, AMD increased the FP scheduler’s entry capacity at least once. The goal of the architecture was to avoid one core’s FP being bombarded, while the abilities of another core were idle. The problem comes when the FP scheduler, or front-end, can’t keep up. They were constrained by TDP on the 32nm process such that they had to make sacrifices on those. Sacrifices they later found ways to improve. Some of it was also due to the block designing of the architecture, which was not efficient for layout or power efficiency, basically a build-to-order design rather than hand-crafted. I think Anandtech has multiple articles pointing out the issues and changes, but it caused a regression in IPC and an increase in pipeline length.
    The last editions of the bulldozer family made huge strides year after year in IPC, gains we would be happy to get with future versions of Zen.
  • Kvaern1 - Wednesday, August 28, 2019 - link

    There are plenty of examples of much worse core advertising. My TV was advertised as having 6 cores...I believe 4 of them are CPU cores, the 5th is the GPU and the 6th I have no idea what is.
  • lmcd - Wednesday, August 28, 2019 - link

    GPU is probably separate from a VPU with codec acceleration
  • patel21 - Thursday, August 29, 2019 - link

    Now that I read you, I remember Google also advertising its first Moto X something similar:

    The device is powered by a chipset branded as the X8 Mobile Computing System, which consists of a dual-core, 1.7 GHz Qualcomm Snapdragon S4 Pro system-on-chip with a quad-core Adreno 320 GPU, a custom-designed Natural Language Processor core and Contextual Awareness Processor core (for a total of 8 cores),
  • eek2121 - Thursday, August 29, 2019 - link

    What TV? I've never seen a company advertising a TV, period.
  • Kvaern1 - Thursday, August 29, 2019 - link

    eek I'm not sure what you mean but when the official company webpage says the TV has a hexacore processor then it's advertised as having 6 cores. It doesn't literally have to be in a commercial or add they mention it.

    It's a Philips TV.
  • Lord of the Bored - Wednesday, August 28, 2019 - link

    It does. Eight processor cores, and four math coprocessors.
  • Cakemaster - Friday, August 30, 2019 - link

    What on earth do you mean by expecting "8 cores to mean 8 identical cores", are you implying that the problem with these CPUs is that the cores weren't identical to each other? These CPUs did have the number of cores which they were advertised as having, they were just very mediocre CPUs overall. They were charging significantly less PER CORE than Intel was, and these CPUs were actually very reasonably priced.

    Furthermore, they actually do run quite well on software which is well optimized for them, and they have actually aged very well, as 8-core FX CPUs have closed a ton of ground vs. Intel CPUs from the same time period. AMD never claimed that each core was equivalent to the cores in other CPUs. Everybody knew back then that most software wasn't even well optimized for 4-core CPUs, and also that you should never make assumptions about the numbers in the product name or the spec sheet. The bottom line is always how well they actually peform according to quantifiable test data. Not even stupidity would explain why someone would have bought an FX CPU based on the assumption that more cores will always translate to higher performance, or that that the cores were equal to the performance of Intel CPUs which cost WAY MORE per core. It's not just a lie that these didn't have 8 cores, the idea that people were being mislead into buying these because they thought that they would perform better than Intel chips is also a lie.
  • Toe - Wednesday, August 28, 2019 - link

    Let's put it this way: nobody would refer to a 486SX - which didn't have an FPU - as a 'zero core processor'.
  • eek2121 - Thursday, August 29, 2019 - link

    Thanks, yeah, while AMD was a bit sketchy here, If I were the judge, you couldn't have sold me this argument with a golden 10 foot pole. What makes it even more hilarious is that AMD and Cyrix both had much better CPUs at the time. We didn't see Intel getting sued.
  • eek2121 - Thursday, August 29, 2019 - link

    It's actually rather sad. I say this for both Intel and AMD. I remember when an FPU was optional. Why AMD settled for this I will never understand.
  • ABR - Wednesday, August 28, 2019 - link

    When I first saw the Bulldozer architecture I thought they were going to market, for example, these "8-core" chips as having 4 cores. Each one packing two integer pipelines together with unified fetch/decode, FP, and cache. If they'd actually done this they might have actually looked pretty good next to Intel. The two integer units would have supported better hyperthreading, and IPC would have been far more competitive. But for whatever reason they decided to present each unit as two cores, perhaps feeling they could win some kind of "core wars" marketing competition that way. They got slaughtered, and the rest is history. Which is too bad, because Bulldozer was a real attempt to innovate. Good to see them get their mojo back with Zen.
  • eek2121 - Thursday, August 29, 2019 - link

    Except they were weak, even against dual core Pentiums or Core i3s. I'm not one that likes to debate either way. However, I've owned so many CPUs over the past few decades that IMO it's AMD's fault for not taking a stronger stance on this one. I get why they would want to settle, but honestly, they likely would have won in the end. AMD need only rely on the 486SX or 386SX to establish their case. I suspect none of the parties involved are even old enough to remember when a CPU lacked an FPU. My first Windows 95 machine lacked an FPU FFS.
  • TheinsanegamerN - Thursday, August 29, 2019 - link

    It's not 1993 anymore. The definition of "core" has changed. AMD didnt advertise a quad core with quad runt cores, it advertised an 8 core, as in 8 identical cores, which they were not.
  • Cakemaster - Friday, August 30, 2019 - link

    Yes. It's totally fucking absurd. Not only did the court believe the lie that these chips didn't really have as many cores as advertised, they also believed the lie that people were being convinced to buy them because they had more cores.

    The truth is that the way AMD marketed these CPUs made people think that they were WORSE than they really were, not BETTER than they really were. The people who did buy these CPUs either did know what kind of performance they were gettting, and anybody who didn't know what they were getting was buying them based on pricing, not the number of cores. For what they cost, they were perfectly decent CPUs. The only people getting ripped off would have been people who trusted lying sales representatives who tried to claim that more cores= always better, but you can't hold AMD responsible for that! Anybody who claims that they were mislead by AMD is themselves a lying scumbag who just wants to cash-in on lies and an incompetent legal system.
  • Smell This - Wednesday, August 28, 2019 - link

    Interesting the '8-core' was included with 6- and 4-core omitted.

    I've got an FX-6350 in a rig that purrs along at 4.5GHz/1.3v ... tested at 4.8GHz/1.35v without breaking a sweat (I didn't want to get greedy!). Works just dandy on a 4K Vizio TV.

    It will work a lot better in 6 months when the RX 5700-series OEM cards start slugging it out with nVidia in a smack-down price war ;-)
  • plewis00 - Wednesday, August 28, 2019 - link

    This travesty of an architecture is why IPC basically didn’t move for 7 years til Zen because Intel didn’t need to put anything substantially better out. It was blatant mis-selling and they should be penalised. If Intel had done this, people would be out for their blood. At least AMD can turn it around with Zen, which let’s face it, surprised us all - an amazing piece of design and architecture.
  • RedGreenBlue - Wednesday, August 28, 2019 - link

    This lawsuit is the result of companies with IT departments that didn’t understand computer architectures. Nothing about the core design was hidden from the public, AMD was very clear about it, and anyone with half a brain would have looked for a review like the one on anandtech and would have understood the floating point side would function as a full 2x128 up to 4 threads and then 128 on 8 threads, assuming the FP was actually being used.
    They would have won this lawsuit and they should have stuck it through to get the term “core” solidified as an integer core, preventing frivolous lawsuits for future architectures. 12 million is less than the cost of the lawyers. I’m amazed this lawsuit was still going after all these years. Somebody must have slow-walked it.
  • Samus - Wednesday, August 28, 2019 - link

    ^^^this

    I read most of the class members are, in fact, corporations. They basically bought cheap shit HP's and Lenovo's with AMD CPU's and wondered why Pentiums and i3's were blowing them out of the water in performance, and IT just shrugged saying "they play our games fine so we thought they was moar bettr"
  • Arbie - Wednesday, August 28, 2019 - link

    Gotta love it when people blame AMD - who struggled through everything, including Intel's illegal dealings, to miraculously come back from the edge and save the desktop world - for Intel's sloth and mismanagement. You're mad because you got very little improvement for ten years. But you're blaming the company you should be thanking. Intel is the one that screwed you, not AMD.

    Who says AMD doesn't get an even shake?
  • Korguz - Wednesday, August 28, 2019 - link

    simple, like one person on here thinks, intel is the god of the cpu world, and can do nothing wrong, but does everything right. and also seems to imply, that intel has done more for the cpu, then any one else in the world, thats why amd doesnt get a shake.
    " Intel didn’t need to put anything substantially better out " but they COULD of, instead, they stagnated cpu development, kept mainstream stuck at quad core, and kept raising prices. if it wasnt for Zen, we could very well be still have quad cores in the mainstream, and cpu prices, could of been even higher. maybe he is blaming amd, because we all have been screwed by intel for so long, we dont know the difference ?
  • plewis00 - Thursday, August 29, 2019 - link

    If you’re stupid enough to read and misinterpret my comment as ‘Intel are gods’ then I’m not wasting more time - it’s easy to take an argument to extremes for your own end. If this was the other way round, everyone would be happy to get the pitchforks out. The Bulldozer setup was awful, it misled buyers into thinking they had a great multi-core setup in the same way as Intel did the clockspeed race before it. And don’t say ‘buyers should have done their research’ as they don’t - I still see people ‘buying an i7’ because it’s ‘better for gaming’ when they’re choosing an ultrabook. Instead of chasing IPC, you got power savings for the last few years. Of course it’s cool to hate Intel but you need two or more chip makers in the market, and you’ll buy the best, which is basically Zen right now.
  • Korguz - Thursday, August 29, 2019 - link

    FYI.. plewis00.. i wasnt referring to you....
  • plewis00 - Thursday, August 29, 2019 - link

    Ok sorry. Miscommunication, my apologies.
  • Oxford Guy - Thursday, August 29, 2019 - link

    AMD also decided to try to get profit from enterprise by making Bulldozer an enterprise-centric design.

    Had it decided to go after gamers it wouldn't have cut the floating point resources, obviously. It wouldn't have gone for 8 cores. It would have, instead, released the Cinebench killer — a CMT part with 8 floating point cores and 4 integer cores!
  • Haawser - Wednesday, August 28, 2019 - link

    Was a garbage lawsuit when it started and is still garbage. AMD probably figured it was cheaper to just pay $12m now than keep hiring lawyers to take it through endless plaintiff appeals.

    Given that the bloodsuckers have already racked up $3.6m... they're probably right.
  • 0iron - Wednesday, August 28, 2019 - link

    It amazed me that the lawyer got 30%, it should be 10% max.
  • Charlie22911 - Wednesday, August 28, 2019 - link

    Tangentially related: Waiting on lawsuits against ISPs marketing “Unlimited” internet. I’ll go ahead and hold my breath.
  • Samus - Wednesday, August 28, 2019 - link

    I'm waiting for the lawsuit against zero calorie "naturally" flavored waters that actually have around 5 calories :)
  • Thunder 57 - Thursday, August 29, 2019 - link

    Didn't you read, zero calories "per serving". The whole serving thing is so stupid since they are unrealistic 98% of the time.
  • plewis00 - Thursday, August 29, 2019 - link

    There was a lawsuit the other day against zero alcohol beer because it’s actually 0.1% or something like that and some Muslim woman was getting bent out of shape because her husband had violated religious code. It’s just the way you’re allowed to advertise things. I’m sure it doesn’t bother the majority. This cores thing was a big deal to people who potentially didn’t know better.
  • Oxford Guy - Thursday, August 29, 2019 - link

    I hope they don't consume citrus, because it has methanol in it. They may be violating their superstition by consuming anything with pectin in it, too:

    "Methanol is also formed when fruits and vegetables are physically prepared for consumption by slicing, chopping, pureeing and juicing. The production of free methanol in all these instances is the result of reactions in pectin, a principal component of plant cell walls and the middle lamella between them. Pectin is heteropolysaccharide contained in the primary cell walls of terrestrial plants. It is produced commercially as a white to light brown powder, mainly extracted from citrus fruits, and is used in food as a gelling agent particularly in jams and jellies. It is also used in fillings, medicines, sweets, as a stabilizer in fruit juices and milk drinks, and as a source of dietary fiber. Pectin will break down to methanol when the plant cell walls and middle lamellae are disrupted, as can happen through physical processes of food preparation. Methanol is also produced when pectin is digested after eating fruits and vegetables."

    Religious insanity should not be indulged, in any form. It's a fundamental attack on efficiency, which means it drags everyone down.
  • Samus - Friday, August 30, 2019 - link

    lmfao for real? so what are they going to call it now, light beer?

    ohh. yeah that...
  • Rickyxds - Wednesday, August 28, 2019 - link

    Like I always said

    buldozer is eight (cores halves)

    In math language is 8 * (1/2) = 4 cores

    And the performance never lied
  • RedGreenBlue - Wednesday, August 28, 2019 - link

    Integer performance was always 8 cores, and mixed performance between FP and integer was much better than 4 cores. Only in a straight FP environment could you argue the performance was the same as 4 cores. Even then there’s the variable of what size the floating point unit was, AMD could have done a 2x256 FP unit instead of 2x128. What then? This is why integer cores are widely accepted as the least common denominator of the term “core” when everything else fluctuates.
  • RedGreenBlue - Wednesday, August 28, 2019 - link

    *4x128 FP unit instead of 2x128
  • Chaser - Wednesday, August 28, 2019 - link

    I can remember back then when the buyers were boasting about their "8-core" CPUs. LOL
  • plewis00 - Thursday, August 29, 2019 - link

    Yes but if you speak out or joke about it, apparently you’re stupid and pro-Intel... that’s the general reaction.
  • Oxford Guy - Thursday, August 29, 2019 - link

    Fact is that code that is reasonably optimized for the Piledriver architecture and inherently compatible with its philosophy can perform more than adequately on it.

    For example, the game Deserts of Kharak showed excellent performance parity with 8 core Piledriver CPUs.

    This comes despite the fact that AMD, being a much weaker company, in large part thanks to Intel's dirty tricks that robbed it of much-needed profit that would have gone to better R&D, released these parts with significant flaws. Those flaws would have been less likely to have appeared had AMD been financially stronger at the time. And, it wouldn't have just let Piledriver rot for so many years.
  • KAlmquist - Wednesday, August 28, 2019 - link

    Here are some scaling numbers for Cinebench R10, calculated by dividing the multithread score by the single thread score:

    12023/3560 = 3.38 (FX-4130, 2 modules, 4 cores, 3.8 Ghz)
    13242/3675 = 3.60 (Phenom II X4 955, 4 cores, 3.2 Ghz)
    19179/5423 = 3.53 (Core i5 2400, 4 cores, 3.1Ghz, Sandy Bridge)
    15060/4512 = 3.34 (Core i5 760, 4 cores, 2.8 Ghz, Nehalem)

    Cinebench uses floating point, so the shared floating point unit comes into play here. The speedup seen on the 4 core Bulldozer is a bit on the low side, but not by a lot. On all four processors there are shared resources which prevent a perfect 4.0 speedup from being achieved.

    For anyone who still thinks the module design can be compared to hyperthreading, here are some numbers for processors with hyperthreading:

    11046/5095 = 2.17 (Core i3 2100, 2 cores, 4 threads, 3.1 Ghz, Sandy Bridge)
    24933/6397 = 3.90 (Core I7 2600K, 4 cores, 8 threads, 3.4 Ghz, Sandy Bridge)
    16598/4490 = 3.70 (Core i7 860, 4 cores, 8 threads, 2.8 Ghz, Nehalem)

    Hyperthreading allows a little more performance to be extracted from each core, but nothing close to what you would get from actual cores.

    You may have noticed that Bulldozer has the worst single-thread performance of any processor listed in this comment. It gets crushed by Sandy Bridge, which was released about three months before Bulldozer, but it's not even really competitive with Intel's Nehalem design. It can't even match AMD's own earlier offering, the Phenom II. <em>That</em> the problem with Bulldozer; it's not that you don't get the number of cores advertised; it's that they are <em>slow</em> cores.

    If you bought an FX-4130 on the assumption that it would be the same speed as an i5 2400 because they both have four cores, you would be severely disappointed. But that's not because AMD misrepresented the number of cores. It's because you can't predict performance based solely on the number of cores; you also have to know how fast the cores are. The premise of the lawsuit is bogus.
  • RedGreenBlue - Monday, September 2, 2019 - link

    Amen.
    I actually have done a lot of experimentation with Engineering Sample Interlagos chips. Overclocking, benchmarking, and plotting IPC with different settings in all scenarios. An odd thing came about, the modules sometimes have incredible single thread performance with CMT off. It’s a tradeoff overall, but for times when single thread efficiency matters in gaming, it can be worth it. That 16 core ES chip is a gold-mine of fun for experimenters, enthusiasts and overclockers
  • RedGreenBlue - Monday, September 2, 2019 - link

    Sad thing is, Zen makes all those aspects obsolete, except good old fashioned overclocking fun.
  • crashtech - Wednesday, August 28, 2019 - link

    It became pretty obvious from maintaining the Cinebench R11.5 thread that classifying the big FX's as 8 core CPUs would have made them look embarrassingly bad versus actual 8 core CPUs, so the category they went in became 4M/8C and 4C/8T, real world results made it pretty clear that a module does not 2 cores make. Heavily overclocked FX's could just keep up with Sandy Bridge i7's, but that was about it.
  • KAlmquist - Thursday, August 29, 2019 - link

    I can't find the thread crashtech is referring to, but I suspect that the 8 core FX processors didn't merely look embarrassingly bad versus "actual 8 core CPUs;" they actually were embarrassingly bad. If it were just a matter of AMD misrepresenting the core counts, then the 8 core FX processors would match the 8 core Intel processors on the single thread benchmark, and only be embarassingly bad on the multicore benchmark. I'm guessing that the Intel 8 core processors trounced AMD on both variants of the benchmark.
  • Oxford Guy - Thursday, August 29, 2019 - link

    1) Cinebench is a primarily floating point workload, so the 8 core Bulldozer design was already going to perform worse than an Intel 8 core — because it has less die space devoted to floating point performance.

    2) The developers of Cinebench are not gods. Their choices are their own.
  • duploxxx - Thursday, August 29, 2019 - link

    What a huge pile of bullshit.

    Stupid lawsuit and judges. As if a "core" is a hard definition.
    HT and SMT are also seen in Windows as a core/processor. It is as misleading as 2 SCU units with 1 FP.
    HT and SMT are also not always able to run.
  • HeavyHemi - Thursday, August 29, 2019 - link

    However, Dimmy, they are not sold with H/T or SMT as being physical cores. Keerist...can you even read?
  • Xyler94 - Thursday, August 29, 2019 - link

    As of Windows 8, Windows can differentiate between logical process and whatnot, but it does show up as a logical process to Windows.

    Windows doesn't understand differences when it's a VM, but it knows it's a VM, so it just reports what it uses.
  • TheUnhandledException - Thursday, August 29, 2019 - link

    The short answer is there is no formal definition and that wouldn't matter anyways. Marketing and lawsuits about marketing are about consumer perception. If an average person would expect 8 cores to be 8 complete independent cores and you don't have that and sell it as that and said product underperforms 4 core competitors depsite being called "8 core" well you are going to payout in a lawsuit.
  • NICOXIS - Thursday, August 29, 2019 - link

    Well, this is just another proof that saving some silicon space in order to boost core count was the wrong choice.
  • Oxford Guy - Thursday, August 29, 2019 - link

    It is no such thing.

    1) A Bulldozer-style design (in terms of there being more integer hardware than floating point hardware) is more efficient for workloads that involve little to no floating point on the CPU.

    2) AMD's designs, Bulldozer and Piledriver, had bugs and other drawbacks — like L3 cache performance that wasn't much better than DRAM and AVX performance that wasn't much better than SSE. AVX may have even regressed from the change of Bulldozer to Piledriver. They were certainly not the pinnacle of the CMT concept and were frozen in time while Intel continued to extract IPC increases with its many incremental improvements over Sandy.
  • Oxford Guy - Thursday, August 29, 2019 - link

    Some other design flaws and drawbacks:

    1) Inadequate micro-op caching, when compared with Sandy.
    2) Too-heavy reliance on automated tools, resulting in transistor count bloat.
    3) Very deep pipeline probably not efficient for desktop-oriented code, particularly of its time.
  • rocky12345 - Thursday, August 29, 2019 - link

    The CPU is a hybrid of sorts yes it can do up to 8 threads and it does do this as advertised. The problem is more how AMD marketing marketed the FX series and made it out to be more than it was. As said it can do 8 threads but at it's heart windows see's it as a 4/8 setup kind of like windows see's Intel's i7's up to the 7700 CPU as a 4/8.
  • Oxford Guy - Thursday, August 29, 2019 - link

    How Windows, Cinebench, or other software's coders decide to classify it only proves one thing: how they chose to classify it. It's my understanding, too, that Windows had both 8 and 4 core classifications for the same 8 core FX chips at different times in its development.
  • ballsystemlord - Thursday, August 29, 2019 - link

    I was hoping someone would do this. A core should be at least one INT and one FLOAT unit, not one INT and one half FLOAT.
  • Oxford Guy - Thursday, August 29, 2019 - link

    That's arbitrary. A workload that offloads FP work to a GPU or other specialized processor has zero need for FP on a CPU — making the 1:1 INT to FP ratio a ridiculously inefficient use of die space.

    Similarly, workloads that use little or no FP do not need your "should" mandate.
  • Oxford Guy - Thursday, August 29, 2019 - link

    It's droll that when AMD led in floating point performance, all the benchmarkers pushed the importance of integer. Then, when AMD focused on integer at the expense of float, suddenly floating point performance is the cat's meow.

    It's also droll that so few realize that Bulldozer wasn't designed for desktop users. It was designed for enterprise. It's not accidental that a good number of supercomputers in top-performance lists used lots of Piledrivers.

    The other things many forget:

    1) Intel cheated on security to improve performance.

    2) AMD's high-performance CPU design was static from Piledriver until the time of Zen 1, which was a very long time for a part that already lagged the competition.
  • AntonErtl - Friday, August 30, 2019 - link

    For those who think that FX-8xxx does not have 8 cores: Consider an alternative Bulldozer design, where half of the 4 64-bit FP adders and half of the 4 64-bit FMACs of a module was assigned exclusively to one core; and likewise, where each core got exclusive access to half of the I-cache and L2 cache, and the instruction fetchers and decoders were halved in size. You would get a CPU that would have even worse single-thread, and probably also worse 8-thread performance than Bulldozer, but even some clueless judge would find no reason not to call such a castrated FX-8xxx an 8-core CPU.

    This shows that buying by the number of cores and suing on this basis is foolish. OTOH, marketing by the number of cores tries to exploit this folly and this makes it ethically questionable; does AMD deserve to get punished for that? Maybe, but then I would like to see other ethically questionable marketing punished, too.

    If you only read one other comment, read KAlmquist's excellent comment https://www.anandtech.com/comments/14804/amd-settl...
  • Hardware Hound - Friday, August 30, 2019 - link

    In the end, my biggest issue with this settlement is in the performance. FX did have solid multi-core benchmarks, indicating that the 8 cores were legit. The main performance issue the series had (as previously mentioned) was the single-thread. Settling in that regard seems like a mistake.

    However, I'm also dealing with a lot of marketing projects to help the hospital I work at try and pass a tax increase vote. Public perception requires weighing out a ton of pros and cons. I think AMD made the smart move by settling. It not only seems to be saving them the potential litigation costs, but it's a solid move for consumer good will, so to speak.

    Had this gone to trial though, I have a feeling AMD could have won the case eventually if they really wanted to fight it.
  • johnniedoo - Saturday, August 22, 2020 - link

    I actually got a check from AMD looks like. i deposited it into my checking /savings account and it has been returned to me and instructs me to deal with the maker of the check...AMD or the Clerk who administered the paltry 12.1 million bucks...
  • AMDFailsAgain - Saturday, June 12, 2021 - link

    I recieved my settlement offer check today...

    for $1.75...

    Seriously?

    The lawyers sure got their chunk and now us defendants are getting less than two dollars. This is the highest form of thievery.

    The lawyers needed us to sign up and "be heard" so they could get the largest amount they could and now I am getting $1.75 of a $12.1 million dollar lawsuit. I really might turn around and sue Tony Dickey and I'm damn sure not cashing this $1.75 check, because I do NOT agree.

Log in

Don't have an account? Sign up now