KitGuru KitGuru.net – Tech News | Hardware News | Hardware Reviews | IOS | Mobile | Gaming | Graphics Cards
Over the weekend, we learned that AMD had been hit with a class action lawsuit over its Bulldozer series of processors. The complaint stemmed from claims that AMD overstated the number of cores contained in the chip, stating that while AMD advertised the CPU as having eight cores, it only really had four due to a unique design that combined two discrete cores in to a single module.
Each module on Bulldozer is seen as two separate cores within Windows. However, the cores share a single FPU, along with the same instruction and execution resources. Intel CPUs on the other hand feature a different design with independent FPUs.
The lawsuit goes on to claim that due to this design, each core is not capable of working independently, which apparently results in performance degradation. The main point here seems to be false advertisement, as not everyone will understand the inner-workings of a CPU and would trust AMD to provide correct specification details.
These are all pretty serious allegations, which would put AMD in violation of the Consumer Legal Remedies Act in the US. After a few days, AMD has had a chance to issue a comment regarding the matter. In a statement sent out to various press outlets, AMD said: “We believe our marketing accurately reflects the capabilities of the Bulldozer architecture which, when implemented in an 8-core AMD FX processor, is capable of running eight instructions concurrently.”
Discuss on our Facebook page, HERE.
KitGuru Says: AMD has now had a chance to deny that it misrepresented its Bulldozer CPUs, now we just have to wait and see how this plays out in the long run. Do any of you think that AMD misrepresented the core count on its Bulldozer CPU?
In theory AMD FX Dose have 8 Cores… I think the real issue here is, What is the Definition of a Core or what counts as a core… (you don’t need a FPU in a Core to make it a core)
Exactly. And this is where the guy pushing for the lawsuit is crazy: Are we basing the definition of a “core” off of Intel, AMD’s rival? Or is it going to be through AMD’s definition because it is their own processor? AMD came up with CMT, and they get to define such things how they want.
Ironically enough Intels original CPU didn’t even have FPU on die. They were separate chips, And optimal. 8086-80386 required the user to buy a fpu chip and install in on the mainboard.
Yep. Didn’t know that until a little while ago… I’m relatively new to PCs (though I have already built my first build…)
yes amd is the true innovator of pc technology.there would not be unified shaders,hbm memory,async compute shaders,hsa foundation, and dx12 would not be no were good as it is if it wasn’t for amd.
This lawsuit is BS, AMD’s cores can do 8 things at once. Most work doesn’t need to use the FPU, and even that can be split in half to be used by 2 cores.
Sure the single-core power isn’t as good as Intel, but it’s good enough for most work and games. I have an 8350 myself, and when I convert music files for example, 8 of them get converted at once. It’s actually pretty impressive to see.
When multi-threaded operations work on an AMD 6 or 8 core, they can work really well. The only weakness I’ve seen is single-core performance in emulators, particularly in software rendering. Even still, an 8350 or 8370 does pretty well even with emulators.
That said, the 6120 I had earlier did struggle a bit sometimes, and of course Intel i5’s and i7’s will let you do better software rendering, when you need to use that for higher compatibility. I learned that while using emulators to run my old games at higher resolutions. Blurry textures when playing my old games is not for me.
Yes, I use emulators to play games I own, and I seriously doubt I’m the only one. I don’t have enough room under my TV for my old consoles, and even my handheld games play better at higher resolutions and with a controller in my hand.
I would say yes amd has 8 cores even more so than intel with an i7 when windows would show 8 cores in task manager for a i7. We all know it is 4 real and 4 fake or as intel like to say hyper threaded at least in AMD’s case there are actually enough hardware to declare 8 cores for real. This case has no merit and will fail.
And what they said is exactly true. I have ran a test and it ran all 8-threads simultaneously. It was tested with BurninTest. Many who have looked into this knows it can run 8 instructions concurrently, this includes electrical engineers etc.
The definition to me is the ability to independently execute and process data within a unit. With AMD cores it can run 1 thread per core, which means it has that requirement to be considered a core. Intel can also, really any core in any CPU has that ability.
In the Bios you can set all 8 cores to perform individually art least on the sabertooth. If they were mated cores this would not be possible.
64 bit extensions, on chip L2 cache, DDR (remember the DDR vs Rambus), Floating point SIMD multimedia extensions (intel initially used integer SIMD), integrated northbridge and memory controller, integrated single multi-core designs, etc… There are so many things in modern computing that we owe AMD.
It’s ridiculous that someone would use Intel’s definition for what a “core” is when it’s up to AMD to define what is a “core” to them. The fact that these tools obviously want to meaninglessly bleed a company of cash, is just a prime example of what’s so wrong with the legal system and basic human stupidity
There does need to be either more standardized definitions or specific details pertaining to the differences, so that a company can’t make something that appears to be the same as the competitor but actually performs worse because of the design.
Similar to how a virus in the tech field and a virus in the biology field are different, but within their respective fields the definitions are the same.
It’s funny how when some specific design makes for improved performance they point that shit out all day every day, but whenever a design doesn’t look favorable they will just leave it hidden in the technicalities that most plebs won’t understand. That goes for both AMD and Nvidia
8 individual sets of 4 integer processing units means it is indeed a 8 core processor, each of these sets can be fed their own instructions to execute concurrently each clock cycle.
Trouble is the majority of consumers don’t even remotely understand the complex pipeline of how CPUs and other processors work, and each stage of the pipeline will operate at a different speed and capacity, which is where sharing parts across cores is useful. Of course the increased complexity of the pipeline and smaller size of the individual integer clusters is what gives the architecture such poor single-thread performance, which only ever affects legacy code/applications anyway (besides draw call latency, which is being rid of anyway).
Ironically we still have the requirement of an extra floating point processor, just that nowadays it’s a multitude larger, more complex and commonly known as a GPU.
XD
Yes, that I must agree with, but without those standards it’s not up to some lawyer to milk AMD of money because no one objectively defined it in the past. That’s absurd.
America. The land of ‘Let’s sue somebody for the heck of it’.
It wasn’t until the first Pentiums that the FPU was included on die. My first pc was a 486 SX33, the SX denotes no FPU (or maths co-processor as we knew it) if it were a DX then the FPU would be on the CPU.
Make that the 80486 DX as first CPU with FPU on die then.
“We believe our marketing accurately reflects the capabilities of the
Bulldozer architecture which, when implemented in an 8-core AMD FX
processor, is capable of running eight instructions concurrently.” I am sorry but what ? your marketing was making it look like i7 killer at the time and we all know how it turned out, barely matching a stock i5, if this would be about false advertising then amd would be in deep shit, and no it does not even remotely show accurate capabilities, it shows best case scenario, and nothing even close to average consumer workload scenario in what these chips are most being used
No core has lacked an FPU for 20 years. AMD could throw together a few registers and call it a core, then go out for beers. I’m not having it. I say we not only face reality, but look after the consumer as well. I’m not a big fan of defrauding the consumer.
In modern graphics oriented computing, the FPU is the beating heart of the entire core. What can a “core” do these days with just an Arithmetic Logic Unit? That’s barely a calculator. “Let’s build a robot core, but to save money, we’ll include everything but the torso.”
There’s real reasons why every core for the last 20+ years has included an FPU. I’ll go as far as to suggest that the *only* reason for an architecture of this kind is to deliberately defraud the consumer, and I’m not having it.
Lol MS & Sony got duped too as according to this Lawsuit AMD sold them the same shared FPU design in Jaguar APU’s claiming it Nextgen. They should both join this Class Action and claim misrepresentation, failing that, console gamers should file too. There’s got to be money in Resolutiongate! :p
actually jaguar cores are fully fledged cores with their own fpu resources, quite different uarch that is more traditional approach than the cmt they used in bulldozer and further
Actually, someone commented that the fpu has been off the chip before, and that your thoughts on this mean nothing as a core is whatever Intel or AMD use to define it. There’s been no open standard as what a core is.
Actually those commentators were citing circumstances before the FPU as we know it were even invented. Some less polite than myself might even call that a disingenuous argument designed to fool the gullible. My statement remains no less accurate – we haven’t seen a core without an FPU for over 20 years.
And definitions don’t matter in the least. Try running virtually anything created in the last 20 years with a “core” which lacks an FPU, then insist to us that it’s still a core. Hint – you better do it in an operating environment that isn’t graphic. Good luck.
This is about maketing… AMD claims 8-cores as “performance” to the crowds… IT world knows that FX is not a “real core” cpu regarding performance… so, yeah… it has 8-kinda-cores.. but for the general consumer it has 8 real cores!.. (and APU is worst… i’ve seen advertising with 10 or 12 cores apu PCs.. ).. … “hey dude!, look!.. i bought a 12 cores gaming pc! and paid half of intel price for their 4 cores alternative!… I rock!”
This make me remember to Athlon … we had to explain to every client that 5200 is not Mhz.. but is kinda RATED to that.. but truly runs at 2700Mhz… (because when you was referring a P4.. was P4 2800Mhz or 2.8Ghz).. there’s no “model number”… so AMD “smart move” was fool the costumers
AMD was doing that since ages… and now is doing it in CORES.. becouse nobody cares about Megahertz anymore…
Due to this is a call to misinterpretation to the consumer… and for the sake of fair competition, they should be sued…
I think this is a lame cash grab by a lawyer who is hoping AMD doesn’t have the cash on hand to defend itself. Reminds me of Intel’s old business practices…maybe somebody should look into that.
First of all, who gets to define what a core is? Second of all, resource sharing isn’t really the problem in the bulldozer architecture. It’s a slower memory cache, 128 bit bus, and other factors, which made sense to do it that way at the time of the architecture being designed. Shared floating point isn’t really a big problem because floating point can be done out of order via smp. AMD actually has pretty reasonably beefy floating point units. AMD is stuck on a 28 nm node, while the competition has a finer node that lets them use more transistors, thus they can afford things like look-aside buffers and cache with redundancy that AMD can’t do yet. The lawsuit basically makes it illegal for anyone to have a different implementation than intel’s, or to be at a toc when another company is at a tic. And BTW, when do I get my money back for all the intel processors that had insufficient GPUs compared to AMD?
You’d have to sue intel too because they do the same thing
Jagua cores are not based on bulldozer
it doesn’t matter if no core has lacked a fpu for 20 years… The main point is cores to begin with never had fpu… The real definition of a core date way back to even cores were first evented. just cuz you don’t like it or don’t agree with it doesn’t mean its not fact
as were getting even feature into the future most code no longer uses the fpu and all that is done on the gpu now days etc some legacy code and drew cells
AMD sucks, more proof right here. They lied about all their Fury benchmarks as well, shocking.
Actually Cyrix 386DX266 did and the other funny thing about this little gem of a processor is that I have searched many Cpu history sites and this processor is never found in their databases. Seems to be quite rare, it was manufactured in 1994 so I assume it was a hybrid of 386/486. I do recall the reason I bought it was because at the time, it was faster that the Intel and Amd 486’s and about 60 dollars cheaper.
It’s still a core. What makes a core a core is by being able to run a task. 4fpus mean with 2 cores a piece is cmt with 8 cores. Get it right.
What were these situations where intel practiced these types of underhanded tactics?
Well, thank you God of Chip Design for solving that problem by providing a spontaneous but unnecessary definition you pulled out of, errrm, you obtained from on high. I’ll take that engraved on stone tablets, please.
But since AMD’s cores can’t “run a task” without its partner because of the need for the shared cache, thank you for proving MY point by demonstrating that it’s not a core. By your “definition”.
Speaking of points, since you don’t have any valid ones, I’ll be concluding our conversation, except to say that AMD is in real trouble, so they will be badly needing much sharper legal eagles than you probably think. You should offer them your obviously stellar insights.
It’s just as well that a fool doesn’t understand the simple concept of its not up to you to decide what a core is. It’s up to the designers which is…. oh yeah, AMD! Idiot.
What? You mean the Skylake 6700K doesn’t actually run at 6700MHz?
I’m calling my lawyer.
/s
http://www.pcworld.com/article/184882/A_History_of_Intels_Antitrust_Woes.html
thanks.
Yeah… nice one.. but i said that Megahertz isn’t a trademark of performance anymore… When it was… Intel didn’t do that.. What would you think if intel release a “new arquitecture” 12-core CPU but cheaper (and weaker) than current quadcores?… bring it on!.. right? Today..people are influenced by “core count” when they buy… and they dont even know what that is
care to expand?… i dont know where intel do that…
Those are model numbers, unlike those deceptive “PRxxx+” (Processor Rating) used back in the days of Socket 7 systems.
The lawsuit is based on AMD’s definition of cores http://www.amd.com/en-us/innovations/software-technologies/processors-for-business/multicore quote “With the power of two or more processors on a single chip, AMD’s true multi-core processors deliver industry-leading performance and unique features that help systems run cooler and more efficient”. That was on page 6 in the complaint http://regmedia.co.uk/2015/11/06/amd_bulldozer_lawsuit.pdf AMD’s Bulldozer “cores” are not processors, and cannot operate outside of the “module” because the “cores” lack the necessary components (like the front-end) to function as a complete processor. From http://www.csl.cornell.edu/~albonesi/research.html#clustered quote ” In a Clustered Multi-Threaded (CMT) microarchitecture, the core is divided into smaller, more scalable, clusters, with communication paths introduced between them. Instructions from different threads are assigned to clusters according to a steering algorithm implemented in the front-end of the machine”. That core became “module” and clusters became “cores” in AMD’s marketing. Even in AMD’s patent http://www.freepatentsonline.com/20090024836.pdf that “module” is originally the physical core, and that “core” was originally a cluster…
This lawsuit was based on AMD’s definition of cores http://www.amd.com/en-us/innovations/software-technologies/processors-for-business/multicore quote “With the power of two or more processors on a single chip, AMD’s true multi-core processors deliver industry-leading performance and unique features that help systems run cooler and more efficient”. That was on page 6 in the complaint http://regmedia.co.uk/2015/11/06/amd_bulldozer_lawsuit.pdf AMD’s Bulldozer “cores” are not processors, and cannot operate outside of the “module” because the “cores” lack the necessary components (like the front-end) to function as a complete processor.
It’s fully up to AMD to define what a core actually is, and just because a cluster is comprised of two cores, doesn’t mean they have a leg to stand on.
Referring to your comment “a cluster is comprised of two cores”, that “module” is not a cluster. Cannot place a core or cores inside a cluster, when a cluster is actually a part of the core itself. From http://www.csl.cornell.edu/~albonesi/research.html#clustered quote ” In a Clustered Multi-Threaded (CMT) microarchitecture, the core is divided into smaller, more scalable, clusters, with communication paths introduced between them. Instructions from different threads are assigned to clusters according to a steering algorithm implemented in the front-end of the machine”. That core became “module” and clusters became “cores” in AMD’s marketing. Even in AMD’s patent http://www.freepatentsonline.com/20090024836.pdf that “module” is originally the physical core, and that “core” was originally a cluster… https://uploads.disquscdn.com/images/96e6a2e4f0dcba7827dab0ad6a677bd3b7cc4130005c74008e5b4ed46060dde0.png
The lawsuit doesn’t have any technical basis.
AMD Bulldozer family uses conjoined cores, whose concept was introduced in the literature years before by the academic people advancing computer science. The paper in question is “Conjoined-core Chip Multiprocessing” by Kumar et al and published in IEEE Micro on year 2004. The definition of conjoined cores is already given in the abstract:
This paper proposes conjoined-core chip multiprocessing – topologically feasible resource sharing between
adjacent cores of a chip multiprocessor to reduce die area with minimal impact on performance and hence improving the overall computational efficiency.
The authors explored different configurations of sharing between pairs of adjacent cores including L1
caches, FPU, interconnect,… If you look at figure 2 you can find something very very close to Bulldozer module.
As mentioned in the MPR article about Bulldozer, AMD uses conjoined cores. The detailed description of the module, the pair of cores, the sharing, and the detailed work of the microarchitecture is found in the paper “BULLDOZER: AN APPROACH TO MULTITHREADED COMPUTE PERFORMANCE” published in the same
IEEE journal in 2011 and authored by the engineers of the Bulldozer project including the chief architect.
AMD is not the only has used conjoined cores. Sun engineers developed a microarchitecture where each pair of adjacent cores shared one FPU and clusters of four cores shared the fetch unit. Their design is described with all detail in the paper “ROCK: A HIGH-PERFORMANCE SPARC CMT PROCESSOR” published in the
same journal.
Intel has always had larger cash reserves than AMD, and they’ve used that advantage to cut deal with suppliers to force them not to use AMD chips. They’ve also engaged in lawsuits that, while ultimately frivolous, kept AMD tied up for long periods of time and forced AMD to sell parts of it’s business including Global Foundries, which used to be AMD’s own fab.
Maybe if you read what he was actually responding to you wouldn’t be so ignorant to say that… And if you don’t care to read, he was being sarcastic.
So if AMD had just called it a Model Number instead of Processor Rating but did absolutely nothing else differently, you would have been perfectly fine with it? I mean, you’re perfectly fine with Intel doing it now.
Nobody I knew was “deceived” by the Processor Rating thing. Everyone I knew understood perfectly well that it was being used to compare AMD and Cyrix chips to their Intel counterparts. Everyone understood that AMD’s 5×86-P75 indicated that it outperformed the 75MHz Pentium, not that it ran at 75MHz. And everyone I knew understood that the Athlon XP 2000+ only ran at 1666MHz but outperformed a 2000MHz Pentium 4.
Nobody was deceived then. The only people complaining about it being deceptive now are the ones who want to be able to complain about being deceived.
You know, I bought 2 intel laptops this year. They’ve been using the blanket statement “the worlds best processor” for a while, and I agree that they’re a bit faster than AMD right now, and their GPUs are getting a lot better, but there’s still features I sorely miss from my AMD machines, so I decided to pick up one of their A10 carrizo based machines, and I’m really happy with it. So, what is the “worlds best processor?” I know the FX wasn’t as fast as it could be, but it’s still a really fast computer. The worst thing we can blame AMD for is badly mis-predicting where the software environment was headed. They thought it would go integer for legacy applications, lots of cores for mainstream, and GPU for heavy floating point. While in reality, it went largely scalar SSE, and CPU concurrency has basically finally caught up, but it tool a while. GPU is still getting there. Really, we can be angry about it, or we can enjoy what innovations the different companies have to offer, because they are free to implement cpu’s and gpus the way they see fit, and that lets them bring unique products to the market. The only thing I can see out of this is maybe getting some revenge, and is that really the way we want to be remembered?
They don’t show as “fake” cores at all. Threads are displayed in task manager. Cores and Threads have been listed separately since Windows 8 btw. It also lists the 8350 as 4 cores and 8 threads.
I do believe the idea behind their shared floating point was something like this. Since floating point units lend themselves to out of order execution, that if they were in separate threads, they would have wasted resources, where having them shared would let a single, somewhat larger set of resources stay more saturated with operations coming from two different threads. It didn’t matter to the resources that they were different threads. That way, on average, they would stay more saturated using less overall resources.
In my experience, a lot of modern software is largely memory bound.
Again, they can use whatever nomenclature they want
Hey Kitguru! Why was my comment deleted? It wasn’t inappropriate or offensive, it wasn’t trolling, it answered the question being asked. Was it flagged? Was it deleted because it linked to a site that wasn’t Kitguru? We wouldn’t be trying to cover up references to Intel’s monopolistic anti-competitive business practices, would we?
Like I’ve said, just a CPU model number and not indication of frequency (in MHz). For your information, the plaintiff bought two units of AMD FX-9590 http://regmedia.co.uk/2015/11/06/amd_bulldozer_lawsuit.pdf however the CPU model number is not the reason nor part of the lawsuit claim. From the contents of the lawsuit, the plaintiff isn’t very well versed in computer and microprocessor technology (note the mistakes made)…
The “PRxxx+” (processor rating) is very deceptive by purporting its performance as equivalent or better than Intel CPU with the rated frequency. However that was mainly based on integer performance and quite often underperforms, example AMD K5 PR-166+ versus Intel Pentium 166 http://www.heise.de/ct/artikel/Schritt-fuer-Schritt-285464.html Furthermore I know a lot of people (mostly average Joes with basic computer understanding) who bought into those “PRxxx+” marketing. Most notorious were those Cyrix 6×86 chips and later many buyers complained games were not running well (from http://www.anandtech.com/show/34 quote “As you can tell by the Wintune 97 scores, FPU performance is the 6x86MX’s major weakness, it is no where near the performance of the Pentium MMX’s FPU not to mention the K6 and Pentium II, yet another reason not to consider the 6x86MX as a gaming processor”). AMD’s K7-based CPUs were different beasts compared to those Socket 7 era CPUs, and the main competitor were Intel’s Northwood-based CPUs (while Intel’s previous generation Willamette-based CPUs were simply bad performers).
You didn’t answer my first question, though. If AMD had said, “Oh that’s just a model number,” instead of saying, “Oh that’s to indicate that it matches or beats the Intel processor that runs at that speed,” you would have been perfectly fine with it?
Traditionally multi-core CPU uses CMP (Chip Multi-Processing), which literally means multi-processors on a single chip. Multi-threading is not multi-processing nor multi-processor (thus is not multi-core). If companies were to use multi-threading technologies marketing them as “cores” then the true meaning of a true (traditional) CPU core will soon be lost and also in dispute. From http://blog.stuffedcow.net/2014/01/amd-modules-hyperthreading/ quote “Ever since Intel’s Hyper-Threading and AMD’s Bulldozer modules, there has been much debate on what qualifies as a real CPU “core”. Unfortunately, I don’t think “core” is easy to define, so marketing tends to name things for their own benefit.”
AMD sold Global Foundries and various other bits long after the Intel settlement. These sales were due to bad designs, bad decisions and poor management from AMD. Nothing more.
Using model numbers are fine, as long as it does not represent any deceptive performance rating.
Not dredging up the past, but I remember people being constantly confused by the AMD rating back then. Unlike now, Mhz was used much more as a performance metric back then. It was fair marketing OMO, but was confusing for the average joe.
Then I guess some idiotic lawyer will force an arbitrary definition.
It definitely has 8 cores. I have it playing 8 games of chess at the same time 24 hours a day. It is Intel with its so called 16 threads in its 8-core that is the lie. Those other 8 threads can’t do much.
The main reason Bulldozer under-performed expectations was that the cache was too slow.
It can be argued that they exaggerated what they thought it could do. But anybody who does not look at actual benchmarks before they buy has no one to blame but themselves.
There are these things called benchmarks you know.
That is an arbitrary choice by programmers, not a measure. It is 8 cores, 8 threads.
Most people do not drive their cars at the red line either. I actually do use all 8 cores 24 hours a day playing 8 games of chess at the same time. It could not do that if it did not really have 8 cores. Just because they give you more than you use does not mean they have misrepresented anything.
Not running right this minute because my water pump failed but that is unrelated.
AMD’s FX series CPU “cores” are not real cores and are actually (integer execution) clusters. AMD is marketing their Clustered Multi-Threading technology as “multi-cores”, that’s why those “cores” have SMT behaviour http://www.tomshardware.com/reviews/windows-7-hotfix-bulldozer-performance,3119.html quote “Arun says that the dual-core modules have performance characteristics more similar to SMT than physical cores, so the company is looking to detect and treat them the same as Hyper-Threading in the future”. From http://blog.stuffedcow.net/2014/01/amd-modules-hyperthreading/ quote “Ever since Intel’s Hyper-Threading and AMD’s Bulldozer modules, there has been much debate on what qualifies as a real CPU “core”. Unfortunately, I don’t think “core” is easy to define, so marketing tends to name things for their own benefit”. As for Intel’s true 8 core CPU, the Core i7-5960X, those 16 threads are all useable, example video below shows an extreme example pitting 4C/4T against 4C/8T, 6C/12T and 8C/16T (using Intel’s Core i7-5960X)…
https://www.youtube.com/watch?v=-GgDZKGA89I
AMD’s Bulldozer “cores” are not “conjoined cores”, but are (integer execution) clusters as described in CMT (Clustered Multi-Threading) technology http://www.csl.cornell.edu/~albonesi/research.html#clustered quote ” In a Clustered Multi-Threaded (CMT) microarchitecture, the core is divided into smaller, more scalable, clusters, with communication paths introduced between them. Instructions from different threads are assigned to clusters according to a steering algorithm implemented in the front-end of the machine”. That physical core became “module” and those clusters became “cores” in AMD’s marketing. Even in AMD’s patent http://www.freepatentsonline.com/20090024836.pdf that “module” is originally the physical core, and that “core” was originally a cluster.
From the original “conjoined cores” paper http://rakeshk.crhc.illinois.edu/conjoining_micro04.pdf quote “For the conjoined-core chip multiprocessor, we consider four optimizations – instruction cache sharing, data cache sharing, FPU sharing, and crossbar sharing” and quote”We add the additional constraint that the sharing is topologically feasible with minimal impact to a conventional core layout”. That “conventional core” is what I’ve previously called “traditional core”. That’s why the paper had propose sharing mainly the cache and FPU only (and not any internal parts of the core itself). The performance degradation (as shown in the paper) is more inline with traditional multi-cores…
http://i.imgur.com/vNDyQCO.png
Sun’s UltraSPARC Rock is also using CMT (Clustered Multi-Threading), from http://www.cpushack.com/2015/01/16/sun-ultrasparc-rock-when-is-a-core-not-a-core/ quote “This type of arrangement is often called Clustered Multi-threading” and quote “This should sound familiar, as its also the basis of the AMD Bulldozer (and later) cores released in 2011”. In the end, that Sun UltraSPARC Rock CPU was a failure http://www.alphr.com/news/enterprise/357943/larry-ellison-slams-suns-astonishingly-bad-decisions quote “This processor had two incredible virtues: It was incredibly slow and it consumed vast amounts of energy”…
That ALU is the most important part of a CPU core (for integer arithmetic and logic operations), and nowadays with superscalar CPU core there are multiple ALUs (and AGUs, as address generation requires integer arithmetic also). Integration of the FPU only started with Intel’s 80386DX CPU. Before that, the FPU was external and as an optional upgrade (ie. as a plugin maths co-processor chip). Since the beginning of Intel’s Pentium era, the FPU has always been integrated. In the consumer desktop PC space, mainly games makes use of the FPU. Thus nowadays many consumers (including average Joes with basic computer knowledge) expects the FPU to be “part and parcel” of a CPU core.
AMD Bulldozer cores are real cores. They are conjoined cores or cores that share resources with adjacent cores. CMT is different to SMT, both techniques have different goals. Microsoft inability to deal correctly with CMT microarchitecture is the reason why had to launch a posterior patch for Bulldozer family. No other OS had to release a posterior patch to support Bulldozer family correctly.
The definition of conjoined cores was given in the post that you replied above. It is evident that Bulldozer cores verify the definition. But in case you cannot apply logic by yourself. You can also search online the Micropocessor Report article about Bulldozer, which states explicitly that Bulldozer uses conjoined cores.
Cornell’s CMT (Clustered Multi-Threading) is not the same than AMD’s CMT (Cluster-based Multi-Threading), and is not the same than Sun’s CMT (Chip Multi-Threading). You confound different techniques that have the same three letter acronym. And you link to that anonymous author of the personal website which is even more confused…
That MPR (Micro-Processor Report) a.k.a “The Linley Group” is more of a PR and marketing company nowadays AFAIK (puts out lots of dubious PR material). Their AMD Bulldozer article here http://www.linleygroup.com/newsletters/newsletter_detail.php?num=4008 They are especially infamous for IBM’s hypothethical performance claims and marketing comparisons, example http://www-03.ibm.com/systems/power/advantages/smartpaper/memory-bandwidth.html used by IBM for marketing (probably was sponsored by IBM also).
The traditional CPU core (e.g. Intel’s 8088 processor used in the first IBM PC) can function without cache and/or FPU, however cannot function without that front end (especially the instruction fetch unit). Likewise AMD Bulldozer “core” cannot function on its own, because it lacks that front end. Anyone can examine up that paper, and never stated anything about sharing internal parts of the core like the front end.
If you are referring to CPUShack Museum, and look his material http://www.cpushack.com/specs.html certainly he’s being following microprocessor technology for a long time. He was a known figure in the computer industry news, for example http://www.manufacturing.net/articles/2013/02/two-cpu-designers-who-changed-the-world quote ““On the original 8086 there was no way to context switch. You couldn’t run a multiprocessing OS on it,” said John Culver, who runs the CPUShack Museum. “Then Intel added protected mode on the 286 and 386 for multiprocessing. That’s how x86 and other architectures have morphed.””. Another example http://www.itworld.com/article/2827767/it-management/why-intel-can-t-seem-to-retire-the-x86.html quote “”I think they tried to do too much at the time, trying to integrate the latest and greatest out of universities that didn’t lend itself to hardware at the time,” says John Culver, owner of the CPUShack Museum and historian on all things CPU”
#1. juanrga quote “AMD Bulldozer cores are real cores. They are conjoined cores or cores that share resources with adjacent cores. CMT is different to SMT, both techniques have different goals.”
AMD’s Bulldozer “cores” are not real cores nor “conjoined cores”. Traditionally multi-core CPU uses CMP (Chip Multi-Processing), which literally means multi-processors on a single chip. Multi-threading (like SMT and CMT) is not multi-processing nor multi-processor, thus it is not multi-core. That is why AMD’s Bulldozer “cores” do not scale like traditional multi-core CPU http://blog.stuffedcow.net/2014/01/amd-modules-hyperthreading/ quote “The two-way multithreading speedup shows that AMD’s greater replication of hardware within a module results in greater performance improvement (1.54-1.57) compared to Intel’s minimal hardware replication for Hyper-Threading (1.23-1.24). In my opinion, neither of these are close enough to 1.9-2.0 to deserve being called a “core””…
#2. juanrga quote “Microsoft inability to deal correctly with CMT microarchitecture is the reason why had to launch a posterior patch for Bulldozer family. No other OS had to release a posterior patch to support Bulldozer family correctly.”
Microsoft’s Windows operating system does not treat AMD’s Bulldozer “cores” as real cores, but rather as SMT threads, otherwise there will be performance penalty. That SMT behaviour was also noted for Linux, example http://www.phoronix.com/scan.php?page=article&item=amd_fx8150_bulldozer&num=4 quote “In terms of any “sweet spot” or an ideal level of Bulldozer Linux support, it is not there quite yet. Similar to how Microsoft Windows 8 will offer up measurable performance improvements for Bulldozer over Windows 7, the Linux support can also be optimized for this distinct AMD architecture”…
He is using his own definition of core. From the lawsuit:
Then he accuses AMD because the FX-9590 cannot do eight calculations simultaneously according to him. He doesn’t explain what mean by “calculations” and on another part mixes calculations with instructions.
His definition of core is wrong. Moreover the FX-950 can execute up to 32 instructions per cycle or 16 ‘calculations’.
You are using your own definition of core, but you are not who decide what is a core and what is not. In the second place you continue confused about what is a core in Bulldozer. Cores in Bulldozer have front-end. Each pair of cores shares the front end. It is evident you don’t understand the concept of sharing. Your claim that Bulldozer cores lack front-end is so ridiculous as if you claim they lack L1 cache or FPU, because the cores share those. The guy that filled the lawsuit is confused but you are not much more clever.
Bulldozer cores are real, are conjoined cores and work as expected.
CMT cores are technically between SMT cores and CMP cores: SMT CMT CMP.
Without the patch the OS scheduler treated the architecture as pure CMP, which generated deficiencies because one module is not dual-core CMP. It is natural that Microsoft uses part of their former SMT codes in the Bulldozer patches to correct its scheduler, because the Bulldozer microarchitecture is between SMT and CMP. Moreover, Microsoft refers to its own patch as optimized for the “AMD Bulldozer module architecture”.
Incorrect, AMD’s FX series CPU “cores” (technically should be integer execution clusters) do not contain any front end. Even AMD’s own marketing slide, under “dedicated cores” shows that (below). That shared front end is in the “module” but was never part of those “dedicated cores” (external to those “dedicated cores”)…
http://www.techpowerup.com/img/10-08-25/bulldozer-8.jpg
AMD’s FX series CPU cores are not “conjoined cores”, nor real cores. As mentioned earlier, multi-threading (like CMT and SMT) is not multi-core. Even in the “conjoined cores” paper http://rakeshk.crhc.illinois.edu/conjoining_micro04.pdf data has shown that “conjoined cores” behaves like real cores, including the throughput and/or scaling (performance degradation was less than 20%), quote…
http://i.imgur.com/vNDyQCO.png
Additionally Linux treats AMD’s FX series CPU “cores” like Intel’s HyperThreading, example http://blog.angulosolido.pt/2014/01/core-confusion.html quote “From what is present on /proc/cpuinfo we see that Linux has NOT adopted AMD’s terminology. In fact it is treating AMD’s core duplication in the same way it treats Intel’s Hyperthreading: the number of cores is exactly the same and the total number of core threads, called “sibblings”, is twice as large. That is, in both cases we have 2 threads per core and each of those appears to the operating system as a virtual CPU”…
That was not my point. The point was task manager does not show an i7 as fake cores either. So suck one
No. Each cores shared the front-end and the FPU with the other core in the module. You continue without understand the concept of conjoined cores and you are confounding the cores with the “integer clusters”. The integer cluster is what is highlighted in green in that slide.
In any case, as mentioned before it doesn’t matter what you believe is or is not a core. What matters is what think the guy is demanding AMD and he disagrees with you and with you definition of core.
That was part of the plaintiff’s complaint in the lawsuit https://regmedia.co.uk/2015/11/06/amd_bulldozer_lawsuit.pdf quote “In its marketing, AMD represents that each module contains two cores, but that is not the case because a Bulldozer module begins as a single core, to which AMD adds some—but not all—of the components from another core. As described above, a core is a processing unit (what once was a single CPU) that is independent from other processing units on the same physical chip or die. AMD’s decision to provide each module with only some (but not all) of the components of two cores means a module contains only one complete core, not two as advertised” and quote “But as Figure 11 reveals, AMD designed its module processing units to share common components. As such, AMD’s advertised “cores” are not independent from each other and are not really cores”
That is only part of his confusion. He doesn’t understand that conjoined cores share parts with adjacent cores. Moreover he doesn’t understand how those conjoinced cores on Piledriver work. He claims that its FX-9590 cannot do eight operations simultaneously, but that is wrong, because its FX-9590 can do up to 32 operations per cycle.
Again, those AMD FX series CPU “cores” are not “conjoined cores”, they never were in the first place. As mentioned elsewhere, the plaintiff is not well versed in CPU technology thus unaware about the superscalar nature of today’s modern CPUs.
Meaning this one?
Nope not that one. It looked like it as far as the green heatsink goes, I think that one may have been manufactured after the one I had, Which may be why I don’t see the one I had anywhere. With the lawsuits going on around that time, I may have one that was pulled from the market. I will have to go dig it up out of the storage shed. I have never thrown away any of my computer parts so I know I still have it out there. lol
Mind you: the first Intel 486DX was introduced in 1989, that’s 5 years earlier than 1994 (Intel 80486DX-25, slow as s**t)
Old topic but the point was at least in AMD’s CPU it actually has the hardware of sorts to actually have support for those 8 core listed. We will call them Full core & mini core because of the shared resources but the hardware is there. In Intels case you have 4 hardware cores & 4 logical core non hardware which yes they can handle 8 threads but it is what it is 4 hardware & 4 logical for the desktop i7 CPU. Also depends on the version or service release of Windows 8.1 on how it shows the AMD’s core setup. I own a Computer store & have worked on a lot of AMD systems & in some it shows the APU as 2/4 & in other systems or in the FX’s case 4/8 or 8/8 depending on the service update on the system. I am not sure how Windows 10 displays the AMD CPU setup I pretty much stopped looking a long time ago as in not caring. This is coming from someone who pretty much only owns Intel based systems so I am not a fanboy type for AMD.