AMD to Cough Up $12.1 Million to Settle "Bulldozer" Core Count Class-Action Lawsuit

[Vya Domus]

Again, if you want to classify this as conspiracy theory, I too see no point in continuing this.

 

[FordGT90Concept]

AMD's Bulldozer architecture has, at the very least 8 independent cores/processing units.

Execution cores/integer clusters, not multiprocessor cores. "Core," since 2005, has only referred to the latter. FX-8350 is a quad-core processor with eight execution units. This is a statement of demonstrable fact that AMD agreed to in settling the case.

Again, if you want to classify this as conspiracy theory, I too see no point in continuing this.

The only "conspiracy" is that AMD tried to oversell their product and they admitted they did this by settling.


There's literally nothing left to debate: it's settled. "Core" in the context of computer technology, henceforth means "multiprocessor core." By that very basic test, FX-8350 is a quad-core.

 

[Vya Domus]

Nope, core is any independent processing unit. There is no context really, not like this anyway.

That's why GPU manufacturers say their GPUs have thousands of cores because well, those are one type of core even though they are very different from CPU cores.

That's why even manufacturers that make AI chips say they have x amount of cores. Those are cores too even though they look nothing like CPU cores.

 

[FordGT90Concept]

NVIDIA calls them "Tensor Cores" and "CUDA Cores," not simply "cores" because "multiprocessor cores" they're not. AMD calls them "stream processors;" they don't use "core" nomenclature at all on GPU products.

I'd have to see specific AI chip models you're referring to in order to comment.

 

[Vya Domus]

This is irrelevant, the meaning or the use of the term core is not contextual unless you want to talk about a specific type. You don't want to let go of the "multiprocessor core" thing even though this is never brought into question here. We are talking about the standalone term that is "core". That's it, don't attach anything to it.

In this regard anyone that makes a chip which sports multiple processing units is free to call them cores, of what kind it's their business but they are cores nonetheless.

 

[FordGT90Concept]

False, AMD can't call an "execution core" a "core" in marketing again or the FTC will string them up for false advertising. $12.1 million fine would be a trifle compared to FTC's damages.

 

[Vya Domus]

If an execution core is not a core ... then what is it ? A waffle ?

Stop attaching attributes to this term. "Core" should work as a standalone term and it does, a lot of people use it for different types of chips.

There are CPU cores, GPU cores, DSP cores, etc. And none of them look similar and they sure as hell don't have to comply to a certain type of description.

An independent processing unit is a core.

 

[FordGT90Concept]

Exclusively when it refers to a "multiprocessor core." Any other use of "core" must be predicated by the type, examples: "CUDA core," "tensor core," "execution core." In the context of CPUs, it means one thing and one thing alone: a complete processor. This box only contained four complete processors:

 

[Vya Domus]

Exclusively when it refers to a "multiprocessor core."

This is a requirement made up by yourself.

A more recent example on how liberal is the use of this term :

https://www.cerebras.net/wp-content/uploads/2019/08/Cerebras-Wafer-Scale-Engine-Whitepaper.pdf

Technology - Cerebras

This is plastered with the word core unbound to any type because it doesn't have to be.

This box only contained four complete processors:

It contained one processor with 8 cores.

 

[FordGT90Concept]

That's another technical document. We're talking about commercial products like that in the image I gave above. Class action lawsuits are about public harm.

That said, technical documents should be clear in their definitions of "core" as well. Your referenced document does this by specifying "compute cores." The very next sentence it stresses FMAC:

Accelerating calculation is most directly achieved by increasing the number of compute cores. More cores—specifically more floating point multiply accumulate units-- do more calculations in less time.

Every use of "core" in that document after that point has an established context: FMAC-heavy compute core.

This isn't rocket science. Giving context to "core" is akin to giving proper units in algebra. If it is without context (like computing products on store shelves), it applies to "multiprocessor cores." Everyone should be on the same page now; there should be no further confusion from vague use of "core."

It contained one processor with 8 cores.

One socketable package with four processors. The industry cleaned that mess up back in 2005 too when multiprocessor no longer implied multisocket. Processors had to be divorced from their physical attachment in software.

 

[Vya Domus]

That's another technical document. We're talking about commercial products like that in the image I gave above.

I am sorry but no, there is no reason to believe this term behaves differently if it's written on a webpage, in a paper or on a box. It's a word that bears the same meaning, all the time.

You are gain moving the goal post, one time is about multiprocessor cores, the next about complete processors and now you insist to convince me that this debate should only be analyzed in a certain context.

There is no way I'll ever agree to any other definition for a core other than an individual processing unit, irrespective of the make of the chip that it's part of.

This goes against every sane assumption that people made about chips for ages. The world isn't full of only Intel and AMD CPUs, there's a lot more out there.

 

[FordGT90Concept]

There is no way I'll ever agree to any other definition for a core other than an individual processing unit, irrespective of the make of the chip that it's part of.

Then there's no reason to continue this; the matter is settled.

 

[Vya Domus]

One socketable package with four processors.

If I ever meet anyone IRL that shows me a single die processor and tells me with a straight face that this is "four processors", I'll feel obliged to slap them.

 

[Adam Krazispeed]

wheres my money??? iv got several Buldozer and up cpus?? where my $35 bucks each??? wtf?

any way there are 4x dual core modules?? did have 8 cores, each pair was 2 cores sharing resources, wtf? people are lame when they dont know what CMT is, its two cpu cores with shared cache, fpu etc..

it was just a flopped design that did suck..cant sue for a bad design, wheres intels lawsuits over only giving us 4 core cpus for 500+ dollars for over a decade

 

[FordGT90Concept]

If I ever meet anyone IRL that shows me a single die processor and tells me with a straight face that this is "four processors", I'll feel obliged to slap them.

Query NUMBER_OF_PROCESSORS environmental variable in Windows. It returns the number of logical processors which is the number of threads the underlying hardware accepts.

This is a relic dating back to at least Windows 2000 where multiprocessors only existed in multisockets. Microsoft had to shape how the operating system sees processors over time in order to best schedule workloads. If you're writing software and you want to know how many threads are required to saturate the hardware, NUMBER_OF_PROCESSORS is as correct today as it was two decades ago; however, Windows (and other operating systems) are now more knowledgeable about the hardware they run on as a function of necessity.

So let's dig up the first thread on this subject here:

AMD Dragged to Court over Core Count on "Bulldozer"

This had to happen eventually. AMD has been dragged to court over misrepresentation of its CPU core count in its "Bulldozer" architecture. Tony Dickey, representing himself in the U.S. District Court for the Northern District of California, accused AMD of falsely advertising the core count in...

www.techpowerup.com

Specifically this post:

"Kerner" is Danish for "core." Would you look at that? FX-9590 is a quad-core according to Microsoft! But look at what AMD named it at the top: "eight-core." False advertising much?

The operating system has to look at the cores when load balancing more so than the logical processors because multiple threads on a single core share resources. This is a fact of CMT and SMT; CMT just shares less.

 

[GreiverBlade]

One socketable package with four processors. The industry cleaned that mess up back in 2005 too when multiprocessor no longer implied multisocket. Processors had to be divorced from their physical attachment in software.

nope actually one processor with 4 modules containing 2 core and 1 FP/Scheduler each ... and it still stay like that : multiprocessor is still a 2P 4P 8P etc system and a multicore processor is still a single socket unit having 2x 4x 6x 8x 10x 12x cores

My SuperMicro H8DCE with 2 Opteron 270 is a 2P system with 2 cores in each Processor

"Kerner" is Danish for "core." Would you look at that? FX-9590 is a quad-core according to Microsoft! But look at what AMD named it at the top: "eight-core." False advertising much?

The operating system has to look at the cores when load balancing more so than the logical processors because multiple threads on a single core share resources. This is a fact of CMT and SMT; CMT just shares less.

well in one revison of W7 my FX6300 shown as 3 cores and later after some patch it did shows as 6 cores ... as if Microsoft was a reference, well WMI was a reference for long
also Logical core aren't Physical and my FX6300 had 6 physical cores ... windows only probably look at the number of scheduler and considere a pair of core and 1 FP scheduler as "1 physical core"

also ... conjoined-core hehehe well that would mean ... 2 core together to if AMD did put "4 conjoined-cores" that would mean ... well ... 8 cores.

They've sucked you into this debate and they will never let you leave now.

well that's why got some mean for killing time during my day off


aaaaaaaaannddd now i am off ... i need to check my notification settings, seriously xD

 

[Vya Domus]

FX-9590 is a quad-core according to Microsoft!

The same Windows which didn't know what to do TR, Zen and Zen 2 basally every AMD CPU for the past couple of years ? I'll take the liberty of assuming they didn't know what to do with this one either. Did they fix the bug with TR yet that made the OS stack the same threads onto the same cores all the time ?

If you seriously consider Windows as a reliable source of information regarding what it does with the cores of the CPU it means you're getting desperate in your search for proof.

This is a relic dating back to at least Windows 2000 where multiprocessors only existed in multisockets.

You called it, it's a relic, not indicative of modern architectures. Don't use Windows for evidence in this case OK ?

A processor is a single die of silicon which may contain multiple cores and threads.

 

[FordGT90Concept]

My SuperMicro H8DCE with 2 Opteron 270 is a 2P system with 2 cores in each Processor

It's equally accurate to call it four processors. How they're arranged is an implementation detail. That's why Microsoft elected to show "sockets" instead of "processors." Have an example:

Calling it a "processor" instead of a socket is too vague.

also ... conjoined-core hehehe well that would mean ... 2 core together to if AMD did put "4 conjoined-cores" that would mean ... well ... 8 cores.

A "conjoined-core" is a "core" which may contain two or more "execution cores."

You called it, it's a relic, not indicative of modern architectures. Don't use Windows for evidence in this case OK ?

I've already presented a mountain of evidence spanning three threads now. Windows confirming it is just icing on the cake at this point.

 

[Vya Domus]

Windows confirming

It doesn't confirm anything, just that it's a inaccurate way of counting the cores of a CPU.

There are tons of other software that recognizes this CPU as having the corect amount of cores but you pick out just one. Not exactly a mountain of evidence, more like a small tiny boulder.

 

[FordGT90Concept]

Oh, it's totally accurate. It just proves you wrong, so you disagree.

 

[Vya Domus]

It just proves you wrong, so you disagree.

In order to prove someone wrong you need arguments and proof. One single case of software mislabeling a piece of hardware is anything but proof.

But then again you said that people were writing wrong definitions to hide AMD's lies and that implies it was all a conspiracy theory to prove AMD right. Oh, and let's not forget the processor with 4 processors, that one was also pretty good. So I expect you to use pretty much everything as proof at this point.

Also, didn't you said you were done like 2-3 times by now ?

 

[FordGT90Concept]

...that implies it was all a conspiracy theory to prove AMD right.

No, it doesn't. AMD threw it out there and people gobbled it up on their own without questioning the rationale for it. The conspiracy is solely in AMD's original lie ("8-core").

Also, didn't you said you were done like 2-3 times by now ?

Sure did. Sadly, I'm easily baited.

 

[Vya Domus]

people gobbled it up on their own without questioning the rationale for it.

Right, and then they proceeded to write books and papers about this one wrong thing. Got it, makes sense. Doesn't sound like a conspiracy at all.

 

[GreiverBlade]

Calling it a "processor" instead of a socket is too vague.

well a socket host .... a processor ... which in turn host one or multiple core (rather a socket is a .... oh, socket where the processor is ... uh, socketed? isn't it so? )
and obviously yes M$ would call it socket ... 2p/4p/8p system have 2/4/8 socket and Windows reference that, that's all.
although they could wrote CPU instead of socket ... but they wouldn't use a synonym of "processor", instead of the interface where the CPU/Processor is connected to the mobo, for that ... that would be confusing, right?

A "conjoined-core" is a "core" which may contain two or more "execution cores."

which are, as you write it, cores...

 

[FordGT90Concept]

well a socket host .... a processor ... which in turn host one or multiple core (rather a socket is a .... oh, socket where the processor is ... uh, socketed? isn't it so? )

It's a chip housing one or more processors. Each processor on said chip may be called a core.

Since the advent of chip-multiprocessing: one or more socket/land grid array -> central processing unit -> one or more physical processors (cores)-> one or more logical processors

which are, as you write it, cores...

[facepalm.jpg]
A "execution core" understands this:

a += b

A "core" understands this:

if (c > 0) a += b

One is math, the other processing. One is a calculator, the other is a processor.

A "core" relies on nothing other than memory subsystems to carry out instructions. A "core" will call on various "execution cores" to carry out instructions by delegating tasks to them. For example, floating point math gets set to the floating point unit (a type of "execution core") while integer math will get set to the arithmetic logic unit (another type of "execution core"). In the examples given above, if "a" is of float type, it will be routed to FPUs, if it is of int32 type, it will be routed to ALUs.

All of the parts combined make a processor and a processor, when packaged with many on one chip, is collectively called a "core." "Core," without context in computing, is synonymous with "processor."