AMD Agrees To Pay Out $35 Per Chip Over FX Marketing Lawsuit

[TechyBen]

6 minutes ago, leadeater said:

I know they can, I'm saying it would be detrimental to the industry to do so. I thought I made that clear. It wouldn't just apply to Bulldozer, a now defunct irrelevant product but everything. Let a standards body take over that roll, not the court. Then if someone works around the accepted standard you can then go legal. There is no current standardized industry definition of a CPU core and you can get experts in the industry to argue many different ways, seek a formalized consensus that can be amended over time which is something you cannot do if you do it via case law.

How? Legally, it happens all the time.

How is it detrimental to the industry to have 1 definition of a core (not retrospective, it would only apply to future products/designs), and only for the marketing/package. How would that be detrimental?

 

Say we add a multi connected 3d layered CPU system, that shares Cache and also HT throughput. The new "definition" means these shared resources are not true single cores. Intel (the ones closes to 3d stacking AFAIK other than IBM or whatever), would then have 2 options, market it as:

1) "12 core plus 12!!!"

2) "12 Coreplusses"

 

Or even better names. Such as "Quads" for 4 way cores, or "dublers" for 2 way limited cores...

 

How is this worse than, the market/engineers automatically going "Oh, nice register you have there, oh, nice float... why don't we stop saying "buy your registers and cache here!" and just instead name it "buy your core here"... I mean, industry and engineers *alreadys renames things when they changed*.

 

PS AMD is already doing "chiplets" and "CCX" to show not all cores (or connections between them) are created equal.

[leadeater]

3 minutes ago, TechyBen said:

How? Legally, it happens all the time.

How is it detrimental to the industry to have 1 definition of a core (not retrospective, it would only apply to future products/designs), and only for the marketing/package. How would that be detrimental?

Because it would limit the industry in how it may want to evolve CPU architecture, you know it has undergone massive changes over time. How do you propose you change the legal definition of a CPU core over time to allow technology to progress? By going to the courts each time to argue your case in a legal setting every time?

 

Why do something that you know will legally cause you issues because it does not fit the current legal definition of a CPU core?

[Rainbow Dash]

50 minutes ago, ChintzyPC said:

Quick question:
Does anyone know if a purchase from Newegg still counts? I've got proof of purchase through Newegg for an FX 8350. But considering that's not through AMD's website and I've never lived in Cali I'm wondering that it might not count.

The claims only work if you live in California when you purchased one, or if you bought one from AMD's website directly which does not matter where you live.

[Guest]

8 minutes ago, Rainbow Dash said:

The claims only work if you live in California when you purchased one or from AMD's website directly.

Imo, that's pretty ridiculous. Newegg uses the exact same advertising as is on AMD's website so there's no reason to think Newegg, Amazon, etc customers are somehow differently informed. Besides, there's enough evidence through Newegg purchase history that indicates it's a completely valid purchase. This should be opened up to anyone who's purchased it through a valid enough seller (of course not ebay, second-hand, etc though). Although, that would probably mean less money spread out for those who qualify.

Once the site is released I'm going to try it anyhow just in case. lol

[Jito463]

4 hours ago, TechyBen said:

AMD did not lose because slower. They did not lose because different. They lost because it was different, but the box said "same and more", when it was "different and possibly, in some instances, more".

I'll just quote myself here.

On 8/29/2019 at 12:50 AM, Jito463 said:

They didn't technically lose.  They settled out of court so they could get rid of the lawsuit, and not risk the judge siding against them in an even bigger judgment (or even just shelling out more money for lawyers over a longer period, only to end up at the same conclusion).

A settlement is meant to make the lawsuit go away without admitting any fault.

4 hours ago, TechyBen said:

AMD did not state "do more" or "you can perform faster", they stated "it has 8 cores".

And it does.  Can all 8 cores perform a task simultaneously?  Yes.  Not in all tasks, but in certain ones it can.  INT and 128-bit FP, for example.  Ergo, it has 8 cores.

 

On that note, I think about all that can be said on the topic has been said (or written, as the case may be).  Given that we're all just going around in circles, I think I'll bow out now, unless someone brings up a particularly interesting rebuttal.

[TechyBen]

10 hours ago, leadeater said:

Because it would limit the industry in how it may want to evolve CPU architecture, you know it has undergone massive changes over time. How do you propose you change the legal definition of a CPU core over time to allow technology to progress? By going to the courts each time to argue your case in a legal setting every time?

 

Why do something that you know will legally cause you issues because it does not fit the current legal definition of a CPU core?

Wait... No. How does a *name* limit the industry? I gave existing examples and future possibilities. How does a *name* prevent you from doing whatever you wish to silicon? PS, such as "This has 12 gigacores". Like, industries do this all the time. Cars have changed engines, they don't call them "6 cylinder" if they invent a 3 cylinder turbo/super charger that does better mpg, but has less cylinders, because "oh noes, we have a legal definition of cores... I mean cylinders, and our engineers are scared they get sued if they combine them to make a wankel* engine."... right? The example you give is fantasy, because industry has definitions.

 

6 hours ago, Jito463 said:

I'll just quote myself here.

A settlement is meant to make the lawsuit go away without admitting any fault.

And it does.  Can all 8 cores perform a task simultaneously?  Yes.  Not in all tasks, but in certain ones it can.  INT and 128-bit FP, for example.  Ergo, it has 8 cores.

 

On that note, I think about all that can be said on the topic has been said (or written, as the case may be).  Given that we're all just going around in circles, I think I'll bow out now, unless someone brings up a particularly interesting rebuttal.

Oh, I agree it's out of court. But it depends on what a "core" is. Industry can do what it like. Computer sciences/engineering can do what they like. Advertising/packaging is regulated on what they name "things". "Core" is a thing named in marketing, thus it can be regulated/defined.

 

I agree it's only the advertising/packaging/use cases being decided here, not necessarily an industry wide case. But everything else in our language and commerce has definitions. I see no problem with having one here (fish/beef... car/bicycle... Kilogram/liter etc).

 

*  https://en.wikipedia.org/wiki/Wankel_engine

[And regulated via tax, not advertising in this case: https://en.wikipedia.org/wiki/Engine_displacement ]

[leadeater]

35 minutes ago, TechyBen said:

Wait... No. How does a *name* limit the industry? I gave existing examples and future possibilities. How does a *name* prevent you from doing whatever you wish to silicon? PS, such as "This has 12 gigacores". Like, industries do this all the time. Cars have changed engines, they don't call them "6 cylinder" if they invent a 3 cylinder turbo/super charger that does better mpg, but has less cylinders, because "oh noes, we have a legal definition of cores... I mean cylinders, and our engineers are scared they get sued if they combine them to make a wankel* engine."... right? The example you give is fantasy, because industry has definitions.

Because every operating system and all software have the same concept around CPU cores and scale to the number presented. The entire industry works around a common point and that is the CPU core, you can't just market some random number then actually present to the OS a different number, or do you expect all the operating systems to be re-coded just to suit these marketing names and each time they change might I add.

 

I'm thinking you are one of those stuck on the fact that CPU cores need to have an FPU, well a CPU core has a hell of a lot more in there than an FPU. Heck I bet none of this would have been a problem if Bulldozer was a dual FPU single INT CMT design and nobody here would be batting an eye about calling it a proper 8 core CPU.

 

There is nothing wrong with externalizing an FPU out of the CPU core, Intel already does it for the second AVX-512 unit, and equally nothing wrong with sharing it either. Different doesn't have to be bad. Heck you could even have a single shared FPU between 8 cores and that FPU could be akin to a GPU in the way it works and be twice as fast as individual FPUs in each core, but no can't do that now it's a single core CPU or I have to market it as such and present it to the OS as that.

 

The car example just doesn't work here.

 

With that I'm not responding further. If you can't think about future problems or don't care then that's on you, I expect as vast changes to come as there has in the past and I expect no out of date case laws to needlessly inhibit that.

[TechyBen]

1 minute ago, leadeater said:

Because every operating system and all software have the same concept around CPU cores and scale to the number presented. The entire industry works around a common point and that is the CPU core, you can't just market some random number then actually present to the OS as different number, or do you expect all the operating systems to be re-coded just to suit these marketing names and each time tho change might I add.

[Bolded mine] But yes, people do. RISCV =/= IBM compatible. ARM 4+4 or 2+2 cores =/= AMD APU units. "Core" is subjective to the product. AMD had cores, described them one way, had the white paper/specs one place, but the packaging made no additional note of it? They released it in an existing industry/market, not a new one.

 

Quote

I'm thinking you are one of those stuck on the fact that CPU cores need to have an FPU, well a CPU cores has a hell of a lot more in there than an FPU. Heck I bet none of this would have been a problem with Bulldozer was a a dual FPU single INT CMT design and nobody here would be batting an eye about calling it a proper 8 core CPU.

 

Nope. They could be made of chocolate, and glow in the dark. But the packaging/advertising would legally need to define or represent that to sell in a legally managed marketplace.

 

Quote

There is nothing wrong with externalizing an FPU out of the CPU core, Intel already does it for the second AVX-512 unit, and equally nothing wrong with sharing it either. Different doesn't have to be bad. Heck you could even have a single shared FPU between 8 cores and that FPU could be akin to a GPU in the way it works and be twice as fast as individual FPUs in each core, but no can't do that now it's a single core CPU or I have to market it as such and present it to the OS as that.

 

With that I'm not respond further. If you can't think about future problems or don't care then that's on you, I expect as vast changes to come as there has in the past and I expect no out of date case laws to needless inhibit that.

 

But Intel does not advertise/label AVX as "more cores for the 4 core i5" etc. It's a "4 core i5 *with* AVX", not "516 core i5". That would be redefining without informing the consumer.

 

" Different doesn't have to be bad." No one said it was. They said it was not defined/different to defined.

 

Quote

Heck you could even have a single shared FPU between 8 cores and that FPU could be akin to a GPU in the way it works and be twice as fast as individual FPUs in each core, but no can't do that now it's a single core CPU or I have to market it as such and present it to the OS as that.

 

We could have a full on AMD RISKV processor... They could release an FPGA. But would an FPGA be "hundreds of cores!!!"? It would. But would it "speed up Windows"? Could I be sly and say "speed up all your workflows"? How far wiggle room should I be allowed before having to admit to the consumer "but Windows will only use 1 of those, and it will have a slow wait chain while it sets data flow"?

 

Quote

The car example just doesn't work here.

 

Why? Like, wow. Why shut down an example, with no comment? We already have definitions for many industries. Pixels in displays, they now have subpixels. VR now uses multiple displays. New tech breaks and changes those definitions. Existing tech, needs to meet them or change.

 

Advertising vs consumers vs legal definitions. These things exist. See the cars as an example (though as said, mainly for taxation purposes, and only in part for advertising).

Advertising can and does have different definitions to industry and engineers. They also can have different definitions to consumers. Language is fluid like this. Legal settings are not though.

 

Consumers and AMD did not reach an agreement, so it went to court. Then AMD did reach an (out of court) agreement.

 

Some things are noted on the packaging "cache" and "TDP" and "boost" etc. If floating point/integer calcs were not noted on the packaging, then this may get a consumer expected change (put it on packaging, but not legally required) or a legal setting (legally state those two things). Cores/integer/float were not legally required, as previously, there was an expectation from the industry and consumers.

 

An example of this, is the 32 bit/64 bit code/hardware. Or MS vs Mac code. While these are open and advertised, people will not generally back a consumer saying "but I thought my Windows PC could run Mac software", but may back a consumer saying "But I thought my 8 core AMD (of performance X) did twice an Intel 4 core (of the same performance X)?" A 6 cylinder engine is not expected to run twice the speed, or have twice the fuel consumption. But a 6 seater is expected to fit 6 people in it. Having a 6 seater with the other 3 seats only fitting children and dogs is a poor move, though totally possible from an engineering perspective.  We do have some cars with 4/5 seats + 2 mini seats in the back, they are small, but people fit in them. Those with kids only seats, are advertised as such that.

[Dabombinable]

27 minutes ago, leadeater said:

Because every operating system and all software have the same concept around CPU cores and scale to the number presented. The entire industry works around a common point and that is the CPU core, you can't just market some random number then actually present to the OS a different number, or do you expect all the operating systems to be re-coded just to suit these marketing names and each time they change might I add.

 

I'm thinking you are one of those stuck on the fact that CPU cores need to have an FPU, well a CPU core has a hell of a lot more in there than an FPU. Heck I bet none of this would have been a problem if Bulldozer was a dual FPU single INT CMT design and nobody here would be batting an eye about calling it a proper 8 core CPU.

 

There is nothing wrong with externalizing an FPU out of the CPU core, Intel already does it for the second AVX-512 unit, and equally nothing wrong with sharing it either. Different doesn't have to be bad. Heck you could even have a single shared FPU between 8 cores and that FPU could be akin to a GPU in the way it works and be twice as fast as individual FPUs in each core, but no can't do that now it's a single core CPU or I have to market it as such and present it to the OS as that.

 

The car example just doesn't work here.

 

With that I'm not responding further. If you can't think about future problems or don't care then that's on you, I expect as vast changes to come as there has in the past and I expect no out of date case laws to needlessly inhibit that.

If I can find the screen shots, I'll show you how my Phenom II P920 and N970 performed comapred to the A8 4555M that sat between them. That includes multi threaded scaling, which was quite terrible. I can easily see why people feel mislead over AMD's marketing over the Bulldozer series and its derivatives. Because the majority of tasks saw the CPU scale like shit, with well under 3x the single threaded performance.
Unlike with the Phenom II x4, which managed anywhere between 3.8x-3.9x scaling.

[leadeater]

40 minutes ago, TechyBen said:

Why? Like, wow. Why shut down an example, with no comment?

Because it doesn't and you ignored what I pointed to or didn't understand it. Was AMD supposed to market Bulldozer as '8 AMD cores' in your eyes and present it to the OS as 4? Yea I bet that won't turn in to a lawsuit at all. Or market it as 4 cores and present it to the OS as 8? Same again. If you start messing around with CPU cores and calling them what ever you wish at what ever number you like you're still going to need to make that work under the current model of how things work without entirely breaking all software optimization.

 

Microsoft SQL Server relies on the number of CPU cores presented to the OS, it uses this to optimize the database engine and allocate queries to CPU resources. Making up what ever you want is not an option, and this is just a single example.

 

I as a buyer of hardware need to know the number of cores, not AMD cores, not Intel cores, not the number of FP or INT units, but the number of x86_64 cores that will be presented to me inside the OS. How these presented hardware cores/threads actually work behind them doesn't matter, everything could be shared it makes no difference at all. You could then claim that this CPU is actually only a single core but as long as it can do 8 concurrent operations and performs the required tasks it thus needs to be presented to the OS as 8 cores which necessitates that it is advertised as 8 cores, not some other thing.

 

Marketing does in fact impact what the engineers can do, or will do. Restrictions imposed here flow through. That is also why as far as the case goes around the marketing matters, because I agree you cannot just say how many cores you have but in reality do not. There's a difference between a poor architecture design that performs poorly and not actually having the number of stated cores though.

 

I hope that better explains the situation to you. This will actually be my last reply.

 

30 minutes ago, Dabombinable said:

If I can find the screen shots, I'll show you how my Phenom II P920 and N970 performed comapred to the A8 4555M that sat between them. That includes multi threaded scaling, which was quite terrible. I can easily see why people feel mislead over AMD's marketing over the Bulldozer series and its derivatives. Because the majority of tasks saw the CPU scale like shit, with well under 3x the single threaded performance.
Unlike with the Phenom II x4, which managed anywhere between 3.8x-3.9x scaling.

For sure, it was a horrible architecture but bad performance doesn't actually mean the number of cores didn't exist or weren't legitimate. In my eyes it's far better to just call bulldozer and complete failure, like it was, and move on. It's not as if lessons weren't learned by AMD, the goal is always to make a good product and it's near impossible to hide if it is or not

[BiG StroOnZ]

2 hours ago, Dabombinable said:

I can easily see why people feel mislead over AMD's marketing over the Bulldozer series and its derivatives. Because the majority of tasks saw the CPU scale like shit...

 

Just some random old benchmarks I found in the archive (Although this is gaming performance, I still think it paints a great picture about what a lot of people here are trying to portray and explain; especially when paying close attention to the other Intel chips (even older AMD) that are performing around the same (or better) as these Bulldozer/Piledriver chips in these benchmarks):

 

Spoiler

 

 

[The King of the Undead]

it is a distinct possability that the problem was that it was the same brand but the cores were built differently and thus performance changed but didn't necessarily increase.

[Dabombinable]

8 hours ago, BiG StroOnZ said:

 

Just some random old benchmarks I found in the archive (Although this is gaming performance, I still think it paints a great picture about what a lot of people here are trying to portray and explain; especially when paying close attention to the other Intel chips (even older AMD) that are performing around the same (or better) as these Bulldozer/Piledriver chips in these benchmarks):

 

  Hide contents

 

 

I was thinking of more a 1:1 comparison between mobile quad cores all at around the same clock speed (1.6, 1.8GHz and 2.2GHz).

[Dabombinable]

9 hours ago, leadeater said:

For sure, it was a horrible architecture but bad performance doesn't actually mean the number of cores didn't exist or weren't legitimate. In my eyes it's far better to just call bulldozer and complete failure, like it was, and move on. It's not as if lessons weren't learned by AMD, the goal is always to make a good product and it's near impossible to hide if it is or not

When the multi threaded scaling is closer to a hyperthreaded i7 instead of 8 actual cores (eg dual cpu Xeon setup),. its misleading to say that there are 8 cores. AMD should have kept on marketing them as having 4 modules instead of later on changing that to cores. Because even their own previous generations of CPU had better multi threaded scaling (as in 4x cores= 3.86x single threaded scaling, not 2.6x single threaded scaling).

 

 

[leadeater]

37 minutes ago, Dabombinable said:

When the multi threaded scaling is closer to a hyperthreaded i7 instead of 8 actual cores (eg dual cpu Xeon setup),. its misleading to say that there are 8 cores. AMD should have kept on marketing them as having 4 modules instead of later on changing that to cores. Because even their own previous generations of CPU had better multi threaded scaling (as in 4x cores= 3.86x single threaded scaling, not 2.6x single threaded scaling).

I still don't think that means the number of cores should be advertised as something different, you don't advertise Atom CPUs are having less cores because they are really slow. Clarity over the product and how it performs is far more important, which is were I think the majority of the problem lies. Thing is I don't exactly expect AMD to have come out and said "Hey our new thing is actually pretty crap" either, they should but no company would.

 

If the shared FPU had 2 256bit execution units instead of 2 128bit execution units the performance wouldn't have been nearly as bad, AMD's assumption like many on this forum about AVX2 is that most things are not optimized for it but not optimized for and not being used were very different things. Intel's compiler is actually very common and will implement AVX2 when it can, even if it's not the best optimization of it so the 2 FP execution units will largely have to combine to run that 256bit operation.

 

CMT has aspects of both dual core and SMT, it's really neither but like I mentioned I don't think that's actually the problem, set expectations were.

[mr moose]

@leadeater we have been over that too many times,  the reality is I don't this sort of case is even slightly a problem for the industry.  The courts can lay down many definitions for cores and marketers will just use those terms instead.

 

As far as the rest is concerned,  misleading advertising happens all time, sometimes we agree with the claims and sometimes we don't.  I can see the arguments in this one.  The language is sufficient to warrant a case as I don't see the technicalities of the CPU is the problem.  

 

[mr moose]

18 hours ago, leadeater said:

Marketing does in fact impact what the engineers can do, or will do. Restrictions imposed here flow through.

I think it would be a pretty shit engineer who couldn't look at the block diagram/spec sheet of a CPU and decide how many cores it had according to their own definition/requirements and instead relied on marketing when designing a product.

 

 

[leadeater]

1 hour ago, mr moose said:

As far as the rest is concerned,  misleading advertising happens all time, sometimes we agree with the claims and sometimes we don't.  I can see the arguments in this one.  The language is sufficient to warrant a case as I don't see the technicalities of the CPU is the problem.

Oh I'm not surprised a suit happened, like Bulldozer was legit that terrible. There's enough scope to go in to it without bothering about if a CMT was a single core or two, but it does look like a stronger case to answer and more likely to be allowed to go ahead with that in it. I'm just super not ok with it being in there.

 

1 hour ago, mr moose said:

I think it would be a pretty shit engineer who couldn't look at the block diagram/spec sheet of a CPU and decide how many cores it had according to their own definition/requirements and instead relied on marketing when designing a product.

That has nothing to do with my pointed out issue. Have a look a block diagram of 486, have a look at the Pentium 4, have a look at a Core 2 Duo and Core 2 Quad, have look at say Sandybridge, and Skylake-SP. Then also consider the growing changes now like Zen, CCXs and chiplets. The block diagram of today is a lot different to the past and will be so in the future. If you rule that in effect a CPU core must have a FPU, per core, or conversely you cannot call something a core if it does not have a FPU then legally you are bound to your architecture having a FPU inside the CPU core.

 

Now as the example I gave before of having a large high performance single FPU that can run many many operations of many bit sizes but still have a front end per core that has L1 instruction cache , TLB, Decoders that feeds in to the execution engine that as either a single scheduler or two (one for the off board FPU and another for everything else) that puts tasks either in to the core local execution units or puts it out to the external FPU. After this is the standard L1 data cache and L2 cache in the core, how the FPU feeds back in (L2 cache or L3) is not important to the illustrated point here. Can such an architecture like this with 8 cores be called 8 cores? When does it not become a core, what things must be there or are not allowed to be shared?

 

You can't just decide how many cores you think it has, that makes no difference. You think the OS cares or your application cares how many you think it has? Nope. How the presented cores perform is entirely another matter but the number of them matters a lot when it comes to application optimization and behavior. Would do me zero good to have a CPU show up under ESXi as a single core because legally that is what it must be to not be sued due to existing case law. A single core in ESXi is literally useless and won't at all work with how VMs are allocated resources, CPU time. All you'd get is massive contention and no performance at all even if this single core was the worlds fastest CPU, faster than 128 Zen2 cores.

 

Edit:

Any FYI if you want to change my mind you're going to have to convince me that significant architecture changes, in more than 5 year scale will not be impacted by some old ruling that someone may wish to dredge up just to be an ass. Unless you can very confidently say cores will look exactly as they do today in 5, 10, 15 years time then I'll be fine with some court ruling that if it lacks an FPU it is not a core, even though I think that is not the place to be determining that.

[TechyBen]

[edit] Never mind.

[Edit]

Ah, that's better. Found my place.

 

21 hours ago, leadeater said:

Because it doesn't and you ignored what I pointed to or didn't understand it. Was AMD supposed to market Bulldozer as '8 AMD cores' in your eyes and present it to the OS as 4? Yea I bet that won't turn in to a lawsuit at all. Or market it as 4 cores and present it to the OS as 8? Same again. If you start messing around with CPU cores and calling them what ever you wish at what ever number you like you're still going to need to make that work under the current model of how things work without entirely breaking all software optimization.

 

Microsoft SQL Server relies on the number of CPU cores presented to the OS, it uses this to optimize the database engine and allocate queries to CPU resources. Making up what ever you want is not an option, and this is just a single example.

 

I as a buyer of hardware need to know the number of cores, not AMD cores, not Intel cores, not the number of FP or INT units, but the number of x86_64 cores that will be presented to me inside the OS. How these presented hardware cores/threads actually work behind them doesn't matter, everything could be shared it makes no difference at all. You could then claim that this CPU is actually only a single core but as long as it can do 8 concurrent operations and performs the required tasks it thus needs to be presented to the OS as 8 cores which necessitates that it is advertised as 8 cores, not some other thing.

 

Marketing does in fact impact what the engineers can do, or will do. Restrictions imposed here flow through. That is also why as far as the case goes around the marketing matters, because I agree you cannot just say how many cores you have but in reality do not. There's a difference between a poor architecture design that performs poorly and not actually having the number of stated cores though.

 

I hope that better explains the situation to you. This will actually be my last reply.

 

For sure, it was a horrible architecture but bad performance doesn't actually mean the number of cores didn't exist or weren't legitimate. In my eyes it's far better to just call bulldozer and complete failure, like it was, and move on. It's not as if lessons weren't learned by AMD, the goal is always to make a good product and it's near impossible to hide if it is or not

The bold bit. The bit in BOLD.

THAT's what AMD did. They made their own "AMD cores" which were confusing to the consumer. They made a definition of core that did not match the OS definition. That is fine, if branded as "AMD core", but they were branded as "pure" cores, which was misleading to consumers as to description.

 

As said, no court case over Intel Atom cores, ARM cores, RISCV cores... but AMD... AMD pushed too far on some advertising.

[mr moose]

1 hour ago, leadeater said:

Oh I'm not surprised a suit happened, like Bulldozer was legit that terrible. There's enough scope to go in to it without bothering about if a CMT was a single core or two, but it does look like a stronger case to answer and more likely to be allowed to go ahead with that in it. I'm just super not ok with it being in there.

 

That has nothing to do with my pointed out issue. Have a look a block diagram of 486, have a look at the Pentium 4, have a look at a Core 2 Duo and Core 2 Quad, have look at say Sandybridge, and Skylake-SP. Then also consider the growing changes now like Zen, CCXs and chiplets. The block diagram of today is a lot different to the past and will be so in the future. If you rule that in effect a CPU core must have a FPU, per core, or conversely you cannot call something a core if it does not have a FPU then legally you are bound to your architecture having a FPU inside the CPU core.

 

Now as the example I gave before of having a large high performance single FPU that can run many many operations of many bit sizes but still have a front end per core that has L1 instruction cache , TLB, Decoders that feeds in to the execution engine that as either a single scheduler or two (one for the off board FPU and another for everything else) that puts tasks either in to the core local execution units or puts it out to the external FPU. After this is the standard L1 data cache and L2 cache in the core, how the FPU feeds back in (L2 cache or L3) is not important to the illustrated point here. Can such an architecture like this with 8 cores be called 8 cores? When does it not become a core, what things must be there or are not allowed to be shared?

 

You can't just decide how many cores you think it has, that makes no difference. You think the OS cares or your application cares how many you think it has? Nope. How the presented cores perform is entirely another matter but the number of them matters a lot when it comes to application optimization and behavior. Would do me zero good to have a CPU show up under ESXi as a single core because legally that is what it must be to not be sued due to existing case law. A single core in ESXi is literally useless and won't at all work with how VMs are allocated resources, CPU time. All you'd get is massive contention and no performance at all even if this single core was the worlds fastest CPU, faster than 128 Zen2 cores.

 

Edit:

Any FYI if you want to change my mind you're going to have to convince me that significant architecture changes, in more than 5 year scale will not be impacted by some old ruling that someone may wish to dredge up just to be an ass. Unless you can very confidently say cores will look exactly as they do today in 5, 10, 15 years time then I'll be fine with some court ruling that if it lacks an FPU it is not a core, even though I think that is not the place to be determining that.

O.K, I don't think this case was ever going to result in a precedent being set that would ban anyone from using the word core,  all it would do is hold up the already existing precedent of not making claims about quantity being directly related to performance when that is not the case.

 

I certainly cannot see a court ruling defining exactly what are core is as having any effect on CPU design or advertising.  At worse it just means companies will have to put in a foot note qualifying why their definition of a core is different to the legal definition. 

 

Besides all that, when technology advances beyond legal boundaries/definitions the technology just seems to exists without legal  repercussion.  It never seems to get hindered by old laws.

[TechyBen]

I think Leadeater misses how *consumer* markets work. You don't get to redefine industries for consumers. For enterprise? Fine. But for consumers, rules of historical expectations exist.

 

If for example, we had 100% saturation of CPUs having integrated GPUs for the last 10 years. Then AMD release a "new" CPU, without a GPU, and advertise it as "the pure CPU core, do everything faster", then the courts would rule it as misleading, because it forgot to state, "everything except the 1 thing we removed, the iGPU".

 

That happened here. "Do everything faster" and forgot "except the floats we removed", if they said "go faster*" and then "* 90% of tasks faster", etc... It was not that it was "poor", it was that it was advertised as an upgrade. Intel Atoms are not advertised as "faster, more pure cores to upgrade your Skylake series!" 

 

[leadeater]

47 minutes ago, mr moose said:

O.K, I don't think this case was ever going to result in a precedent being set that would ban anyone from using the word core

I didn't say anything about not being able to use the word core. I said specifically what is legally considered a core as to the number you can say you have, if legally something is a single core then you can't advertise it as 8, make it show as 8 to the OS or anything like that. I'm sure you can see the problem is saying something has one number but yet it is actually another when you go to use it.

 

Making up your own numbers isn't an option, it isn't and why you can't see that is baffling. Windows and Linux alike treats cores and threads a certain way, show it 8 and it will act accordingly. Why wouldn't anyone object and not sue if the number of cores seen by the OS is not the same as advertised. Like the accusation with Bulldzoer.

 

Saying anything like "AMD cores" is redundant, it's already there on the box, the brand name AMD. 

 

A foot not means nothing, totally nothing. When all operating systems and all software have zero sense or caring for this. The foot note is Bulldozer architecture, look it up if you want to know how the core is structured.

 

47 minutes ago, mr moose said:

Besides all that, when technology advances beyond legal boundaries/definitions the technology just seems to exists without legal  repercussion.  It never seems to get hindered by old laws.

Yet yourself have more than once pointed to the law not keeping up with technology in other topics. You'll have to forgive me but this doesn't hold true even by your own words.

[TechyBen]

16 minutes ago, leadeater said:

I didn't say anything about not being able to use the word core. I said specifically what is legally considered a core as to the number you can say you have, if legally something is a single core then you can't advertise it as 8, make it show as 8 to the OS or anything like that. I'm sure you can see the problem is saying something has one number but yet it is actually another when you go to use it.

 

Making up your own numbers isn't an option, it isn't and why you can't see that is baffling. Windows and Linux alike treats cores and threads a certain way, show it 8 and it will act accordingly. Why wouldn't anyone object and not sue if the number of cores seen by the OS is not the same as advertised. Like the accusation with Bulldzoer.

 

Saying anything like "AMD cores" is redundant, it's already there on the box, the brand name AMD. 

 

A foot not means nothing, totally nothing. When all operating systems and all software have zero sense or caring for this. The foot note is Bulldozer architecture, look it up if you want to know how the core is structured.

 

Yet yourself have more than once pointed to the law not keeping up with technology in other topics. You'll have to forgive me but this doesn't hold true even by your own words.

It is mundanely easy to show 100000s of cores to the OS, but all these be virtual. So AMD can release a "million core" CPU, but as long as the OS sees them, it's not false advertising?

 

Advertising =/= legal definitions of objects. (Again, see car manufacturing/mileage/fuel consumption etc)

 

No, it's not redundant. AMD make X86 compatible CPUs. If the CPU uses some other method, then that would be a problem. If AMD use some *other* metric to an *industry standard*, then that's a problem. If the new AMD CPU was an ARM CPU with x86 (and x64) in virtualisation, then they would get *in trouble*.

[leadeater]

1 hour ago, TechyBen said:

I think Leadeater misses how *consumer* markets work. You don't get to redefine industries for consumers. For enterprise? Fine. But for consumers, rules of historical expectations exist.

No and it isn't any more ok for enterprise. Please actually actually consider what I have said, not just ignore it or not bother to try and understand it. No matter what you keep saying the case put forward was directly challenging (specifically in the court filing with expert witnesses) what a core was,  going through with this would set case law, that is a fact. You might want to argue against me about how this would effect the industry and we might disagree, we already do, but it would be irrefutable fact that case law would have been set.

 

The most likely way for it to not become an issue is to not making a ruling on it, zero chance of effect.

 

This is why I said I wouldn't reply to you anymore, because you didn't show any sign of reading what I wrote or attempt to understand it and came back with totally wild nonsensical replies that were just taking too long to decypher.

 

CPU vendors using their own numbers is not an option when nothing else uses it or would.

 

1 hour ago, TechyBen said:

That happened here. "Do everything faster" and forgot "except the floats we removed", if they said "go faster*" and then "* 90% of tasks faster", etc... It was not that it was "poor", it was that it was advertised as an upgrade.

Ahh actually they did, this is why I think you're not reading my posts at all. 

 

On 8/31/2019 at 10:52 AM, leadeater said:

And it was very clearly stated on AMD's website that the FP scheduler was shared and could either perform a single large operation or two, as I highlighted earlier in the screenshoot. This information was stated, failure to read is not deception. So even a lay person could figure out that if two could work together as one then 8 / 2 = 4.

Or operating separately with each core. Hmm if it can do 1 per core and there is 8 that must mean it can do 8, but what about this teaming together part? Guess that means half of 8?

Every retailer that had the AMD marketing page embedded on the product page had this information on it, most did.

[TechyBen]

5 minutes ago, leadeater said:

No matter what you keep saying the case put forward was directly challenging (specifically in the court filing with expert witnesses) what a core was,  going through with this would set case law, that is a fact. You might want to argue against me about how this would effect the industry and we might disagree, we already do, but it would be irrefutable fact that case law would have been set.

Yes! And it did not succeed! AMD backed down. We are not saying it should be court decided. But it can be. If AMD don't want it to be, they need open documentation even *on the box*.

 

If consumers and markets get confused, then this happens. Thus AMD were not completely open, if they were, confusion would not be there. If the market had not already set a standard (through historical content of "cores" and "OS") then consumers would not expect the performance of the *floats* to equal core count.

 

A "layman" does not need to know how many float/integer performance, but they *can* see it was lower performance, thus not the "new/same/better". If AMD released "power efficient AMD ATOM Cores!" they would not be sued, because those are not advertised as "better/faster" but "better/powerefficent" instead.

 

Quote

Ahh actually they did, this is why I think you're not reading my posts at all. 

On the box, next to the "faster!!! Better!!!"???

Quote

most did.

Ah, not 100%? Also, courts don't care what the markets advertised... it's what AMD advertised that was in court. Hence the disagreement here, we are talking past the points being made.