AMD on suicide watch
Source: http://legalnewsline.com/stories/510646458-amd-faces-suit-over-alleged-misrepresentation-of-new-cpu
Stupid American law system.
It's an 8core architecture.
>wah this is not how I define cores ~Dickey
>>51243569
Since Intel can afford to buy the definition if core, Amd is done.
>>51243569
>It's an 8core architecture.
It's not tho.
Like AMD has the money to fight lawsuits like this.
This isn't a real claim, it's designed to get AMD to waste money on a lawsuit.
>>51243592
I don't know, man. I'm seeing 4 modules with 8 cores.
>>51243593
It was ze joos?
>>51243706
>shared FPU
>shared L2 cache
Yeah, no. It's 4 cores.
>>51243736
Is a 386SX not a CPU core then?
It doesn't have an FPU period...
Or L2 cache.
More to the point, this is the actual argument:
>AMD’s Bulldozers suffer from material performance degradation and cannot perform eight instructions simultaneously and independently as claimed.
You can indeed run 8, independent threads simultaneously.
>>51243526
poo
o
o
>>51243712
>>51243712
No it's not intel. Intel doesn't want AMD to die. They want AMD to be alive but locked up in a little chastity cage, they don't want them gone.
This is most likely an opportunist law firm.
>>51243736
Nah, the schematics clearly show eight cores.
Everything else is irrelevant.
>>51243810
>You can indeed run 8, independent threads simultaneously.
If that's the case why don't they just boot up an amd computer and prove it?
>>51243872
They probably will.
There hasn't even been preliminary hearings yet, it's simply filed...
>>51243872
Because that's not how lawsuits happen.
>>51243810
>You can indeed run 8, independent threads simultaneously.
Not 8 threads that rely entirely on floating point calculations.
>>51243883
Lawsuits are pretty fucked up then, if they can't just provide the evidence and walk away. This is going to look bad for amd.
Fuck dickey.
Fuck him in the dick.
This is the last thing they needed just before zen.
>>51243904
Yes, yes you can.
'Shared' FPU does 2x 128bit (or 2x 32bit or 2x64bit or 2x80bit) calculations simultaneously or if put into a special mode, will disable the other int core in the module and you can have a 256bit fmac.
The reality of the FPU is it's essentially just the same idea as paired singles used in some PPC cpus (it' just paired quads instead)
The real thing actually being shared is fetch/decode/scheduling, which is fucking stupid and slow - but the lawsuit as it stands, doesn't have a leg to stand on.
>>51243907
They can, they'll get to make their arguments after the litigators make their case.
After they present their evidence, they can also file a motion for dismissal
>Or even better, dismissal with prejudice.
>>51243907
If you have a sure thing, you don't want to win the case right away. You want to draw it out then win it and make the other side pay both legal fees.
Laying down some base facts for all you tech illiterate retards, and Tony Dickey himself since he was shitposting here for hours yesterday:
There is no such thing as a standard/true/real X86 core. There are only varying hardware implementations of the ISA. Anything and everything is an X86 core so long as it complies with and can process the instruction sets defined therein.
FPUs are not a necessary unit in an X86 core, and for many years CPUs didn't have them at all. FPUs came about as add in accelerator cards before companies began incorporating their own on die.
AMD's Bulldozer and derivative family of chips have "real" cores as much as any other architecture
The 4 module Bulldozer and Piledriver chips can fetch, decode, execute, and retire 8 ops simultaneously, including FPU ops. The 4 modules can handle 8 128bit FPU ops. The only thing they can't do is handle fused 256bit ops in the same manner.
Each core in the module is independently addressable.
l.
>>51244019
The performance issues with AMD's Bulldozer family do not stem from the module design itself, there are three broad strokes responsible:
1. Front end misses and stalls.
The moderate long pipeline and frequent branch misses meant that 20~ cycle stalls happened frequently in the front end. This stalls the entire pipeline and robs performance of the arch consistently. You can't get ops to the ALUs at all when the front end is stalled.
2.) Cache latency and bandwidth
Ops that run through the cache hierarchy suffer the performance penalty directly induced by how long it takes them to get written back/ written through the cache. AMD's cache is considerably slower than intel's, though this isn't entirely their fault by design as cache performance is heavily influenced by the libraries offered by the foundry.
3.) GloFo's 32nm PD-SOI node massively missed target frequencies and displayed horrendous leakage.
Bulldozer was originally supposed to target 4ghz+ operation. By design it was meant to be a narrow core to save die area, and clock high to make up for the lower serial performance. In theory this is sound for multi threaded workloads, as 2 sets of 2 ALUs have higher utilization than a single set of 4 ALUs. For compute performance per mm2 its great if everything else were to workout. Though the front end issues combined with the current leakage and low stock clocks of the Bulldozer family chips significantly crippled their potential.
>>51244008
If amd drags this out they're smearing they're own name.
Normies will think zen is the same after hearing about this.
I had a customer refuse to buy a micro usb cable because they thought the made in China cables kill people. It's the wall plugs that was the cause of death, but people will be people.
They will think a zen quad core is dual core, even though zen is dropping cmt and going with smt.
>>51244042
>normies
>buy AMD
>care about corporate court cases
pick none
>>51244031
>>51244019
That out of the way, we can see that Tony Dickey's complaint is based off of a false premise. He asserts a few things which are empirically false, and they are central to his claims of being mislead and incurring damages.
His claims have no validity whatsoever, and thats readily apparent. The case will get thrown out at the first hearing.
Now enjoy the rest of the ensuing troll thread.
>>51244055
It's all about reputation.
Why do you think Intel is so popular among normies?
If you want amd to release good products, they can't afford to lose popularity and sales.
>>51244088
>If you want amd to release good products, they can't afford to lose popularity and sales.
This. If they don't make any money how can they be expected to release good products?
It's because of that, that we went from the beastly 7970 and 290x to the fury x...
>>51244118
Each of those gpus is 40-50% faster than the last one - exactly in line with Nvidia's lineup.
>>51243526
do you stupid americans sue everything you can't comprehend?
>>51243526
Cool, does this mean i can sue NoVideo over 4gb misrepresentation?
>>51244288
Some people, like this scum fuck in question, make a living or supplement their income by constantly filing suits against companies.
Its ridiculous how eager people are to use the legal system as an ATM.
>>51244309
No - the 970 does have 4gb of vram on it. Nvidia lying about rops and L2$ however could get them sued.
>>51244019
Then why do the FX chips see ~50-60% scaling going from single thread to multiple when every other architecture ever sees in excess of 80%?
Why did Windows task scheduler get a patch to assign workloads to not spread across multiple modules to address the performance issues?
FX chips are 4M/8T like most software reads them as
>>51245087
>Then why do the FX chips see ~50-60% scaling going from single thread to multiple
Actual scaling is nowhere near this low, unless you were fallaciously comparing to intel core scaling in raw performance. Per core performance is lower so thats to be expected. Running Linpack on an FX 4300 vs an FX 8350 with normalized clocks shows ballpark figures you'd expect to see in any well threaded workload.
Multicore scaling and coherency is largely affected by cache. Even Kaveri had patently lower multicore scaling than Llano because even the Steamroller cores passed data among themselves slower than the old K10.5 derivative Husky cores.
One of the main performance improvements to the Steamroller design over Piledriver was multi core scaling, though they unfortunately made a trade off of cache latency for increased bandwidth.
>Why did Windows task scheduler get a patch
The Windows scheduler patch saw performance gains of 1-3%~ and it was primarily aimed at stopping the OS from over loading the front end of certain modules while ignoring the others. Thats entirely an issue with Microsoft's software, not AMD's architecture.
>FX chips are 4M/8T like most software reads them as
Software only reads things how its written to. This is a total nonstatement.
>>51243736
Each module has 2x128bit FPUs in it. The two 128bit FPUs can be combined into one 256bit FPU for AVX work.
>>51243526
I knew this was gonna come up:
>>51243810
>Is a 386SX not a CPU core then?
You are thinking of the 486SX (the 386DX didn't have the internal x87 FPU either), but that's fine.
And no, you would not get away with releasing a 386-style x86 CPU without an x87 FPU today, unless clearly stating this in the specifications. Consumers would otherwise be misled, when in fact most of their x86 software would not even run, let alone perform as expected. This is to protect consumers, not just geeks who read up on the architecture.
Now, Bulldozer is even an x64 CPU, and x64 CPUs do include x87 FPUs. Releasing an FPU-less Bulldozer was never an option for AMD. A consumer can't be expected to anticipate that these allegedly 8-core x64 CPUs will scale considerably worse with floating point workloads than integer when going from 4 to 8 threads.
This is not about determining who is technically correct, it is about protecting the consumer. We're all ignorant in some field of life, and this is how we try to protect each other.
Like the rest of you, I don't want to see AMD bled dry of resources that would be better spent on R&D either. I just think the plaintiff has a case.
>>51244031
>Bulldozer was originally supposed to target 4ghz+ operation. By design it was meant to be a narrow core to save die area, and clock high to make up for the lower serial performance.
iow AMD set out to create their own Prescott. brilliant
>>51245810
It doesn't matter how you get there, so long as you actually get there without using a ton of power.
If the FX 8350 had 15% higher IPC, and a 95w TDP it would have been a solid chip to compete against Ivy Bridge. It wouldn't win out in serial workloads, but the multithreaded advantage would have been really significant.
Looking at some of the highest binned FX 8370e chips shows what clocks could have been reached if the foundry delivered.
Stable 4ghz clock, 1.17v, and 90watts measured at the rail.
If they delivered this on launch then at least they would have had serious overclocking headroom before power consumption exploded.
>>51244019
De facto standards are a thing.
This is not a game of mental gymnastics and being technically correct. Besides, show me a precedent x64 core without an FPU.
>tech illiterate retards
Yes, the very same the laws are meant to protect.
>The only thing they can't do is handle fused 256bit ops in the same manner.
Do you agree float workloads don't scale as well as int workloads beyond 4 threads because of this?
>>51246000
>De facto standards are a thing.
No they are not, especially not when it comes to ICs and instruction set support. Vendors routinely create their own supplementary instructions set. One vendor can offer entirely different tertiary instructions, while their competition offers their own internally created instruction sets.
One vendor can offer hardware for handling legacy code, while the competition handles it through emulation.
One vendor can offer media accelerators, strong RNG, crypto accelerators, and etc while the competition does not.
There are no standards whatsoever apart from the ISA itself. Everything not included in the ISA is supplementary and not standard.
>Yes, the very same the laws are meant to protect.
Consumer protection laws are in place for the average person, not the absolutely retarded.
People purchasing sports cars know that they can check a publication like car and driver for reviews.
People purchasing computer hardware know that they can check a site like Anandtech, TomsHardware, Techreport, or any number of other sites for reviews. Their web traffic alone proves they are well known among consumers. Mr Dickey claims to have purchased two FX Centurion chips well into 2015. If he managed to find these special binnings of AMD's Vishera line and had literally never heard of them before then he is quite simply a liar.
The fact that he was purchasing components off of Newegg to build a system objectively proves he is hardware savvy, and would have known their performance characteristics before electing to purchase them.
>Do you agree float workloads don't scale as well as int workloads beyond 4 threads because of this?
The only 256bit ops one would be using are AVX, and they are by no means common. They're exceedingly niche as a matter of fact.
You're trying to weasel around the facts of the matter because you can't form a consistent logical argument.
The guy filing the suit is a liar and a shitty conman.
>>51245988
Those e chips are fucking magical compared to their non-e counterparts and you are absolutely right, had they been released way back when they would've been a solid competitor. Though even now for the money they are damn tasty chips.
>>51246156
>still not getting it
The consumer is not expected to understand the intricacies of the x86 ISA and its extension.
Where is the precedent x64 core without an FPU?
>the average person
According to you /g/ is tech illiterate retards. You probably don't want to meet the average person... (And it would be anyone within a couple of sigma of the average.)
>The only 256bit ops one would be using are AVX, and they are by no means common. They're exceedingly niche as a matter of fact.
A simple "yes" would've sufficed.
Whether it's common enough is up for the courts to decide.
>You
Your problem is that you keep looking at this through the spectacles of an engineer. Regrettably, most people are NOT engineers.
TL;DR: Bulldozer scales worse because AMD equipped it with fewer FPUs than marketed cores.
>>51246652
>The consumer is not expected to understand the intricacies of the x86 ISA and its extension.
And they don't have to. They don't have to know a single thing about a core, exactly as a consumer doesn't have to understand a single thing about a motor for auto manufacturers to advertise the specs of them freely.
You are not making an argument, you are still just lying through your teeth constantly.
>A simple "yes" would've sufficed.
There you go weaseling around facts again. Not including a 512bit FPU means that intel's Skylake is at a disadvantage when it comes to handling 512bit AVX. Think anyone is going to try and sue them for not printing this in all of their marketing material? Of course not, because its entirely unreasonable, exactly as this case is.
>TL;DR: Bulldozer scales worse because AMD equipped it with fewer FPUs than marketed cores.
No, this is only relevant to strict FPU ops and literally nothing else.
Tony Dickey is a liar and a conman, and I'm near 100% certain that you are him.
The average consumer does not build their own PC.
The average consumer certainly does not purchase enthusiast level hardware to build their own PC.
These things are relegated to the enthusiast market, which is a small niche of the market to begin with.
Tony Dickey, you, actively went out of your way to pick TWO FX Centurion chips, compatible mobos, heat sinks, and adequate power supplies. You having enough knowledge to seek out enthusiast level hardware want people to believe you did so in utter ignorance, and that thousands of others did the same, all without looking up a single review ever? The Vishera processors came out in 2013. You bought two of them this very year.
You're a liar through and through. Your case has absolutely no merit whatsoever, and its going to get tossed out instantly without ever going to trial. All you do is lie and shitpost, time and time again.
>>51245686
Show me examples of situations where the typical consumer actually needs 256-bit float capability.
Because my experience shows that we still use 128 for most of everything. AVX is a niche case and isn't really a consumer thing. All it really does, in the eyes of common functionality, is technically make a CPU able to cram two 128 ops into one cycle.
Which is what the "flex fpu" of the bulldozer design does.
Where bulldozer fails is the instruction decoder, scheduler, and branching units.
In one aspect they aren't fast enough and/or wide enough, in another aspect there should have been two sets for each module (this is how they saved space over Phenom designs).
The proof of this is that if you turn off the other integer unit of a module, throughput of the first can increase as much as 30%
But please, please. Keep telling me more about what you know.
>>51244063
>The case will get thrown out at the first hearing.
No it won't, since he's claiming false advertising, and he's right.
It doesn't matter if it's technically 8 cores. They're not comparable to Intel's cores, and AMD DID compare their core count to Intel in their ads. You can't compare apples to oranges implying it's apples to apples. AMD misled by not disclosing that their chip could not be judged on a core-count basis for all workloads because each core had only half the FPU resources, and they misled by not disclosing that their integer cores could only manage half Intel's throughput.
If they had just described their own chip it would've been fine. But they advertised it compared to Intel, and you can't do that without coming clean on the relevant differences.
"B..but the data's on every review site!" That means nothing. False advertising is a charge about the advertising. Doesn't matter if AMD was giving tours and handing out informational brochures about the entire design process -- if their advertising was false and misleading, it's false and misleading.
>>51247179
>8 x86 cores
Nothing there mentions FPUs
>>51247179
>making the exact same fallacious nonsense post you did yesterday.
Give it up, you pathological liar.
A core is a core, thats all there is to it. Intel is not the authority on what constitutes a core, and their architecture is in no way the defacto standard that all other X86 cores must fall in line with. In fact, their implementation of SMT is different to all others seen throughout the industry.
You make wild baseless assertions, and constantly pretend that they are truthful and factual. They are not. You are lying over and over.
It is not a technicality that AMD's 4 module Bulldozer/Piledriver chips have 8 cores. They have 8 cores. Thats all there is to it. They have 8 cores, and in multithreaded workloads they perform better than intel's quad core i5s, and even their higher binned i7s in a number of instance.
Your case is built upon a false premise, you assert a position that is factually incorrect in every single possible way it could be interpreted. Trying to contort things to fit your twisted narrative because you're a mentally ill lying retard. The marketing is not deceptive, or misleading. You were utterly blown the fuck out yesterday for making the same argument, and the same thing is going to happen today.
The case will be immediately thrown out, Tony Dickey. You're a fucking nutcase.
>>51244310
kek, AMD employee detected
>>51246156
Who wants to bet this is a disgruntled AMD employee who probably has his job on the line after making such a shitty CPU
>>51247316
The firm I work for is based out of North Dakota, and it sure as hell doesn't come with an AMD grade paycheck or level of job security.
Putting one lying conman in his place doesn't mean I'm defending the years and years of AMD's management shitting their pants and wasting valuable R&D funds.
>>51247179
This is why I buy intel only.
>>51243526
wow, and they havent done jack shit to 3.5, housefire fermi or lapfire Shield?
man, i guess when you take appledrone...i mean nvidiadrone money you can pay off a justice department.
>>51247362
>muh firm
Why is everyone on 4chan a lawyer or working at Google?
>>51247374
Firm does not imply law firm.
Thoufh fun fact: two EE students who used to post here got jobs at Freescale and a TSMC office. Going to college pays off when you enter the right field.
>>51243589
This.
>>51247267
>A core is a core, thats all there is to it
False advertising is false advertising. That's all there is to it.
>>51247262
Here you go. Without the "x86."
>>51247937
What kind of cores do you expect them to be, dumshit?
There is nothing misleading about that. Nothing even remotely implies that their cores have 1:1 performance with intel's Sandy Bridge.
>>51247966
It's not a kindergarten exercise in labeling and counting things, it's an advertisement. Its entire purpose is to give you reasons to buy.
"You should buy the FX 8150 over the i5 because it has 4 more cores," rather implies that those 4 extra cores make a difference in a metric that the consumer cares about, which in this case would be performance. Yet they do not. Those "4 more cores," only just gets it just up to i5 levels in heavily multithreaded tasks.
The FX does not have 4 more cores than Intel, it has 8 shitty cores to Intel's fantastic 4.
> AMD FX(tm)-8350 Eight-Core Processor
> Cores: 4
Microsoft ain't buying AMD's load of bullshit, either.
INTEL
MASTER
RACE
.
>>51248280
>which in this case would be performance. Yet they do not.
There you go lying again. Why is it that you cannot be honest even for a moment no matter what? Consistently with every single post you make you lie to try and get your utterly insane points across.
It does not work. People here are smarter than you are.
The 4 module FX chips have 8 cores. Those 8 cores give a solid performance advantage in multithreaded workloads. There is no obligation anywhere to state that one core architecture is different from another. Again, just as there is no obligation anywhere for auto manufacturers to state that their V8 differs from another competitors V8. No reasonable and prudent person would assume they are the same.
The architecture used in intel's Atoms is not the same as the arch used in their mainstream CPUs.
The arch used in AMD's Kabini/Beema is not the same as the arch used in Thuban and Deneb.
The arch used in VIA's latest quadcores is not the same as the arch used in anything AMD or intel make.
The same is true of all the ARM SoC vendors who create their own cores.
These fact do not need to be explicitly advertised. There is literally not a single thing that is misleading about this. You were intimately aware of all of this, Tony. You know that your case is fraudulent. You're a lying nutjob trying to get money just like every other shameless patent troll out there.
>The FX does not have 4 more cores than Intel
Yes it does. This is an objective fact that cannot be argued.
> it has 8 shitty cores to Intel's fantastic 4.
"I'm a petulant child who can't make a real argument." - You, Tony Dickey the pathological liar and shitposter extraordinaire.
>>51248280
>it has 8 shitty cores to Intel's fantastic 4
And this is where the case falls apart. There's no specific definition of a "core" that AMD cites in their advertisement. It's how every game gets away with labeling their game a "GAME OF THE YEAR." It's hardly deceptive, unless you're retarded.
>>51248513
>state that their V8 differs from another competitors V8
This is bullshit. If you look from V8 from the side, you can clearly see only 4 cylinders. V8 are 4 cylinder engines, and this is false advertisement.
Only i8 engines can be called true 8 cylinder.
>>51248670
>gaymur fags
>retarded
Pick one get the second absolutely free
>>51248513
>The architecture used in intel's Atoms is not the same as the arch used in their mainstream CPUs.
And they don't advertise the Atom against the i5 on core count, now do they?
>The arch used in VIA's latest quadcores is not the same as the arch used in anything AMD or intel make.
And they're not advertised against them on core count, now are they?
>new CPU
>fucking bulldozer
For fuck sake guys.
>>51248670
>There's no specific definition of a "core" that AMD cites in their advertisement.
They don't have to make one. The comparison to Intel on nothing but core count makes the claim they're comparable. It was their legal responsibility to make clear how that was not the case.
You can't just leave that hanging there. They did.
>>51248734
Yet again, intel's arch in the Core2Duo chips was nothing like AMD's Athlon 64 X2.
As a matter of fact when the C2D chips were released people were saying they weren't "real" dual core chips because they shared resources.
Guess what? No one sued intel over this.
There is literally no requirement for marketing material to explicitly state "our core is different from that other core."
You're so mentally ill I'm starting to feel bad for entertaining you for so long.
You're getting your laughably flawed case thrown out immediately, Tony.
>>51248842
>The comparison to Intel on nothing but core count makes the claim they're comparable
They are comparable because they're cores. Putting core counts side by side implies absolutely nothing whatsoever. AMD, or any company whatsoever, is not responsible for your deranged assumptions.
>>51248513
I'm sorry you fell for AMD's false advertising and I realize your fragile ego will do anything to avoid the conclusion that you've been duped, but this isn't about your post-purchase rationalization. It's about whether AMD specifically advertised their cores as though they were comparable to Intel. It's not about whether a dozen Raspberry Pis have 12 cores, it's not about whether your GPU has 1500, it's about whether AMD compared their CPU to Sandy Bridge on core count as though they were comparable.
The FX has half the FPU resources per core and half the integer performance. You can't compare them as though it's apples to apples.
>>51248899
>There is literally no requirement for marketing material to explicitly state "our core is different from that other core."
There is when you're making a comparison along those lines.
>>51249046
I do not own any Bulldozer family CPU, you shitposting conman.
>>51249092
No there isn't. Just like there isn't any need for an auto manufacturer to list specifics on bore, stroke, or compression of their engines. They can list their i4/V6/V8 whatever directly against their competitor's. There is literally no legal precedent whatsoever which obligates them to list something that a reasonable and prudent person would assume is different.
The core of your complaint is based on a false premise.
The 8 cores in a 4 module chip are independently addressable, they can each execute operations simultaneously. This right here disproves every single thing about your case. You filed a suit against a company and asserted something that is patently false. Its empirically wrong, can be validated in mere seconds.
Everything else you post is just constant lies and spin, contorting reality. You're not going to convince any judge that your mentally ill and factually deficient ramblings hold water by constantly lying on 4chan.
It's amazing how butthurt you faggots are getting over this. Who gives a fuck.
>>51249390
a lot of people
>>51248449
This is their front end which is also shared per module, they had lots of scheduling problems due to this, also compiler code generation issues.
See: http://www.anandtech.com/show/5448/the-bulldozer-scheduling-patch-tested
Overall this is about the definition of "core", I can see the case for a loose definition being harmful to the average consumer, but then this is a industry wide problem. Particularly ARM and their big.little scheme, which is wise, but an average consumer will likely misunderstand.
>>51243526
op, your amd logo was stretched so i fixed it
>>51250965
>Overall this is about the definition of "core",
No, this is only about AMD misleading customers by comparing their core count to Intel.
