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

[Vya Domus]

None of that has anything to do with your claim that cores can be defined as such.

The instructions set isn't tied to the hardware, instructions can't define hardware, they can't tell you what is a core or CPU and what isn't. Any turing complete computer can be made to carry out any kind of instruction no matter how simple or complex, people have figured this out a 100 years ago.

Point me to any instance in a any book or article that says something along the line of "this thing has a MOV instruction therefore it's a core", whenever this is brought up it's done so from a pure hardware perspective. As far as I am concerned you are literally making all of this up.

 

[R-T-B]

Ridiculous.

1) Meritless suit.

2) Only covers California purchases.

3) Doesn't even cover all of the 8 core consumer-grade FX chips.

Agreed on all but part 1. I used to say it was meritless, but the definition of what a processor core is has changed since the 90s, and AMD is paying for that. Honestly, they should. I'm just sad so little of it goes to consumers.

 

[londiste]

Any turing complete computer can be made to carry out any kind of instruction no matter how simple or complex, people have figured this out a 100 years ago.

True. But at the same time turing completeness has no restrictions on how complex it is to program said things or time it takes for the computer to perform these instructions.

 

[R-T-B]

When "Bulldozer" was released how many PC CPUs (sold as new) had less FPU cores count compared to integer cores?

A lot of arm cores back then did. Devil's advocate, not sure we want to call early arm cores "performance."

 

[ManofGod]

Bulldozer and it's iterations trough out the years had modules with two pipelines in them, so technically they did sell people said amount of processing cores albeit some of the front end and FP was shared.
Guess you can view these modules as double wide INT pipelines and call these 4C/8T instead of 8C/8T, but that would not be completely true now would it?

Did AMD lie about the architecture? Not really. Block diagrams show whats what. G
Did AMD lie about core counts? Bit of grey area depending on how one views the architecture as a whole.
Was Bulldozer uarch good? Hell no.
Does this entitle anyone to get money back cause they had chosen poorly even after the reviews? Not really at least in my opinion.

But seemingly you can sue back money if you band together with other people who had bought into FX even after reading trough the benchmarks, not once but even after years of Excator and Piledriver reviews.
'Murica at work right there.

The Bulldozer uarch was good, absolutely. However, they never fixed the weaknesses in the uarch, like too slow cache and other such stuff and instead, moved on. Probably best for them, since they never produced a new chipset beyond the 990FX.

 

[londiste]

Where are these definitions ? Can you link books, papers anything ?

http://inspirit.net.in/books/academic/Computer%20Organisation%20and%20Architecture%208e%20by%20William%20Stallings.pdf

Just a few excerpts. I would definitely recommend reading more, starting with part 1.

central processing unit (CPU): That portion of a computer that fetches and executes instructions. It consists of an Arithmetic and Logic Unit (ALU), a control unit, and registers. Often simply referred to as a processor

Each of these components will be examined in some detail in Part Two. However, for our purposes, the most interesting and in some ways the most complex component is the CPU. Its major structural components are as follows:
• Control unit: Controls the operation of the CPU and hence the computer
• Arithmetic and logic unit (ALU): Performs the computer’s data processing functions
• Registers: Provides storage internal to the CPU
• CPU interconnection: Some mechanism that provides for communication among the control unit, ALU, and registers

The processor does the actual work by executing instructions specified in the program. This section provides an overview of the key elements of program execution. In its simplest form, instruction processing consists of two steps: The processor reads (fetches) instructions from memory one at a time and executes each instruction. Program execution consists of repeating the process of instruction fetch and instruction execution. The instruction execution may involve several operations and depends on the nature of the instruction (see, for example, the lower portion of Figure 2.4).

A multicore computer, also known as a chip multiprocessor, combines two or more processors (called cores) on a single piece of silicon (called a die). Typically, each core consists of all of the components of an independent processor, such as registers, ALU, pipeline hardware, and control unit, plus L1 instruction and data caches. In addition to the multiple cores, contemporary multicore chips also include L2 cache and, in some cases, L3 cache.

The architectural organization of cores comes primarily from the multiprocessor systems (type of which a multicore system is). With Zen and talking about memory/core access latency, NUMA is a term to keep in mind.

The term SMP refers to a computer hardware architecture and also to the operating system behavior that reflects that architecture. An SMP can be defined as a standalone computer system with the following characteristics:
1. There are two or more similar processors of comparable capability.
2. These processors share the same main memory and I/O facilities and are interconnected by a bus or other internal connection scheme, such that memory access time is approximately the same for each processor.
3. All processors share access to I/O devices, either through the same channels or through different channels that provide paths to the same device.
4. All processors can perform the same functions (hence the term symmetric).
5. The system is controlled by an integrated operating system that provides interaction between processors and their programs at the job, task, file, and data element levels.

Nonuniform memory access (NUMA):
All processors have access to all parts of main memory using loads and stores. The memory access time of a processor differs depending on which region of main memory is accessed. The last statement is true for all processors; however, for different processors, which memory regions are slower and which are faster differ.

 

[neatfeatguy]

About every 6 months one of these posts shows up about Bulldozer....

 

[Vya Domus]

http://inspirit.net.in/books/academic/Computer%20Organisation%20and%20Architecture%208e%20by%20William%20Stallings.pdf

Just a few excerpts. I would definitely recommend reading more, starting with part 1.





The architectural organization of cores comes primarily from the multiprocessor systems (type of which a multicore system is). With Zen and talking about memory/core access latency, NUMA is a term to keep in mind.

Thanks but I looked really carefully throughout all that and not once was a core described by it's ability to execute instructions. It's always done so under the more generic "processor" label and it's not clear at all how you go from processor to core, are these two really interchangeable and equivalent ? The author doesn't think so :

Processor: A physical piece of silicon containing one or more cores. The processor is the computer component that interprets and executes instructions. If a processor contains multiple cores, it is referred to as a multicore processor.

So the attribute of executing instructions is given to the higher level abstraction that is the processor. I also find it funny you ignored the really interesting definition that was just below the one about the CPU:

■ Core: An individual processing unit on a processor chip. A core may be equivalent in functionality to a CPU on a single-CPU system. Other specialized processing units, such as one optimized for vector and matrix operations, are also referred to as cores.

"May be equivalent", not a requirement. Also the authors reckons even SIMD like executions units can be called cores, all of these things match AMD's claims. This is getting very murky, this endeavor to prove AMD's cores aren't cores sees no light at the end of the tunnel no matter where you look.

 

[londiste]

"May be equivalent" also the authors reckons even SIMD like executions units can be called cores, all of these things match AMD's claims. This is getting very murky, this endeavor to prove AMD's cores aren't cores has no light at the end of tunnel no matter where you look.

This is AMD's heterogenous computing initiative (read: APUs) that matches this description. Any type of processing unit still works the same way, processing instructions.

 

[Vya Domus]

This is AMD's heterogenous computing initiative (read: APUs) that matches this description. Any type of processing unit still works the same way, processing instructions.

As far as I am concerned this is some independent author's definition irrespective of AMD's initiatives. So this is meant to apply to any type of processor not just AMD's APUs.

So what do we do after all of this, what is a core at the end of the day ? The material out there is in accordance with AMD's claims, or rather, it doesn't contradict them in any way if you prefer it that way.

Why would I consider that AMD settled because they thought they couldn't win ? I seriously doubt that, the plaintiffs could never prove this, there is simply no way to construct a counter argument that makes sense and which doesn't trump a million other definitions as well. They know this and that's why they accepted the settlement too.

They settled because this is actually a really small amount of money that they need to pay under obscure terms and this way they don't look like a big evil corporation squashing some displeased customers.

 

[londiste]

So, do you think GTX970 4GB scandal was bullshit as well? They did deliver 4GB as promised, after all.

 

[Vya Domus]

So, do you think GTX970 4GB scandal was bullshit as well? They did deliver 4GB as promised, after all.

Don't try to get get this subject astray, that had nothing to do with cores or changing the definition of things.

Nvidia provided incorrect specifications about one of their products and not only that but there was no way you could verify this easily. AMD didn't, even if you thought their modules didn't actually contain two cores you could simply go and look at the diagrams and information that they released about their micro-architecture.

Nothing was hidden about what was going on under the hood of their products as was in Nvidia's case.

 

[londiste]

Given how eager you are to try to twist things and find loopholes to nitpick, I was just curious if you could tell me why Nvidia decided to settle that one or why they shouldn't have.

But ontopic - How do you define a core?

 

[Vya Domus]

How do you define a core?

Id say a core is one of those two things AMD put in a module. The book you provided made that pretty clear if you want to stick by it.

 

[HD64G]

So, do you think GTX970 4GB scandal was bullshit as well? They did deliver 4GB as promised, after all.

It wasn't DDR5 all of it as it was written in the specs...

Whereas AMD had the 4-modules with 2 cores each that shared some resources and that was clear from the start. No matter what we use to call cores, it has 8 of those. Might be weaker in some calculations and better in others (it was a server focused arch afterall not helped by the manufacturing process of 32nm vs the 22nm used by Intel back then) but it wasn't a lie. Those arch's shortbacks were clearly shown in the day-1 reviews' results where the previous gen AMD CPUs were almost equal in some tasks and fell back in others. In conclusion, AMD didn't gain market share or sales from the 8-core label on the box.

 

[Vya Domus]

why Nvidia decided to settle that one or why they shouldn't have.

Sure, they wrote that their product had something which it didn't and no material that they provided could have indicated that it didn't.

In other words, they were falsely advertising something and actively hiding information. There was no way that they could have denied this, if I remember correctly they even admitted they did a "mistake". That's why they settled.

But these two cases aren't alike, they deal with different matter. You brought this up because I think you like to twist and nitpick at things.

 

[londiste]

It wasn't DDR5 all of it as it was written in the specs...

It was. GTX970 had 4GB of DDR5.

Sure, they wrote that their product had something which it didn't and no material that they provided could have indicated that it didn't.

In other words, they were falsely advertising something and actively hiding information. There was no way that they could have denied this, if I remember correctly they even admitted they did a "mistake". That's why they settled.

It wasn't memory. That they could have gotten away with.
It was ROPs and L2. Whether their admitted 'mistake' was a mistake is arguable but I am pretty sure it wasn't. With correct specs the memory problem would have been found much quicker

Whereas AMD had the 4-modules with 2 cores each that shared some resources and that was clear from the start. No matter what we use to call cores, it has 8 of those. Might be weaker in some calculations and better in others (it was a server focused arch afterall not helped by the manufacturing process of 32nm vs the 22nm used by Intel back then) but it wasn't a lie. Those arch's shortbacks were clearly shown in the day-1 reviews' results where the previous gen AMD CPUs were almost equal in some tasks and fell back in others. In conclusion, AMD didn't gain market share or sales from the 8-core label on the box.

Not some resources. Most of them. This is why definition of a core matters.

Given a sufficiently parallelized workload a perfectly valid expectation for 8-core CPU is 8 cores to perform 8 times better than 1 core (perhaps 7.9 times). With Bulldozer, that is not the case. It will perform about 80% of that, sometimes worse (with FPU load).

 

[Vya Domus]

Given a sufficiently parallelized workload a perfectly valid expectation for 8-core CPU is 8 cores to perform 8 times better than 1 core

No, it isn't. You can get widely different results and even witness a regression in performance, despite the fact that a task can be parallelized.

parallel =/= faster (not always faster)

more cores =/= faster parallel execution (definitely not always faster)

That's what your average Joe might think , that 8 means times 8 performance. But if he feels cheated by AMD he should feel cheated by any CPU manufacturer that does multi-core CPU because that's the case for all of them.

But it always gets down to the same complaint in the end, that AMD's CPU wasn't as fast as others. In the end this core debate garbage isn't even relevant to most people. I am sure as hell they wouldn't have given a damn if AMD didn't have real cores but they were 10 times faster.

 

[londiste]

Post your Cinebench R23 Score

Download Cinebench R23 Benchmarks FFXV Benchmark Cinebench R23 "Multi" Scores are clickable, leading to the original post. NameCPUCoreClockSingleMultiCooling mirrormaxEPYC 7742 (x2)128C/256T@ 3160 MHz 100981 cbAir nepuEPYC 770264C/128T@ 2499 MHz 48844 cbAir Bret WeeksRyzen Threadripper...

www.techpowerup.com

Going through the list looking for x-core/x-thread CPUs with single core results and OC (to be kind of sure sc runs at the same clock):
i5 8600K - 6C/6T @ 5282 MHz - 557 cb - 3276 cb - 5.88x (98%)
FX 8370 - 8C/8T @ 5080 MHz - 263 cb - 1719 cb - 6.54x (82%)
i3 8350K - 3C/3T @ 4988 MHz - 525 cb - 1537 cb - 2.93x (98%)
FX-8320 - 8C/8T @ 3813 MHz - 195 cp - 1264 cb - 6.48x (81%)

FX-83x0 is Piledriver with minor improvements over Bulldozer.

 

[Vya Domus]

That's cool listing of results for Cinebench, I don't know what you are trying to prove though.

Not everything is Cinebench and let's just say that the user that was expecting perfect scaling and felt lied to wasn't probably thinking about running just Cinebench.

There is a ton of software that hits a dead end after a certain point in terms of multithreading, as a matter of fact that's generally the rule not the exception.

 

[londiste]

Cinebench is an example of a benchmark where CPU should scale perfectly to cores. It does and always has.
For some reason, it does not work like that on Bulldozer (or Piledriver).

20%-ish performance deficiency is noteworthy, isn't it?
This would be an awesome result for SMT and is quite noteworthy for CMT but is very worrying for independent cores.

 

[Mistral]

Cheaper to pay up than to drag this through court and win, I guess...

 

[Vya Domus]

20%-ish performance deficiency is noteworthy, isn't it?

Not really, I don't find it earth shattering. I suppose you can call anything nonzero noteworthy.

For some reason, it does not work like that on Bulldozer (or Piledriver).

For some reasons it works in the other way around too for other CPUs were you get higher than the number of cores scaling. What about that ? Do those have extra cores and the manufacturers forgot to tell us ?

Or is it that software scales in all sorts of ways irrespective of how many cores there are ?

 

[londiste]

Edit:
I get it, you are just trying to be a contrarian. OK.

For some reasons it works in the other way around too for other CPUs were you get higher than the number of cores scaling. What about that ? Do those have extra cores and the manufacturers forgot to tell us ?

Examples?

 

[Vya Domus]

Examples?

i9 9980XE

18C/36T

@ 5200 MHz

525 cb

11687 cb

22x scaling. This one does have SMT but hey, 18 cores are 18 cores. Your time x cores math is a valid expectation, according to you. Cores are cores and everything is set in stone.

Also, you are aware that in your amazing comparison of Piledriver with other CPUs that those newer Intel processors have much more aggressive single core and multi core turbos, right ?

Check this out :

i5 6500

4C/4T

@ 3200 MHz

374 cb

1326 cb

1326 / 374 = 3.54x , 88.5% , interesting. Not quite 100% but much closer to 80% the deficiency goes down to about 6%, noteworthy not noteworthy what do you say. Is this really a quad core I wonder ?

I get it, you are just trying to be a contrarian. OK.

And you replying to every comment of mine for the past couple of pages are ... not ... right ?