Email about AMD class action lawsuit

[MrGRiMv25]

How does one prove that though?

Unless someone hasn't cleared their browser history since Bulldozer arrived then they have no chance at proving they went to the site - if they'd changed ISP's in the time since then doubly hard to prove you visited the site.

I'd probably say that I went to AMD's site to look at the specs before buying the CPU to see if it "met my needs" - they can neither prove nor disprove it so any person claiming should be in the clear as long as they just say they did look at the site - it's probably worded that way so people are put off claiming or something

 

[moproblems99]

Whatever someone wants it to be, but generally it does math of some kind. Bottom line it's ambiguous as heck.

If my understanding is correct, addition. The only thing it really does is addition.

 

[notb]

If my understanding is correct, addition. The only thing it really does is addition.

Every arithmetic operation can be written as a sequence of additions (or an inverse operation).

Fundamentally, a CPU is performing bitwise operations (comparing, changing and shifting 0s and 1s).
So actually, multiplying (by 2 or 2^-1) is the most natural arithmetic operation.

Practically speaking (on the interface visible to the OS), CPU performs instructions and these can be pretty surprising (because of the bitwise operation nature).
For example: CPUs are very good at performing a MAC, which is this:
a <- a + b * c

 

[R-T-B]

Fundamentally, a CPU is performing bitwise operations (comparing, changing and shifting 0s and 1s).
So actually, multiplying (by 2 or 2^-1) is the most natural arithmetic operation.

Makes it triply interesting to think about the old 8-bit motorolas that completely lacked a multiply instruction, doesn't it? heh.

 

[seronx]

There's no strict definition of a core. It's not an abstract mathematical entity.

There is a strict definition of a core.

The sum of the parts of a core are only the:
Retire and Scheduler unit
Execution EX0/EX1/AGLU0/AGLU1
Load/store unit

Control unit equals the Superscalar OoO control unit: Retirement(in-order) and Rename/Scheduler(out-of-order).
Datapath unit equals the Superscalar execution/address generation units: EX0/EX1/AGLU0/AGLU1.
Data bus equals the Load/store unit: LSU
Instruction bus equals the Retire -> Scheduler bus: Macro-ops get converted to micro-ops, etc.

Core = "the central or most important part of something."
So, when some says the Processor core => they are talking about the middle of the processor, not the processor.

Bulldozer module has two central cores, that are required to exist because they are important.

None of these go beyond their external bus.
{Processor Core} <-> {Processor Unit} <-> {Processor Device}
Bulldozer Core <-> Bulldozer Module/Compute Unit <-> Bulldozer Processor

Core is only a portion of the unit. The unit is only a portion of the device.

 

[notb]

Makes it triply interesting to think about the old 8-bit motorolas that completely lacked a multiply instruction, doesn't it? heh.

I said: multiplying by 2 or 2^-1. That's just shifting bits.
You're right: there wasn't a straightforward multiplication instruction. You had to do it by shifting and adding.
Modern CPUs offer a multiplication instruction, but deep down it's implemented the same way. We just don't have to worry about it anymore (thankfully).

There is a strict definition of a core.

There isn't, but I admire your imagination.

 

[Totally]

I said: multiplying by 2 or 2^-1. That's just shifting bits.
You're right: there wasn't a straightforward multiplication instruction. You had to do it by shifting and adding.
Modern CPUs offer a multiplication instruction, but deep down it's implemented the same way. We just don't have to worry about it anymore (thankfully).

There isn't, but I admire your imagination.

Even if there is one today there definitely wasn't one when bulldozer came out as I remember "what constituted a core?" or "what was the definition of a core?" and were two very contested topics, and I think the only thing agreed/accepted at the time by everyone was that a core needed to be able to handle and execute it's threads independently of the others.

 

[seronx]

There isn't, but I admire your imagination.

There is... it is defined by four units: a control unit, a datapath, a data bus, and an instruction bus.

Even if there is one today there definitely wasn't one when bulldozer came out as I remember "what constituted a core?" or "what was the definition of a core?" and were two very contested topics, and I think the only thing agreed/accepted at the time by everyone was that a core needed to be able to handle and execute it's threads independently of the others.

The definition of a core always existed.

The core is the center of the processor. In a single core processor, it is the middle of that processor. In a multi-core processor, it is in the middle of each of the multiple processors.

The definition of a core by the technical standard by the industry is extremely strict. That ONLY the aggregate execution portion of the processor: Retire Buffer/Schedulers/Datapaths/Load-Store, is the core.

The front-end of the processor isn't the Core. (FP, IF, ID, etc)
The external bus unit attached to the processor isn't the Core. (CU, L2, etc)
The co-processor of the processor isn't the core. (FPU, etc.)
The caches of the processor isn't the core. (L1i, L1d)

The core is the smallest portion w/ nothing beyond L0i and L0d being allowed to be considered part of the core.

 

[Totally]

There is...

I didn't say there was it was ambiguous at the time, pretty much every chip maker had their own definition of what constituted what was a core. Main reason why I thought the whole thing was stupid, everyone couldn't agree on what a core was but also called into question whether a bulldozer core was a core or not.

it is defined by four units: a control unit, a datapath, a data bus, and an instruction bus.
There was always a definition of a core.

In 2011, was this the universally accepted definition of a core?(rhetorical question, btw). Therefore it's an opinion not a fact, even it is a popular opinion, it is still an opinion.

 

[seronx]

In 2011, was this the universally accepted definition of a core?(rhetorical question, btw)

Yes, what is a processor's core was defined well before the 1990s.

Unlike word definitions, technical definitions do not change.

 

[notb]

There is... it is defined by four units: a control unit, a datapath, a data bus, and an instruction bus.
The definition of a core always existed.

You're repeating yourself.
Can you point us to a respected source that you got your "definition" from? A well known book? A journal?

 

[R-T-B]

There is... it is defined by four units: a control unit, a datapath, a data bus, and an instruction bus.

If that was true this settlement wouldn't exist.

Even technical definitions change. Evidence? Settlement we are discussing.

 

[Totally]

Yes, what is a processor's core was defined well before the 1990s.

Unlike word definitions, technical definitions do not change.

For a logic core (what AMD claims, weakened due to windows reporting 4 cores in TM early on and the cause of this mess IMO), the earliest for a physical core that I know of was made by Intel in 2006 (lawsuit claimed, this is what is under contention in this case).

 

[notb]

If that was true this settlement wouldn't exist.

Precisely that.
This is what we need law and lawyers for. Because some things aren't strictly defined and interpretation (judgement) is needed. So we create a legislation system, regulations, statuses, treaties, agreements, standards - all with just a sole purpose of adding some order to the chaos of everyday life.

So that when we wait for a bus, we're picked up by a big box that moves by itself - no pedaling necessary. When we order a pizza at the restaurant, we don't get a haircut instead.
And when we buy a 4-core processor, it's a 4-core processor IN A WAY WE'RE ENTITLED TO EXPECT. Because we've seen or bought some-number-of-cores processors before.

No one sues anyone about a definition of circle. It's a definition. There's nothing to discuss or interpret.

Some comments in this thread are so out of touch with reality that it's just really difficult to respond - let alone convince the author that he's wrong...

Even technical definitions change. Evidence? Settlement we are discussing.

Because most of them aren't definitions. They are just standards.
We have definitions of mathematical objects and operations. We have definitions of chemical elements.
But we don't have, for example, a definition of a gram (unit of mass). People often call it a definition, but it's a standard - so it can change. And it did earlier this year (for the first time since XIX century).

We've had a "standard" for a car for over a 100 years and at every moment a car was "defined" in a way that most people knew what to expect.
But most cars in history - despite being perfectly fine when they came out - couldn't even be sold today as new.

It's the same story with a core. The only problem is: it's too complex and too abstract to easily comprehend. That's why PC forums so often use car analogies - not the other way around.

 

[DeathtoGnomes]

That's why PC forums so often use car analogies - not the other way around.

What are you driving at?

 

[CountMike]

We all know what apple core looks like but in lieu of processors that only came to attention when multi-core processors came to be. At first it equaled number of processors in the same package but cennecting those 2 (or more) processors inside the package (instead by chipset on the MB in case of "multi processor systems) muddied the waters considerably. Intel kinda made it worse by labeling their CPUs with "Core" (Core 2 Duo/Quad for instance). As an old timer i still think of "Core" as processor with as many of them stuck in one package along with interconnects and IO where applicable.
On the other hand, SW works with threads not directly with processor.