Lawsuit claims AMD lied about the number of cores in its chips

This will hurt them if found guilty

http://www.engadget.com/2015/11/07/amd-processor-core-class-action-lawsuit/

 

I thought their marketing team clarified perfectly well exactly what a 'core' is to them... problem is, intel have coined hyperthreading, and people refer to these as '8 cores' (lamens) and if anyone should be held responsible for the mis-clarification its salesmen. The few times I've ventured into HN, Domain etc, the reps would always push the '8 cores' from an AMD CPU, but wouldn't clarify what a 'core' is.

SO basically, AMD uses its own terms for its CPU's architecture, with an abundance of information clarifying what it is online, countless reviews etc... and consumers fail to do any research on it...

Seeing as AMD didnt even try and hide anything, and were very open about what their product is, I feel this will fall flat on its face...

 

except they wont be... pro tip, look at rock, i386 etc a "core" doesn't mean anyone thing. its a collection of units that does stuff in various ways. All cores on a soc share something at some point, voltage rails, cache, memory controller etc.

 

This is stupid in heaps of ways and shouldn't even be allowed in court, but the worst part is the guy who filed the lawsuit will probably get millions of dollars.

 

It was AMD who first introduced multi-core CPUs (Opteron and Athlon 64 X2 lines)... before then, the term "core" didn't really exist in the computer world. Certainly it was a non-existent concept in the consumer market.

Back then you can have multi-processor systems or "Dual CPUs", but the concept of "core" didn't exist. The very first products that is marketed as cores had dedicated FPUs. And every processor in the consumer market since then have dedicated cores that never shared compute resources. Until Bulldozer came along...

Personally, I'm partial to the concept that the definition of cores should have dedicated compute resources. So 2-core indicates that the 2nd core is the same as the 1st. Its simple.

Edit: On another point, the reason Bulldozer sucks has pretty much nothing to do with its shared "FPU". The FPU isn't even shared... each core is capable of executing its own standard FPU instructions. The difference is that newer 256-bit AVX instructions reuses both FPU so the cores in the module had to share. AVX was first introduced in Bulldozer and Sandybridge, so its not like previous AMD CPUs had dedicated AVX functions either.

AVX application is fairly scarce and generally have minimal impact.. so it pretty much has nothing to do with Bulldozer woes. The issue is basically in order for AMD to cram so many cores in, it reduced the compute resources of each core. So its sort of like if Intel made an 8-core Atom. Thats why it had poor single-thread performance.

 

Will depend on how much the courts know about the technology I suppose.

 

I didn't say x86 market.. I said consumer market.

The courts would look at terminology based on individual markets.

Edit: what happened the comment I was replying to..

 

If I remember they called them modules, right?

But that got thrown out the window at some point, and now we have hexa- and octa cores

 

Retail boxed FX-8150. "8-core" in every side. Even has a prominent statement "World's first 8-core desktop processor".

 

And then you go into the BIOS, disable all, but 1 core, and are still left with 2

 

I wonder if they will argue that once upon a time the math co-processor was external to the "core" and is the bit that is now shared.
so logically the 8 "core" is still correct, just that it has 4 co-processor shared on the same silicon.

its not like there's eight any other part thats now on the main wafer. eg. you dont see eight memory controllers. eight L1/L2/L3 cache units. etc

 

In terms of the last level of cache it's best to have it shared between all cores because it allows the cores to share cache which I imagine is why Intel have the L3 shared between all of them and I'm pretty sure amd had to create a link between modules to make up for that

To all the people who say it's not a real core if it sharse resources well I guess intels cpus aren't real multi core cpus either

 

i use the cache as a (poor) example of other stuff in the silicon other than the "cores"
intel never said a quad core with HT was 8 cores. they have always maintained a 4c/8t etc approach to their cpus

 

After a bit of googling, there seems to be no definitive description regarding specifically what a 'core' is obligated to contain to actually be classed as a core. It is a very general term and has been used loosely through technology for many years on this basis alone I think this case will be thrown out

 

One of those cases of a little bit of knowledge can be a dangerous thing...

I hope he gets awarded a slap upside the head.

As far as I am concerned after 30+yrs messing with computers, if it runs a thread its a core.

Intel i7 4c/8t is ass covering done well but really, if it runs a thread on its own it counts. If it was all about the core count AMD would have a larger market share but its about thread performance. Particularly as multithreaded apps become the norm.

my 2 cents....

 

This has bene discused to death in the other thread, but something to add to the discussion

One simple way to define the vailidty of the term might be to distinguish between SMT and discrete core would be the Performance scaling with extra threads.

Say, on an industry standard benchmark set anything over 50% performance uplift from a doubling of threads (at the same frequency), can have each thread considered a core

In this case, AMD could use the term core for FX (they scale form 60-70%) , whereas Intel , nor AMD's Zen could not as as scaling from SMT (in the form both processors use) is in the range of 20-30%

 

AMD did not in anyway try to hide what their 'cores' were in their 8 core CPU. Given there is no absolute definition of what constitutes a core, and AMD were in no way dishonest or deceptive in their marketing of this 8core CPU, I suspect the judge will take some arguments and then kindly tell them to GTFO.

More cores does not necessarily mean more performance in the same way that more clock speed does not necessarily equal more performance. If this customer was confused, then its his own fault for being fooled by the pitch of computer salesmen, not the fault of AMD for making the product.

 

it is a typical money grabbing useless piece of crap.
https://www.reddit.com/r/buildapc/comments/1e8226/discussion_amds_module_architecture_the_fx_8350/
an example of if he learned to read in his trailer with MA he may have made in an informed decision ,the amount of ma and pas that think their intel has 8 cores would surprise you...they have hyper threading totally different thing.



I wish they kept the 1100t's going and maybe stretched them out and developed something other than the bulldozer but that's life.

 

STARS needed to go, the fundamentals of the core were just to old ( the entire ROB/scheduler/ execution stage and load store systems). Bulldozer ended up being the wrong answer because single thread perf was to low. if bulldozer shipped with +20% IPC then it would have been interesting to see how that played out. Given that PD saw about 10% IPC , SR about 8% and EX about 5% so amd would have kept pace.

L2 to slow,L1 to small and execution stage to narrow. look at all other modern high performance cores vs BD and thats what stands out.

 

ExtremeTech have good article on this topic, kinda hits the nail on the head....

... enough said.