[Ars] AMD sued over allegedly misleading Bulldozer core count

[AtenRa]

Intel did.

I think they have a decent shot at winning. AMD should have used the term modules which most of the tech sites use but they obviously wanted to deceive that it was better than Intel's products.

At the time of release, the 8-Core FX was faster in Integer Multi-Thread than Intels CPUs at the same price.

 

[LTC8K6]

At the time of release, the 8-Core FX was faster in Integer Multi-Thread than Intels CPUs at the same price.

Yeah, but I think that was it's one bright spot.

http://techreport.com/review/21813/amd-fx-8150-bulldozer-processor/17

 

[mrmt]

A question I'd like to ask this idiot who is suing AMD -- how exactly do you think you'd be able to squeeze an additional 4 floating point units onto a chip fabbed at 32 nm? Because we're all ears. There just doesn't seem to be enough physical real estate IMO.

I suggest you check a Vishera layout, plenty of space for a lot of things there.

 

[svenge]

I suggest you check a Vishera layout, plenty of space for a lot of things there.

The die space is big enough that I heard that the DoD is assessing if the new F-35C carrier variant can land on it effectively!

 

[Dresdenboy]

Don't forget the FPU's power consumption and usage efficiency. Due to the many multi-cycle instructions and 3-4 pipelines, there are enough bubbles during execution of most FP codes, that a second thread can make use of it, while the fixed power cost of the block doesn't increase.

As I wrote in the Zen thread, the power consumption of Intel CPUs up to HSW is roughly 3/4 fixed cost and 1/4 caused by instruction execution. It's likely similar for Vishera.

 

[MiddleOfTheRoad]

I suggest you check a Vishera layout, plenty of space for a lot of things there.

What the hell are you smoking? The die size was already 315 mm² (An i7 4790K is about 177 mm² by comparison). What do you want? 350 square feet with a refrigerator and a bed too?

 

[C.Cardinale]

CMT results depending on the application between 120% and 180% (or so) performance of a simple core without Hyper-Threading. I.e. a CMT module provides, substantial and expressed in simple cores between 1.2 and 1.8 cores. AMD has never brought it to 8 cores in any of its FX CPUs. So one must conclude the bizarre fact that 4 module of FX CPUs have between 4 x 1.2 = 4,8 and 4 x 1.8 = 7.2 cores, in substantial terms.
Therefore illiterate consumers may have been deceived if AMD marketed 4-module of FX CPUs as "8-core CPUs".
In addition, the company AMD maintains a certain tradition in grotesque gaffes, about the failed concept of ACP (in order to make better look the Opteron CPUs), triple-core Phenoms, or 220-Watt bulldozer derivatives, or just the oblique bulldozer architecture...

 

[mrmt]

What the hell are you smoking? The die size was already 315 mm² (An i7 4790K is about 177 mm² by comparison). What do you want? 350 square feet with a refrigerator and a bed too?

You wouldn't talk about smoking something if you bothered to have a look on the Viscera/Zambezi layouts, there's plenty of space there.

Btw, 315mm^2 is not something out of the ordinary for >>real<< servers chips, it's only by AMD own incompetence that its 315mm^2 chip competes with chips half the die area.

 

[VirtualLarry]

CMT results depending on the application between 120% and 180% (or so) performance of a simple core without Hyper-Threading. I.e. a CMT module provides, substantial and expressed in simple cores between 1.2 and 1.8 cores. AMD has never brought it to 8 cores in any of its FX CPUs. So one must conclude the bizarre fact that 4 module of FX CPUs have between 4 x 1.2 = 4,8 and 4 x 1.8 = 7.2 cores, in substantial terms.
Therefore illiterate consumers may have been deceived if AMD marketed 4-module of FX CPUs as "8-core CPUs".
In addition, the company AMD maintains a certain tradition in grotesque gaffes, about the failed concept of ACP (in order to make better look the Opteron CPUs), triple-core Phenoms, or 220-Watt bulldozer derivatives, or just the oblique bulldozer architecture...

Even eight "real cores" wouldn't perform 8x a real single-core, due to how software multi-core scaling works, so you can't really use a performance argument to say that proves it's not a "real" 8-core.

 

[kalmquist]

I read the complaint.

To begin with, I suspect that the lawsuit is fraudulent. According to the complaint, the plaintiff is an individual named Tony Dickey who purchased two FX-9590 CPU's from Newegg. The fact that he ordered two at once, and successfully installed these CPU's (which require a liquid cooling system) suggest a certain level of computer knowledge. Never the less, he was not aware that the processors used a CMT design. Prior to purchasing this product, he went to the AMD web site and read that it had 8 cores, but overlooked the text that advertised the "Shared FP Scheduler: Dual 128-bit Floating point engines &#8211; capable of teaming together for 256-bit AVX instructions or operating separately with each core." He chose the $300 AMD processor over an unspecified Intel 6-core processor (presumably the i7-4930K at around $570), but would not have purchased the AMD processors if he had known they used CMT. Although he is unhappy enough with AMD to sue AMD, he would consider purchasing AMD products in the future.

While I don't have any direct evidence that Tony Dickey is a liar, the story he tells strikes me as too improbable to be believable.

The complaint seeks to convert the lawsuit to a class action on behalf of, "All individuals in the United States that purchased any of the following AMD Bulldozer processors: FX-8120, FX-8150, FX-8320, FX-8350, FX-8370, FX-9370, and FX-9590." The majority of these are actually Piledriver rather than Bulldozer, but whatever. In any case, the class action only extends to people who have the expertise and desire to install their own CPU's, not to people who buy complete computers.

The justification for converting the lawsuit to a class action comes down to the claim that people who build their own computers are stupid. The complaint asserts, without evidence, that "Average consumers in the market for a CPU lack the requisite technical expertise to understand the underlying design of the Bulldozer processors."

In their complaint, the lawyers cite Bulldozer reviews on Tom's Hardware and Anandtech which explain CMT. Why, you may ask, did these reviewers bother to explain CMT if their readers are too stupid to understand it? The lawyers claim that the sites that explain CMT are "technical trade publications (i.e., publications not read by average consumers)." Remember that, in this context, "average consumers" refers to people who buy CPU's as separate products, not in prebuilt systems.

The "About AnandTech" page on this site states that, "AnandTech serves the needs of readers looking for reviews on PC components, smartphones, tablets, pre-built desktops, notebooks, Macs and enterprise/cloud computing technologies. We are the largest independent technology website doing all of this with over 12 million unique readers per month." It is hard to see how the claim that Anandtech isn't targeted at consumers can be anything other than a bald-faced lie.