AMD Dragged to Court over Core Count on "Bulldozer"

[FordGT90Concept]

I had a dual, quad-core Xeon server a while back. Motherboard died. Anyone that has access to a system with 8 cores is free to try it and I'll graph it:
https://www.techpowerup.com/forums/threads/random-password-generator.164777/

1. Set max threads to 8.
2. Set random seed to 0.
3. Set minimum characters to 32.
4. Set "Maximum number of attempts to generate a password" to 10000000 (default is 1 million, add a 0 to make it 10 million).
5. Uncheck "include special characters."
6. Check "number of special characters required."

Run it, note the time in the corner and the number of threads., change max threads to 7, rinse and repeat through 1.

 

[FR@NK]

I'll test it on my 8 core.

I will need to turn off turbo boost and lock all the cores to the same speed. Did you do this when you tested it ford?

Update:

1 threads: 29.8149422
2 threads: 15.7981694
3 threads: 10.7508806
4 threads: 8.2506762
5 threads: 6.7012710
6 threads: 5.6937791
7 threads: 5.0004093
8 threads: 4.5316209

 

[cdawall]

Have you used a CPU with more than 4 "real cores" by your own definition to show that your application is actually capable of speeding up to a reasonable extent past 4 threads? Most applications don't speed up very well and start hitting some form of limitation when it comes to concurrent processing of data, even more so if it's just making the full calculation run in parallel and not using different threads to handle different stages of the task. I suspect that if the trend for the times for both the 8350 and the 6700K are the same, that the way it's written merely doesn't speed up past so many cores. This doesn't make AMD's CPU not have real cores, it's just part of the reality that is designing software.

If you don't believe me than maybe you should send the binary to @cdawall to run on his Opterons to see if it scales past 4 or 5 threads and maybe another member who has a 6c or 8c Intel CPU to generate some numbers for us. This is a claim that can be validated, so it should be because not all software scales and unless it has been tested on a machine fitting your "real core" criteria with more than 4 of them, I would say that you have insufficient data to assert that your benchmark is even capable of showing such optimistic speed up with additional cores.

I have a 5960x as well as the opteron. When testing on the 5960x I couldn't get consistent results so I didn't post them.

I always due my research before buying too but this guy may not be responsible for his mistake if he can convince the jury that AMD failed at "truth in adverting" and that burden is on AMD and not the customer to advertise their CPUs honestly and clearly.

They have 8 cores is why this will go in the trash. Without a previously written definition there is no malicious intent nor is there a way to prove these aren't cores. Hell even ford calls them cores, he just tried to preface it and make it seem like he disagrees.

 

[FordGT90Concept]

I'll test it on my 8 core.

I will need to turn off turbo boost and lock all the cores to the same speed. Did you do this when you tested it ford?

Update:

1 threads: 29.8149422
2 threads: 15.7981694
3 threads: 10.7508806
4 threads: 8.2506762
5 threads: 6.7012710
6 threads: 5.6937791
7 threads: 5.0004093
8 threads: 4.5316209

Big surprise here, it's right where I expected it to be (compare "8-core" column to 80% scaling column). I'm taking a screenshot of the raw data to prove I'm not bullshitting anyone:

What was this 8-core by the way? I mean, model of the processor(s) so I can make it match.

This proves not only is the application extremely predictable, the curve you see where it deviates above 2-7 is mirrored in quad-core (deviates above on cores 2-3). This is deliberate, again, to account for the UI overhead that doesn't show until all of the physical cores are loaded.


I tested with turbo enabled. Turbo being enabled might explain why it is 20% and not, say 10%, because I'm using the single thread as the point of reference. If it overclocked that single core, it would exaggerate the single thread test making the multicore look worse. That said, I'm not really worried about it because FX processors would do the same.


Out of curiosity, is this octo-core system running about 3.246 GHz?

I have a 5960x as well as the opteron. When testing on the 5960x I couldn't get consistent results so I didn't post them.

My previous instructions didn't change the random seed. These instructions should get a more consistent result because that was a pretty big omission on my part.

You'll also get wild results if anything CPU intensive is running.

It will kind of be all over the place when running the same test over and over. A single test isn't really important... it's the trend that matters: generally shorter with each thread added. Exception: 4th and 8th (with SMT) thread quad-core processors may be about equal to preceding and the same goes for 8th and 16th (with SMT) thread on octo-core processors because of the UI thread.

 

[FR@NK]

Out of curiosity, is this octo-core system running about 3.246 GHz?

Yea its a 6900k set at all cores to 3.2GHz.

 

[FordGT90Concept]

I added names + clockspeeds to spreadsheet.

 

[FordGT90Concept]

I'm going to change OS SSDs soon and cleaning my computer desktop. Here's the incomplete OpenOffice Document Spreadsheet should it ever be completed...

 

[Aquinus]

You can all thank @FordGT90Concept for reminding me of this thread because I had honestly forgot. It sounds like:

For the foregoing reasons, the court GRANTS defendant's motion to dismiss plaintiffs' claims. Within 14 days, plaintiff shall submit an amended complaint that corrects the deficiencies identified in this order.

IT IS SO ORDERED. Dated: October 31, 2016

https://casetext.com/case/dickey-v-advanced-micro-devices-inc-1

Just wanted to throw that out there since this happened after the last post on the matter.

 

[FordGT90Concept]

Defendant argues that (1) Alabama law applies to Dickey's claims, so Dickey's California claims should be dismissed; (2) plaintiffs fail to state a claim for fraud; (3) plaintiffs fail to state a claim for breach of warranty; and (4) plaintiffs lack standing to seek injunctive relief. Dkt. No. 52.

1. Declined
2. Granted

Lack of Factual Basis for Plaintiffs' Expectations
The court is not convinced that Dickey and Parmer were required to identify a particular statement by AMD or anyone else representing AMD's cores as completely independent so long as plaintiffs alleged a particular, plausible understanding of the term "core" as independent such that AMD's use of the term would be misleading. However, as discussed below, plaintiffs have failed to allege their expectations with sufficient particularity.

Lack of Specificity of Plaintiffs' Expectations and Industry Standards
The court finds that plaintiffs' amended allegations fail to cure the deficiencies previously identified by the court, in particular whether plaintiffs believed that a core could not share resources as well as plaintiffs' particular understanding of what constitutes a core. See Dkt. No. 46 at 7. Accordingly, the court grants defendant's motion to dismiss plaintiffs' fraud-based claims. Plaintiffs will be given one final opportunity to amend to attempt to cure these deficiencies.

3. Granted because of #2.
4. Not addressed because of #2 and #3.

In other words, as observed before, the plaintiff doesn't understand the technology of it well enough to prosecute the case. I think he could win if he argued that the dispatcher can create blocking scenarios which degrades the performance of both integer clusters thereby proving they are not independent.

Example:
http://pds.ucdenver.edu/document/hardware/AMDbulldozer-IEEE-Computer-2011.pdf
The authors use "core" in contexts that are confusing and inconsistent.

It opens saying "It combines two independent cores intended to deliver high per-thread throughput with improved area and power efficiency." Core refers to the integer clusters (article usually refer to them as "interger cores") here. Also note the word "combines" which contradicts the word "independent." They were apart, but now they are together.

Look on page 9, it says "Figure 4 shows how the Bulldozer core uses these different mechanisms." Core refers to the AMD's so-called "module" here and it repeats in the Figure 4 text: "Multithreading model that shows how the Bulldozer core uses different mechanisms."

If a bunch of geeks are using the word "core" interchangeably to describe two very different things, how is Joe Public supposed to know the difference when they see "8-core" in marketing? AMD never made the distinction themselves on their products. I think someone could easily make this argument and win the lawsuit but Dickey doesn't understand the tech well enough to.


I believe the case is no longer being pressed because Dickey didn't motion to appeal within the 14 day limit.

 

[Aquinus]

If a bunch of geeks are using the word "core" interchangeably to describe two very different things, how is Joe Public supposed to know the difference when they see "8-core" in marketing? AMD never made the distinction themselves on their products. I think someone could easily make this argument and win the lawsuit but Dickey doesn't understand the tech well enough to.

You say that but, his legal team was literally twice the size of AMD's. I find it hard to believe that someone who dumps that much time and money into something would just give up if there were options to "easily win." All things considered, there was a motion to appeal earlier and the court still struck it down. Not appealing it is basically saying that they concede and your entire premise runs on the assumption that his team or he didn't know what they were doing but, that doesn't seem to be the case. Even if that was the case, it also could indicate that their argument was flawed in the first place... but if you're such an expert on it, maybe Dickey should have added you to his team.

 

[qubit]

So, is the breakdown of this case good or bad from the consumer's viewpoint?

Personally, I always thought those siamesed cores were a bit of a stretch to count them as two, because the operation of one hampered the other due to those shared resources, so I can see why the lawsuit happened. Whether it's enough to call it fraudulent and worthy of suing over I'm not so sure.

 

[FordGT90Concept]

I find it hard to believe that someone who dumps that much time and money into something would just give up if there were options to "easily win."

They literally only cited two sources which the judge reviewed:
http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043.html
http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested

All things considered, there was a motion to appeal earlier and the court still struck it down.

Because of the lack of standing (#1). They fixed that in this appeal and added a second plaintiff to bolster the case but they failed furnish evidence explain why calling it a "core" is misleading.

Not appealing it is basically saying that they concede and your entire premise runs on the assumption that his team or he didn't know what they were doing but, that doesn't seem to be the case.

Actually, it does, with any court case, standing has to be proven and they had to appeal to prove standing. The law firm that handled this was clearly treading in untested waters. A law firm that deals with class action suits regularly wouldn't make that novice mistake.

Dickey is not wrong: people see "8-core" and the price compared to Intel's 8-core offerings and they think it's a steal. AMD did that deliberately too.

So, is the breakdown of this case good or bad from the consumer's viewpoint?

Largely irrelevant because that Bulldozer/Excavator/Piledriver were the beginning and end of that processor design. It's shit and I doubt any other chip manufacturer will revive it. Because this case wasn't dismissed on specific substance, it likely wouldn't be cited against future misleading designs.