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

[xtreemchaos]

ive a bulldozer and 2 piledrivers which I love, ive moved on now but still keep them for all times sake, to tell the truth it don't mean much if there 4 or 8 core to me there a part of my life and I enjoyed them and thay made me happy.

 

[RichF]

CMT is a little bit of this, a little bit of that. SMT implies there are no added execution resources for additional threads. AMD added an Integer Cluster for CMT. Aside from that they are very similar.

To many consumers, SMT is also very similar to a real core. To others it's not. This lawsuit is about consumer perception, apparently. A man is upset that his CPU wasn't as fast as he expected it to be.

Keep in mind that added Integer Cluster does not necessarily mean a huge boost in execution resources. Bulldozer Integer Clusters contain 4 pipes each (2 ALU, 2 AGU) and FPU contains 3 pipes (2 FMAC + MMX). At the same time, Zen's Integer Cluster has 6 pipes (4 ALU, 2 AGU) and FPU has 3 pipes (2 FMAC + MMX).

Zen is also a much more recent CPU that has the benefit of a smaller node for a larger transistor budget (as well as the talent of Keller and 20/20 hindsight vision). AMD made decisions with Bulldozer that resulted in weak cores. Weak cores don't mean justified lawsuit.

 

[64K]

law suite for just 35 bucks each lol

Well, Nvidia settled the GTX 970 class action lawsuit against them for $30 each.

 

[RichF]

Well, Nvidia settled the GTX 970 class action lawsuit against them for $30 each.

That was a cut-and-dried case, by comparison. The 512 MB partition had abysmal performance, nowhere near the level of GDDR5. Plus, to top it off, it came with XOR contention.

Selling a "4 GB" card with 512 MB of its VRAM being sneaky hidden slug RAM is fraud, pure and simple. This is especially true given the context of marketing it as a 4 GB part when 970 SLI was a big thing because of the high cost of the 4 GB 980.

Plenty of people felt duped because they would have more strongly considered a 980, gone AMD, or waited for something else before spending their money. But, hey, look at the great deal on the 970 in SLI — it has the same amount of VRAM as the expensive 980! Oops.

I remember how Witcher 3's visuals were downgraded to fit into the 970's 3.5 GB. Nvidia is powerful, indeed.

 

[Deleted member 163934]

1) FPU emulation has always been vastly slower than having a hardware FPU. So, pointless.

2) I'll repeat my question: Zen 2 also performs better with AVX-256 than Zen 1 because it can execute 256-bit instead of combining 128s. Does that mean Zen 1 didn't have any real cores in it?

Zen 1 can't execute 256-bit AVX independently. It has to combine at the 128-bit level. So, it doesn't have any real cores, eh? Not only is it slower at doing 256-bit AVX, it can't do it independently.

Let's just say I have an workload that I can happy split it easily in an even way in 8 threads or 4 threads, it's a finite worload. Spliting it in 8 threads should finish it considerable faster than 4 threads.
If I do this with 8 FPU only threads and 4 FPU only threads on the 8150 the workload is gonna finish in an similar amount of time in both cases. Main problem is that for an real 8 core cpu this is not what I would expect. If I don't say what type of threads I used, because I don't have to say, you advertised an 8 cores 8 threads cpu so I can put whatever type of threads I want on that cpu, then anyone that see the result will start to wonder why a properly thread splited workload is showing no improvement at all when using 8 threads compared to 4.

You gonna say again about faster. And you are wrong. (If the 8 and 4 threads have nothing to do with FPU the 8150 starts to behave as expected...)

If 8150 would had been advertised as an 8 cores 4 threads cpu I wouldn't had anything to say. (I would wonder why you need 2 cores for 1 thread and how exactly you split that 1 thread internaly between the 2 cores...)
You are free to point to a cpu advertised to have more cores than threads. I'm not aware of such case.

 

[64K]

That was a cut-and-dried case, by comparison. The 512 MB partition had abysmal performance, nowhere near the level of GDDR5. Plus, to top it off, it came with XOR contention.

Selling a "4 GB" card with 512 MB of its VRAM being sneaky hidden slug RAM is fraud, pure and simple. This is especially true given the context of marketing it as a 4 GB part when 970 SLI was a big thing because of the high cost of the 4 GB 980.

Plenty of people felt duped because they would have more strongly considered a 980, gone AMD, or waited for something else before spending their money. But, hey, look at the great deal on the 970 in SLI — it has the same amount of VRAM as the expensive 980! Oops.

I wasn't usisng the GTX 970 as a comparison to AMD's case. I was just pointing out to PerfectWave that these mini settlements ($35) aren't unusual. I doubt anyone expected to get compensated the full amount that they paid for the Bulldozer CPUs.

 

[RichF]

I wasn't usisng the GTX 970 as a comparison to AMD's case. I was just pointing out to PerfectWave that these mini settlements ($35) aren't unusual. I doubt anyone expected to get compensated the full amount that they paid for the Bulldozer CPUs.

Ok. The comparison is worth making.

Another further point about it, too: The 970's VRAM wasn't even apparent initially. By contrast, anyone who had bothered to read reviews from legitimate sites would have seen that Bulldozer wasn't delivering (and Piledriver, although, at least, it was very cheap near the end). 970 buyers truly were tricked. Bulldozer and Piledriver buyers were either too impatient (pre-orders) or too lazy (didn't check reviews) and deserve the product their received. AMD was under no legal obligation to make its cores as powerful or more powerful when compared with Intel's.

It's a good thing for Intel that this guy didn't buy an Atom. It wasn't out-of-order, after all. He would have sued on the basis that any modern CPU is expected to have out-of-order design for performance since in-order hadn't been on the market since the Pentium 1. Of course, there's also the bait-and-switch of the chipset using far more power than the CPU, negating the justification for the performance-killing in-order design to conserve power.

 

[FordGT90Concept]

Citation please. Everything I've read said FX CPUs' CMT is not SMT. I am also interested in how that BIOS setting works, considering that you are saying it's mislabeled by Gigabyte.

CPU Instructions | pclt.sites.yale.edu

pclt.sites.yale.edu

The CPU has to fetch the data, then turn the instruction over to one of the processing units.

CPU = core
processing units = integer cluster

 

[londiste]

CPU = core
processing units = integer cluster

Incorrect. At least from CS point of view.
Integer cluster is execution unit.

 

[FordGT90Concept]

Execution units are inside the integer clusters.

1+ CPU -> 1+ core -> 1+ integer cluster -> 1+ execution unit
FX-8350 -> 4 core -> 8 integer cluster -> 64 execution unit (don't quote me on that, fast research)
R7 1800 -> 8 core -> 8 integer cluster -> 48 execution unit (again, fast research)

 

[RichF]

CPU Instructions | pclt.sites.yale.edu

pclt.sites.yale.edu

CPU = core
processing units = integer cluster

And yet Gigabyte didn't label the BIOS setting "One thread per core". Differences of opinion apparently surround this hardware all over tech.

 

[FordGT90Concept]

I'm not talking about Gigabyte at all.

I just stated a simple fact that as far as software is concerned, they're the same. Each core is either given one thread or two. It's no more complicated than that without drilling deep into the hardware design.

 

[Vya Domus]

Do a 256-bit FMAC operation and the thread on the other integer cluster is stalled until it completes. Impossible on a dual-core processor.

How can you still go back to this, has it not being made clear to you that this has nothing to do with whether or not something is a single or dual core processor ? In what world is it required for a core to do some arbitrary 256-bit instruction ?

Cores share nothing other than memory ("core replication is obvious"). Bulldozer violates that by sharing fetchers, decoders, dispatchers, and Core Interface Unit. That's shared logic, not just memory.

You have been proven wrong many times on this, cores share caches, crossbars, memory controllers, and even other DSP like units without which they could not function independently.

 

[RichF]

I'm not talking about Gigabyte at all.

You cited the words of a software engineer from Yale. I am pointing out that other software engineers may see things differently.

Unrelated to this post but related to the previous one about Atom... In the case of that bait-and-switch, it actually was worse for the shopper if one had done his/her homework early-on because there was a lot of hype around how Intel's engineers had worked so hard with a pure philosophy of only adding functionality if it could be put in without using too much power. It was very dreamy in its presentation, at least that was the intent.

The cold reality was that the chipset was a terrible pig that made the entire thing into a sad joke. Despite that, the hype machine was in full force and netbooks with this terrible tech sold very well as a result.

 

[londiste]

As for the claim that Bulldozer doesn't function with intra-modular cores disabled... I guess you're unfamiliar with BIOS settings that can do that. I've run both the 8320E and 8370E with 4 cores via the 1 integer core per module setting.

'1 integer core per module' is honestly a straightforward setting and equates directly to disabling HT or SMT. In this case what it does is disabling CMT.

BIOS settings are sometimes a bit of smoke and mirrors. In the long thread there were examples of settings that allow disabling each core separately which is technically impossible. I mean, you can disable part of a module (Integer Cluster) in Bulldozer and the rest will be a functional core. But you cannot disable anything else without compromizing core functionality. This goes back to the problem of not being independent.

 

[Mephis]

Cores are like porn. I know one when I see one.

 

[FordGT90Concept]

In what world is it required for a core to do some arbitrary 256-bit instruction ?

It's not the type of operation that matters, its the fact that it blocks a parallel thread. Dual cores can't block because they're separate processors. A single core can block when there are shared logic circuits (e.g. Hyperthreading and Bulldozer's design).

You cited the words of a software engineer from Yale. I am pointing out that other software engineers may see things differently.

We're talking about truth in advertising. Your average Tom, Dick, and Harry needs to understand it, not engineers. AMD made a critical mistake by omitting "integer" on their products.

 

[ProPain]

Big Companys like AMD sould learn from there mistakes .... otherwise CLASS ACTION LAWSUITES are a good reminder for them ....

AMD - will stop flase Marketing - and I like that !

 

[RichF]

Cores are like porn. I know one when I see one.

That is actually one of the worst judicial atrocities in modern history.

Of course, being utterly corrupt, it's widely praised and influential to this day.

We're talking about truth in advertising. Your average Tom, Dick, and Harry needs to understand it, not engineers. AMD made a critical mistake by omitting "integer" on their products.

Too vague a statement to be useful here. One guy's problem with AMD's marketing managed to get some success in the courts. That's proof of very little.

Atom is actually the bigger fraud but since it was backed by Intel and not presented as a high-performance architecture people were more content to be duped. Being duped by the illusion of some grand power-conservation Zen magic is more tolerable, I guess — despite the plethora of netbooks that filled stores and which vanished like the mighty dinosaur in a comparative flash.

Atom would have been lame with a good chipset but it would have been lame on its own terms, not false pretenses. At least it would have been quieter, with more battery life, and with less heat.

 

[FordGT90Concept]

Too vague a statement to be useful here. One guy's problem with AMD's marketing managed to get some success in the courts. That's proof of very little.

What's vague about this?

AMD screamed the lie from the top of the mountains.

New processors only have the core/thread count on the tiny little sticker sealing the box now.

If they just snuck the little word "integer" in there, this would have never happened because the packaging is truthful. But no. They had to make their product stand out...

 

[RichF]

You're back to recycling the whole FPU thing? Are we doing the time warp again?

(makes Magenta face)

quote: "AMD made a critical mistake by omitting 'integer' on their products."

 

[FordGT90Concept]

No, this is what an 8 core looks like:

Not this:

This guy is many billion transistors short of what it claims to be.

 

[Vya Domus]

Dual cores can't block because they're separate processors.

They can totally block each other and they do it all the time, for example when two threads from different cores try to write to the same cache line of say the L3 cache which is shared.

There isn't a single multi-core processor on the planet that can truly operate 100% independently and that doesn't share anything. This notion of separate processors does not exist in this context, stop using it.

The only truly separate processors are the ones on different boards. End of story.

 

[londiste]

L3 cache is not core.

 

[RichF]

So, a wafer-scale processor could have 3,000 "quasi-cores" but it's a single-core chip if those "quasi-cores" share things other than cache?