AMD sued over false advert claims regarding their bulldozer CPUs

Maelstorm

Maelstorm

AMD is being sued over their bulldozer architecture processors over false advertisement. The suit claims that the 8-core FX-8000 series CPUs are really 4 cores.

http://www.pcworld.com/article/3003...ozer-cpus-arent-really-8-core-processors.html

So a 8-core processor contains 4 modules which contain 2 cores each. Apparently, the problem is that those 2 cores cannot operate independently so they are in fact one core. Also, there is only one FPU per module as well.

One thing that I want to point out about that last bit: Sun Microsystems (now Oracle) has several CPUs that have an asymmetrical CPU/FPU count. Case in point is their UltraSPARC-T1 processors which have one FPU for 2 CPU cores (I have a machine that uses the T2 CPU). So I don't think that would be a valid point. I mean, really: How often do you use software that requires floating point? On my windows PC, I figure that CAD, mathematical analysis/modeling, and calculators would use floating point by their nature. The graphics engines of games like World of Warcraft, Aion, and EvE Online which feature full 3D immersive graphics may be added to that list.
 
  • Thanks
Reactions: Oko
This lawsuit really has no merit at all IMO. A victory for the plaintiff would rely on an established industry definition that doesn't exist of what, design wise, a CPU core is.
 
I find this hilarious. Its unsurprising to me as AMD has not been competitive in a while with Intel CPU performance, their GPUs suck due to poor driver-hardware interfaces and their drivers are obfuscated by using lots of microcode even in the open source AMDGPU drivers. This court ruling will have industry wide repercussions if the lawsuit is in favor of the plaintiffs, and I hope it happens as there needs to be accountability from CPU manufacturers about what their products provide. I look forward to the results of the court case.

Here's an example of AMD's uncompetitiveness against a Haswell CPU, since it is relevant, Piledriver being the successor of Bulldozer: https://www.reddit.com/r/buildapc/comments/1qs2ku/discussion_i54670k_vs_fx8350/cdfwqex
 
TeamBlackFox said:
I find this hilarious. Its unsurprising to me as AMD has not been competitive in a while with Intel CPU performance, their GPUs suck due to poor driver-hardware interfaces and their drivers are obfuscated by using lots of microcode even in the open source AMDGPU drivers. This court ruling will have industry wide repercussions if the lawsuit is in favor of the plaintiffs, and I hope it happens as there needs to be accountability from CPU manufacturers about what their products provide. I look forward to the results of the court case.

Here's an example of AMD's uncompetitiveness against a Haswell CPU, since it is relevant, Piledriver being the successor of Bulldozer: https://www.reddit.com/r/buildapc/comments/1qs2ku/discussion_i54670k_vs_fx8350/cdfwqex
Click to expand...

So you are arguing this lawsuit should be successful on the unrelated basis that you don't like AMD's products because they're bad and not competitive? Not that I am in complete disagreement with your opinion, but I'm not seeing the relation.
 
  • Thanks
Reactions: Oko
Actually, it's not false advertisement. If could be considered misleading, however, the articles say, it's total computing power is disclosed of the Bulldozer AMD chips. The fact is, AMD clusters different component parts together in CPUs, some of which have shared resources. This is an advancement in technology actually. AMD has at once been ahead of Intel, then Intel took the lead, afterwards AMD pioneered the APU. This is another such advancement from AMD.

At least if an article is going to be posted about, get the thread title right, and keep your opinion separate from that of the article.
 
protocelt said:
So your arguing this lawsuit should be successful on the unrelated basis that you don't like AMD's products because they're bad and not competitive? Not that I am in complete disagreement with your opinion, but I'm not seeing the relation.
Click to expand...

I'm arguing it because there has to be causality behind AMD's poor competitiveness, and perhaps this is one of the keys.
 
I have actually read into some of the more intricate details of the claims. One of the claims made in the lawsuit is false advertisement, and since there is not a lot of room in the subject line, I chose the false advertisement part as that seems to be the crux of the issue at hand. As for having my opinion, I am the OP.... Besides, this case revolves around the definitions of what a thread is and what a CPU core is.

My definition of these are as follows:

Thread: An independent stream of instructions for the CPU to execute which are part of one or more programs.

CPU Core: A unit inside the CPU which can execute a single thread as defined above.
 
Maelstorm said:
I have actually read into some of the more intricate details of the claims. One of the claims made in the lawsuit is false advertisement, and since there is not a lot of room in the subject line, I chose the false advertisement part as that seems to be the crux of the issue at hand. As for having my opinion, I am the OP.... Besides, this case revolves around the definitions of what a thread is and what a CPU core is.

My definition of these are as follows:

Thread: An independent stream of instructions for the CPU to execute which are part of one or more programs.

CPU Core: A unit inside the CPU which can execute a single thread as defined above.
Click to expand...

According to that definition (and pretty much any other definition of a "core"), the AMD CPUs pass muster.

There are two separate integer units (ALUs) that can run two separate integer threads simultaneously. The do share a single fetch/decode unit, though, which has been shown to be a bottleneck, and which has been greatly improved in later versions (Bulldozer --> Steamroller --> Piledriver).

There's also a shared 256-bit capable floating-point unit (FPU) that can run two separate 128-bit FP threads simultaneously (at the same time as the integer threads are running). However, it can only run a single 256-bit FP thread.

So, each dual-core module is capable of running 2 integer and 2 128-bit FP threads simultaneously, making it a dual-core CPU by pretty much any metric. It's only when you try to execute 256-bit FP threads that things drop down to single-threadedness through the single FPU. However, AMD would prefer that developers write HSA-compliant FP tasks that can be run on the GPU cores inside their APUs, as those are much better at running FP tasks than FPUs inside CPUs.

This is a baseless lawsuit that should get thrown out before it ever actually hits a court room. The only people who'd "win" anything from this would be the lawyers.
 
I come from an electronics background (specialty is digital electronics which is the basis of computers), and ALU is the Arithmetic Logic Unit. All it does is perform math/logic operations under the direction of the instruction decoder. So, you are correct in the sense that a module can perform two simultaneous integer threads, but only in regards to math and logic operations, which is only a subset of what programs do. There's a lot more than just math that goes on in a program. You have jumps, loops, and moves (data transfers) which do not use the ALU at all. Those are pretty much fetch, decode, and bus operations. So there is a major bottleneck there if you are trying to do two things at once as the module only has one decode unit, and the system has only one bus. In order to get two true threads running simultaneously on a module, you would need two execution units. Unless AMD has figured out how to multiplex it, which would explain the performance issues.

As for the floating point, what language would be utilized (other than assembler) to take advantage of 256-bit floating point? And what would the application be? In C/C++, a long double is 80-bits (96-bits on a 32-bit machine and 128-bits on a 64-bit machine. This is because machines pull memory in 32/64 bit chunks.) Remember back in the 1980's when Intel first came out with the 80286 and 80386? How long did it take for software to be developed that utilized the capabilities of the new hardware?

Along time.

Now it seems that hardware capabilities are advancing in non-standard directions again. Consider that AMD's 3DNOW is mostly unused unless the compiler (clang/gcc) is given the -march and -mtune options to actually generate the instructions to take advantage of it. Then it's useful only on linear algebra operations. MMX (which has since been superseded by SSE) does the same thing, which is to bring vector processing hardware to desktop computers. Before, only mainframes and supercomputers had dedicated vector processing hardware.

Now moving back on topic, as for this being a baseless lawsuit, I'm not so sure. I'll keep an open mind about it. However, you are correct though: The only people who will get anything out of this will be the lawyers.
 
Intel's Core 2 "quad-core" processors (Q8600, Q9500, etc) from ~5 years ago did exactly the same thing (and AMD's Phenom quad-cores *were* real quad cores at the time)... Strange reversal of roles...

AMD's failings saddens me. I bought a new computer a few weeks ago, and I went for an Intel CPU; it was just the best choice on all fronts except price/performance (and I could spare the extra 150 euro). ARM isn't a direct competitor on the desktop/server front (and probably won't be for quite a while, if ever). A single company making all the chips is *not* good for the free market. Hell, even *two* companies isn't good, but it's still a hell of a lot better than just one.
 
sidetone said:
It's difficult to compete. Cyrix is out of the CPU picture. Maybe AMD has been more successful on GPU's.
Click to expand...

The x86 world went through a round of consolidation a decade or so ago. Centaur and Cyrix were bought up by AMD and VIA (not exactly sure who bought which). Via still makes x86 CPUs, but they're pretty much limited to small form-factor systems where everything is built into the motherboard. They're not really competitive, performance-wise, with Intel CPUs. But they are very inexpensive and low-power. Graphics drivers aren't the greatest, but they make great HTPCs and casual gaming systems.
 
phoenix said:
The x86 world went through a round of consolidation a decade or so ago. Centaur and Cyrix were bought up by AMD and VIA (not exactly sure who bought which). Via still makes x86 CPUs, but they're pretty much limited to small form-factor systems where everything is built into the motherboard. They're not really competitive, performance-wise, with Intel CPUs. But they are very inexpensive and low-power. Graphics drivers aren't the greatest, but they make great HTPCs and casual gaming systems.
Click to expand...

Via's last CPU was from 2011, and even this seems little more than just an update to their 2006 CPU's...
 
You must log in or register to reply here.
Back
Top