Bulldozer Lawsuit

[likeman]

It was properly implemented
Bulldozer recognized as 4c/8t have better performance in games and many applications
It being recognized as 8c/8t would not make any performance improvement in applications which can utilize 8 threads so they made better choice to configure it as 4c/8t

And wasn't this 4c/8t thing done after some Windows path anyway? And wasn't it advertised as path which improve performance for which people actually wanted and waited for?
This itself doesn't mean CPU had just 4 cores, just that the whole idea to name it as such was stupid...

Windows 7 there was an manual patch to make it work more or less correctly, windows 8 and higher made sure it was built in to split them into HT like layout so the task Scheduler would not put the same workload on the same module

Later bulldozer cpus (piledriver type A cpus) ms did not bother to implement the module splitting patch so the 2 modules would show as 4 cores again and just let the task Scheduler to work it out on its own again

It should I've been marketed as core assist so the FX kind of 8 core CPUs should have been sold as 4 core with core assist as it's not smt and it's not HT as they are using split cores (this would of Ment there was no quad core parts in the type A cpus)

 

[XoR_]

I don't think it was risky to call Bulldozer an "8 core CPU" at all. Obviously, had I bet on that I'd have lost. Performance levels and core count aren't necessarily related, so I don't see how that feeds into the perception that it was a quad core. OS tweaks? OS'es need adjustment for new CPU's all the time. Intel's first HT CPUs are an example of this. Next, it was the P6 Microarchitecture that they returned to in order to beat the Athlon 64 and Athlon X2. While Intel called it P6, (Like the Pentium III, itself a form of P6) it wasn't a Pentium 3. Core 2 was vastly different. It also took more than five years of development to make (or revisit) that architecture and Intel stopped trying to push the clock frequencies into 4GHz and beyond by that point and concentrate on efficiency. There are many reasons Intel did this, chief among them was the shift from desktop to mobile systems in the market. They needed something that provided performance per watt, rather than outright performance.

Yes, Pentium M Banias took many years to be built and shares more with Pentium 4 than 3.
Execution units of NetBurst were replaced with design based off Pentium Pro. Optimizations added in Pentium 2 and 3 were result of compromises which they had to make in order to not exceed transistor count budget. Some optimization which Intel added to Pentium 2/3 were no longer relevant eg. fast execution of 16bit software, some were implemented in broken way eg. SIMD instructions, and also some new optimizations would take more effort to add to Pentium 3 than with cleaner design based off simpler base that was properly designed from the start.

Pipeline stages of different Intel processors:
P5 (Pentium) - 5
P6 (Pentium 3) - 10
P6 (Pentium Pro) - 14
NetBurst (Willamette) - 20
NetBurst (Northwood) - 20
NetBurst (Prescott) - 31
NetBurst (Cedar Mill) - 31
Core - 14
Bonnell - 16
Sandy Bridge - 14
Silvermont - 14 to 17
Haswell - 14
Skylake - 14
Kabylake - 14

AMD most probably did exactly the same with their Zen architecture. Jim Keller who also helped to design K8 in the past most probably trashed most changes AMD made up to K10 and extended his own K8 the way it should be done.

It was risky because actually where is definition of a processor core? Dual core processors were first made by gluing two single cores together with some logic (including higher level cache) or even by just putting them on the same package. In the latter case separateness of these cores is pretty much unambiguous =)

AMD Bulldozer does not look like having 8 processing units that can be separated in to 8 working processors if AMD was to make split processor in to separate components and put cores on 8 separate dies. Some components like instruction decoded would need to be duplicated for each core and this is a big problem in case of a lawsuit like the one we are discussing and reason why AMD might in the end loose in the court. Making sure integer execution units are not starved by widening instruction decoder and other components to accommodate increased needed throughput is not the same as making sure you are safe from legal standpoint. Actually it is pretty easily provable that eg. 2m/2t configuration is faster than 1m/2t as even in integer workloads

Again: before multi core processors came out there was no really talk about processors as having any "cores" and because from the start multi core processors were multiple separate processors that you had on one die or processor package (with which there already was some ambiguity as at least some people saw Presler as not "real dual-core"...) it was risky move to call Bulldozer as having twice as much cores as it had modules. Do we have 8 completely separate fully functional processors on AMD Bulldozers?

I do hope AMD win this but I would not put my money on this...

Funnily enough AMD shamelessly copied Intel design with their Threadripper

And if problems got resolved with brilliant solution but this solutions happen to be obstacles you can alway drop to older design... of course with some improvements.

I can clearly see they reverted to old ways of doing things and North Bridge and CPU have now their own houses and are connected with FSB+++++++++++++... and you are supposed to add these pluses up to infinity hence name "Infinity Fabric" =)
BTW. Does this even count as "integrated memory controller" design? Maybe AMD should avoid it calling as such and never state that and be just be silent about it. Like seriously...

 

[tangoseal]

Yes, Pentium M Banias took many years to be built and shares more with Pentium 4 than 3.
Execution units of NetBurst were replaced with design based off Pentium Pro. Optimizations added in Pentium 2 and 3 were result of compromises which they had to make in order to not exceed transistor count budget. Some optimization which Intel added to Pentium 2/3 were no longer relevant eg. fast execution of 16bit software, some were implemented in broken way eg. SIMD instructions, and also some new optimizations would take more effort to add to Pentium 3 than with cleaner design based off simpler base that was properly designed from the start.

Pipeline stages of different Intel processors:
P5 (Pentium) - 5
P6 (Pentium 3) - 10
P6 (Pentium Pro) - 14
NetBurst (Willamette) - 20
NetBurst (Northwood) - 20
NetBurst (Prescott) - 31
NetBurst (Cedar Mill) - 31
Core - 14
Bonnell - 16
Sandy Bridge - 14
Silvermont - 14 to 17
Haswell - 14
Skylake - 14
Kabylake - 14

AMD most probably did exactly the same with their Zen architecture. Jim Keller who also helped to design K8 in the past most probably trashed most changes AMD made up to K10 and extended his own K8 the way it should be done.

It was risky because actually where is definition of a processor core? Dual core processors were first made by gluing two single cores together with some logic (including higher level cache) or even by just putting them on the same package. In the latter case separateness of these cores is pretty much unambiguous =)
View attachment 137760

AMD Bulldozer does not look like having 8 processing units that can be separated in to 8 working processors if AMD was to make split processor in to separate components and put cores on 8 separate dies. Some components like instruction decoded would need to be duplicated for each core and this is a big problem in case of a lawsuit like the one we are discussing and reason why AMD might in the end loose in the court. Making sure integer execution units are not starved by widening instruction decoder and other components to accommodate increased needed throughput is not the same as making sure you are safe from legal standpoint. Actually it is pretty easily provable that eg. 2m/2t configuration is faster than 1m/2t as even in integer workloads
View attachment 137761

Again: before multi core processors came out there was no really talk about processors as having any "cores" and because from the start multi core processors were multiple separate processors that you had on one die or processor package (with which there already was some ambiguity as at least some people saw Presler as not "real dual-core"...) it was risky move to call Bulldozer as having twice as much cores as it had modules. Do we have 8 completely separate fully functional processors on AMD Bulldozers?

I do hope AMD win this but I would not put my money on this...

Funnily enough AMD shamelessly copied Intel design with their Threadripper
View attachment 137762

And if problems got resolved with brilliant solution but this solutions happen to be obstacles you can alway drop to older design... of course with some improvements.
View attachment 137763
I can clearly see they reverted to old ways of doing things and North Bridge and CPU have now their own houses and are connected with FSB+++++++++++++... and you are supposed to add these pluses up to infinity hence name "Infinity Fabric" =)
BTW. Does this even count as "integrated memory controller" design? Maybe AMD should avoid it calling as such and never state that and be just be silent about it. Like seriously...

That's one massive post to say you hate AMD. Would have been easier.

 

[XoR_]

That's one massive post to say you hate AMD. Would have been easier.

I have apparently nothing better to do than hate some tech company, especially one which pushes technology forward (at least from time to time)

If there is no strict technical definition of what "processor core" is or if it doesn't exactly support AMD claims then they can get slapped in their face. That is why I saw claimed Bulldozer core count as very risky and something they should not really do, even if performance was great... which it was not.

Hopefully AMD did their legal research beforehand properly and thoroughly...

 

[juanrga]

If there is no strict technical definition of what "processor core" is or if it doesn't exactly support AMD claims then they can get slapped in their face.

Check #32 for the definition of core used by AMD on Bulldozer. The terminology used by AMD is in full agreement with terminology used in computer science and engineering.

 

[XoR_]

Check #32 for the definition of core used by AMD on Bulldozer. The terminology used by AMD is in full agreement with terminology used in computer science and engineering.

Hopefully you are right and it does agree with agreed upon and written down technical terminology and which definitions are strong enough to succesfully use in court. God bless AMD

Now I will shut up to not say any more of these supposedly hateful things which I could say some more XD

ps. I hate AMD so much I will buy Ryzen 3xxx as soon as it comes out

 

[DuronBurgerMan]

Bulldozer had 8 cores. Yes, yes, shared FPU (that could be split). So what? The 386 was a CPU. It lacked an FPU - instead requiring a separate 387 (or in some cases, a 287). It didn't have "0" cores. It had 1.

Sure, it was a weird design. Still had 8 cores. They were just slow and not well utilized. In the spirit of the thread, I tried to think of a good car analogy for this - but really, they don't work for this. It had 8 cores. Full stop.

I don't get the 'pay a premium' part at all. AMD was generally very cheap in the Bulldozer years. That was the only advantage AMD could really summon with Bulldozer: price.

There are a myriad of things to criticize AMD's marketing moonies on. This isn't one of them.

 

[funkydmunky]

I think how AMD priced these chips in comparison to Intels 8 core CPU's at the time shows they were not being deceitful. Even the 9000 series Bulldozer's when introduced at crazy $$, were nowhere near Intels. As well they were introduced long after the 8000 series Bulldozer's which established the CPU's price performance.