Bulldozer Lawsuit

[bigdogchris]

https://www.techpowerup.com/251758/bulldozer-core-count-debate-comes-back-to-haunt-amd

I have not owned an FX series chip, but I did not realize that the conclusion was to make it at 4c8t chip in Windows to fix the issues with it.

AMD's defense is that people knew what it was when they bought it. I recall that AMD was advertising it as an 8 core chip, not 4 core with SMT.

Also going back to look, the reviews it was being described as an 8C8T chip. If [H] didn't know the difference based on what AMD was saying, how would a normal consumer?

https://www.hardocp.com/article/2011/10/11/amd_bulldozer_fx8150_desktop_performance_review/3

 

[Algrim]

When is a core a core? I guess that lawsuit aims to answer that question.

 

[IdiotInCharge]

When is a core a core? I guess that lawsuit aims to answer that question.

Yup. It's not a question that's easily answered, either. The existence of SMT clouds it quite a bit- when you compare a Bulldozer 'module' with two 'cores' that share an FPU with a single Intel Core or AMD Ryzen core with SMT enabled, you find that total resources, addressable resources, and actual performance track closely between the Dozer module and traditional core.

At the same time, the Dozer module does have two physical cores which are addressable separately, but which share an FPU.

So should 'cores' refer to 'complete cores', i.e. total number of 'blocks' that support all processing functions native to the instruction set, so a four module Dozer has four FPUs and thus has four 'cores'?

 

[bigdogchris]

Explaining modules with perty moving pictures.

 

[nightanole]

Is it 4wd, 8wd, or 12wd?

 

[XoR_]

It was inevitable to happen and even dummy such as myself knew such stupid advertising strategy will backfire on AMD

They basically made broken architecture which in theory was supposed to be good but they did not manage to tweak it to expected performance levels and then it was too late to scrap it and start anew or even to develop Phenom 2 into something better.

Phenom2 was much better base. Still something that was at most like Core2 than even Nehalem (first 4c/8t Core i7 was faster than 6 core Phenom 2) but at least not pathetic and with some tweaking and adding 2 cores they could have pretty decent real 8 core CPU

AMD instead looking at whole picture (Core i7 performance levels with which they competed on the market) they started slapping "Eight-Core" on boxes like only comparison was their earlier Phenom2... which was still better architecture with higher IPC.

Saddest thing from Bulldozer fiasco is that "module" idea wasn't really bad in itself and certainly not reason Bulldozer was slow. It was pretty clever design but rest of core design was flawed and marketing it as 8 core would be risky from legal standpoint even if it was faster than Sandy Bridge let alone when it was slower... I do not believe they did not realize that... but apparently there a lot of idiots in IT industry. Like raise hand anyone who did not know Windows 8 GUI design will be fiasco

I hope AMD win because it would hurt them if they did not and also they had technical reasons to call it 8c despite it being stupid and risky idea. If they loose then who wins? No one except some greedy lawyers

 

[Master_shake_]

When is a core a core? I guess that lawsuit aims to answer that question.

if you read the lawsuit the plaintiff doesn't even know.

 

[Dan_D]

Yup. It's not a question that's easily answered, either. The existence of SMT clouds it quite a bit- when you compare a Bulldozer 'module' with two 'cores' that share an FPU with a single Intel Core or AMD Ryzen core with SMT enabled, you find that total resources, addressable resources, and actual performance track closely between the Dozer module and traditional core.

At the same time, the Dozer module does have two physical cores which are addressable separately, but which share an FPU.

So should 'cores' refer to 'complete cores', i.e. total number of 'blocks' that support all processing functions native to the instruction set, so a four module Dozer has four FPUs and thus has four 'cores'?

This is a tough debate, but back in the day a processor was a processor even when it didn't contain an FPU. The FPU is arguably separate in the x86 architecture as it has been in the past. These days, you wouldn't dream of buying a CPU without one, but at one time they were indeed split from the rest of the design.

 

[nightanole]

This is a tough debate, but back in the day a processor was a processor even when it didn't contain an FPU. The FPU is arguably separate in the x86 architecture as it has been in the past. These days, you wouldn't dream of buying a CPU without one, but at one time they were indeed split from the rest of the design.

https://www.vogons.org/viewtopic.php?f=46&t=63288

Or an option entirely.

 

[IdiotInCharge]

This is a tough debate, but back in the day a processor was a processor even when it didn't contain an FPU. The FPU is arguably separate in the x86 architecture as it has been in the past. These days, you wouldn't dream of buying a CPU without one, but at one time they were indeed split from the rest of the design.

I remember needing a 386DX to play Lucasarts' X-Wing

 

[Dan_D]

https://www.vogons.org/viewtopic.php?f=46&t=63288

Or an option entirely.

That's largely my point. The two weren't always integrated until the Pentium days. There is no list of arbitrary rules (or even technical ones) that say you have to integrate the FPU into the CPU. That said, you can argue that the shared architecture of Bulldozer's design doesn't constitute a full core, but that's a stretch as far as I'm concerned. But, if your argument is that there are only four FPUs for 8 cores, then I think that's a non-starter.

 

[KazeoHin]

As much as I think this Lawsuit is trash, honestly, AMD deserves one hell of a slap on the wrist for Bulldozer....

 

[sover]

As much as I think this Lawsuit is trash, honestly, AMD deserves one hell of a slap on the wrist for Bulldozer....

The years long ass kicking they took in the market was quite enough. This lawsuit should have been bounced. As has already been noted, there isn't a strict definition of what a cpu core is.

 

[Dan_D]

As much as I think this Lawsuit is trash, honestly, AMD deserves one hell of a slap on the wrist for Bulldozer....

They made a bad product and suffered financially for doing so. Not only that, if you bought one given all the reviews out there you got what you deserved. Many people knew what they were getting and were happy enough with their purchase that I don't see this being an issue. This case smacks of someone creating a lawsuit based on greed and a flawed understanding of CPU architecture.

Like the Pentium 4's Netburst microarchitecture, AMD's approach was a gamble that didn't pay off. Everyone who knows anything about this industry knows that CPU architectures from different manufacturers can't necessarily be compared from an architectural standpoint. They have a different bus, platforms and all kinds of other design differences that would make defining just what a CPU core is difficult at best. The last thing you want happening is for people who do not know anything about a product to start crafting laws defining that product from a technical level. If you need proof of this, all you need to do is look at our firearms laws in the U.S. and how ridiculous those are. That's how you end up banning bayonet lugs even though most guns don't have them and no one has been killed by a bayonet on a rifle in the U.S. since the Civil War.

Defining CPU cores could open the door for a lot more frivolous bullshit cases. AMD used to market its Athlon X4 as being a "native quad-core" CPU and trashed Intel for using two dual-core CPU dies in its Core 2 Quad CPU's. With certain arbitrary definitions made by laymen, you could end up with Intel getting sued over a non-native quad-core design that was better than anything the competition had to offer. That doesn't make sense and all that's going to do is make more lawyers even more money. It won't benefit us (consumers) in the least.

 

[Gideon]

It's like people forgot about DX and SX processors. AMD made no attempt to hide that it shared resources so it seems like a pointless lawsuit. But you never know how a court will rule on such things, most judges have little knowledge on such technical things.

 

[Dan_D]

But you never know how a court will rule on such things, most judges have little knowledge on such technical things.

That's precisely my point.

 

[Algrim]

if you read the lawsuit the plaintiff doesn't even know.

It was a tongue-in-cheek question.

 

[bigdogchris]

It's like people forgot about DX and SX processors. AMD made no attempt to hide that it shared resources so it seems like a pointless lawsuit. But you never know how a court will rule on such things, most judges have little knowledge on such technical things.

That's true, it's not like the technical documents were not made available. Had the CPU been properly implemented into Windows, like they are doing now with Windows 10, I think there would have been less problems.

But after reading Dan's post I can see where this type of lawsuit would lead. The case should probably be dropped.

 

[NKD]

the entire goal of this is to get a settlement. That is what the lawyers are going to seek I think. Settle for like a few million and lawyers get paid and everyone gets 10 bucks.

 

[Dan_D]

the entire goal of this is to get a settlement. That is what the lawyers are going to seek I think. Settle for like a few million and lawyers get paid and everyone gets 10 bucks.

AMD should tell them to pound sand.

 

[funkydmunky]

As much as I think this Lawsuit is trash, honestly, AMD deserves one hell of a slap on the wrist for Bulldozer....

Like the late model P4's, AMD had projected the chip to ramp to a much higher scale. It didn't and hasn't since. As with all tech you gamble. Sometimes you win, often you loose. Surprised you think a tech company needs punishment for guessing wrong.
BTW AMD is still struggling to have its core advantage recognized properly by MS. Some things never change

 

[bigdogchris]

The point has been made that AMD has already paid for Bulldozer mistakes.

The hiring of Jim Keller and Zen is your thanks for continued business.

 

[XoR_]

That's true, it's not like the technical documents were not made available. Had the CPU been properly implemented into Windows, like they are doing now with Windows 10, I think there would have been less problems.

But after reading Dan's post I can see where this type of lawsuit would lead. The case should probably be dropped.

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...

 

[bigdogchris]

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...

Well, it's correct now. I am talking about correct from the beginning.

The fact that AMD didn't work with Microsoft to have it run as 4c8T from the beginning kind of shows you that they were hoping it would be a real 8 core chip.

 

[pillagenburn]

From my understanding, Bulldozer has 8 cores..... 8 integer cores... so it's an 8 core if you're doing integer stuff (mostly productivity/office type stuff) and 4 core if you're doing FP like gaming, rendering etc.

If AMD gets nailed on this then I say bullsh*t - Intel has done VASTLY worse things for the consumer.

 

[Dan_D]

From my understanding, Bulldozer has 8 cores..... 8 integer cores... so it's an 8 core if you're doing integer stuff (mostly productivity/office type stuff) and 4 core if you're doing FP like gaming, rendering etc.

If AMD gets nailed on this then I say bullsh*t - Intel has done VASTLY worse things for the consumer.

That's the rub. People think it has 8 incomplete cores because they mistakenly believe that a core should consist of an FPU and there isn't a single reason why you would define a complete core as containing an FPU. For much of the x86 CPU's history, this was NEVER the case or was only the case optionally. I agree with you in that Intel has gotten away with far worse, but companies behaving badly shouldn't be excused either way. That said, AMD never hid the design details about its FPU's from the public.

And for those of you who have been playing this game for as long as I have or longer, are aware that almost every processor maker has had worse FPU performance than Intel for a number of reasons. Bulldozer, for lack of a better term only had to compete with a quad-core CPU with Hyperthreading anyway. There was nothing wrong with that design other than, it didn't work that well with software at the time. Its not illegal to make a product worse than your competition. AMD stopped with its bullshit claims about the processor shortly before launch when they knew they wouldn't hold up. Every erroneous claim it made was also well before the product launched. That whole "subject to change without notice" in the fine print of any published documentation comes in real handy there. This is why it under promised and over delivered with Ryzen and Threadripper.

I agree with the sentiment that AMD suffered enough. There is no sense in beating AMD down any further over a chip that hasn't been relevant for several years now. AMD didn't make much money on its CPU's. People voted with their wallets as they should.

 

[Elf_Boy]

I'd say it has 12 cores.

They are each laid out with their own area on the die, no?

CPUs are still on the market that do not have an fpu. You can still buy 6502 based equipment, hidden in keypads and other places where only a little logic is needed, right?

 

[Legendary Gamer]

The point has been made that AMD has already paid for Bulldozer mistakes.

The hiring of Jim Keller and Zen is your thanks for continued business.

What I don't understand about this is, how in the hell can they win this or expect AMD to settle unless the Judge/ Jury is composed of complete retards... I have to agree, AMD has paid (dearly) for the Bulldozer architecture. The CPU used Modules of a pair of CPU cores that shared resources. They were still individual CPU cores on that same module. No so different from Intel placing two cores on the same module with their older processors, just that AMD stripped a couple components off each core. Threads in BD/EX went one per core... so... No SMT... I would fight to educate the Judge and the Jury so something like this doesn't ever come back to haunt them or any other CPU maker for that matter. This is some of the dumbest shit I've seen in a while.

 

[MMitch]

This lawsuit can only hurt the consumer. Seems staged at this this point euh

 

[jbjmed82]

It was inevitable to happen and even dummy such as myself knew such stupid advertising strategy will backfire on AMD
View attachment 137032
View attachment 137033
They basically made broken architecture which in theory was supposed to be good but they did not manage to tweak it to expected performance levels and then it was too late to scrap it and start anew or even to develop Phenom 2 into something better.

Phenom2 was much better base. Still something that was at most like Core2 than even Nehalem (first 4c/8t Core i7 was faster than 6 core Phenom 2) but at least not pathetic and with some tweaking and adding 2 cores they could have pretty decent real 8 core CPU
View attachment 137034

AMD instead looking at whole picture (Core i7 performance levels with which they competed on the market) they started slapping "Eight-Core" on boxes like only comparison was their earlier Phenom2... which was still better architecture with higher IPC.

Saddest thing from Bulldozer fiasco is that "module" idea wasn't really bad in itself and certainly not reason Bulldozer was slow. It was pretty clever design but rest of core design was flawed and marketing it as 8 core would be risky from legal standpoint even if it was faster than Sandy Bridge let alone when it was slower... I do not believe they did not realize that... but apparently there a lot of idiots in IT industry. Like raise hand anyone who did not know Windows 8 GUI design will be fiasco

I hope AMD win because it would hurt them if they did not and also they had technical reasons to call it 8c despite it being stupid and risky idea. If they loose then who wins? No one except some greedy lawyers

I wonder why they never released a legit 8 core CPU after releasing a 6 core.

 

[XoR_]

I wonder why they never released a legit 8 core CPU after releasing a 6 core.

There was a meeting and teams in AMD presented their plans and one team presented solution A which was more evolutionary, tweak here, tweak there, do not break what is not broken and fix what is there to fix, etc. and other team presented solution B with a more revolutionary approach with lot of cheap high GHz cores and added support for features Intel did not have at the time...

Those who cannot remember the past are condemned to repeat it ~George Santayana

 

[juanrga]

AMD Bulldozer uses the concept of **conjoined cores** introduced in academic literature many years before. Below image shows ordinary core (1) and a pair of conjoined cores (2) sharing L1 caches and the FPU

The original work by Kumar et al published in 2004

https://www.microarch.org/micro37/papers/18_Kumar-Conjoined-Core.pdf

 

[Dan_D]

It was inevitable to happen and even dummy such as myself knew such stupid advertising strategy will backfire on AMD

What "advertising strategy?" AMD really didn't have much advertising for the product beyond sending review samples and slide decks to sites like this one. If you are referring to the shared resources of the execution units in Bulldozer, I think you are way off base. Again, there is no set definition of what constitutes a core.

They basically made broken architecture which in theory was supposed to be good but they did not manage to tweak it to expected performance levels and then it was too late to scrap it and start anew or even to develop Phenom 2 into something better.

The architecture isn't "broken." It works as intended. The problem is that it wasn't as good as anything Intel was doing at the time. AMD banked on the software evolving a certain way towards parallelism which would have shown Bulldozer in a better light. This simply didn't happen. Interestingly enough, AMD wasn't wrong as Bulldozer has aged somewhat better than expected with more modern software. As for tweaking it, there is only so much they could do. Even if you manage to polish a turd, its still a turd. As for developing Phenom II into something better, that was obviously not possible. If AMD could have developed Phenom II into something better than Bulldozer rather than going with a brand new architecture and remained competitive, it would have. It would have been cheaper and faster to market.

Phenom2 was much better base. Still something that was at most like Core2 than even Nehalem (first 4c/8t Core i7 was faster than 6 core Phenom 2) but at least not pathetic and with some tweaking and adding 2 cores they could have pretty decent real 8 core CPU

Phenom II may have had better IPC, but it couldn't scale for shit in terms of clock speeds. It was still slower than Intel's aging Core 2 regarding IPC. It was obvious that the design couldn't be scaled up to make it competitive. You can say Phenom II was a better base, but that doesn't make it true. Being faster than the successor product per clock cycle doesn't make something a better base. Especially when that architecture was already at its limits. AMD had years to refine it and catch up to Core 2 using Phenom as a "base" and couldn't do it. Using it as a "base" instead of Bulldozer would likely have turned into a bigger fiasco than Bulldozer was. AMD took a chance and it didn't pay off because the software industry didn't do what AMD thought it would.

AMD instead looking at whole picture (Core i7 performance levels with which they competed on the market) they started slapping "Eight-Core" on boxes like only comparison was their earlier Phenom2... which was still better architecture with higher IPC.

No, Bulldozer is an eight core CPU. To say otherwise would be to arbitrarily define what a CPU core is. Where does that stop? Who decides that? Ultimately, a judge would have to. How many judges out there are electrical engineers who understand modern semi-conductor design well enough to make that distinction that's both technically correct and in the legal sense? As far as I am concerned, if you want to get basic, integer units are really all that matter. FPU and cache / other shared resources aren't important to the definition of what a core is or isn't. Many of those shared functions such as L3 cache, etc. weren't always even present on CPUs in the past. This is also legal territory that could impact innovation going forward as companies try to meet design demands to match marketing strategies rather than letting the product be what it was meant to be and marketing that product for what it is.

You can sit there and say things like: "A CPU core must contain an integer and an FPU units", but that wouldn't make it true and it opens a Pandora's Box as far as litigation goes. Traditionally, x86 CPUs didn't have FPU's. x87 "MathCo" processors were optional. Eventually, these were integrated together and its a good idea but there is absolutely no reason why this has to be the case. If you make that the case, then it could open a flood gate for litigation around CPUs that never contained FPU capability going back decades. Some older CPUs are still sold today for certain specialized work loads. It could have a dramatic impact today even if those older CPUs were taken out of litigation by some exception clause based on age. Such decisions could impact non-x86 CPUs as well. Its just not something I think should be done as it doesn't benefit the consumer, the legal system or the companies that build these products. If anything, artificially defining a CPU in a way that would penalize AMD would jurt everyone I just mentioned. Only the lawyers benefit from that and I'm not for that. Now, if the legal definition of a CPU core ends up being something that favors AMD in this case, that might not be as bad but it could still be problematic outside of the x86 arena, or could be a problem going forward. That's well into speculative territory but I always side with less laws are better than more laws.

Saddest thing from Bulldozer fiasco is that "module" idea wasn't really bad in itself and certainly not reason Bulldozer was slow. It was pretty clever design but rest of core design was flawed and marketing it as 8 core would be risky from legal standpoint even if it was faster than Sandy Bridge let alone when it was slower... I do not believe they did not realize that... but apparently there a lot of idiots in IT industry. Like raise hand anyone who did not know Windows 8 GUI design will be fiasco

I hope AMD win because it would hurt them if they did not and also they had technical reasons to call it 8c despite it being stupid and risky idea. If they loose then who wins? No one except some greedy lawyers

The design wasn't so much "flawed" as it is slower than the competition's design. It obviously works, just not as fast as we would have liked. I don't think calling it an 8-core CPU was all that risky, but I suppose that turned out to be problematic after all. Again, I think having separate integer units, pipelines etc. are good enough to constitute a core in the legal sense. Things like dedicated L3 cache or an FPU aren't part of the basic x86 CPU requirements and changing that now simply doesn't make sense. As you said, only the lawyers would benefit from such decisions anyway. Fuck those guys.

It's worth noting that Intel took a similar chance with its "Netburst" microarchitecture which was thought to allow scaling to 5GHz and beyond. Intel knew it was worse in terms of IPC, but felt that this disadvantage could be negated by ramping up clock speeds to ridiculous levels. Unfortunately, Intel wasn't capable of doing this and it didn't have enough of a clock speed lead over AMD's much more efficient Athlon 64 and its variants to out perform it in most tasks. A big part of business is understanding where your market is going and predicting how things your product depends on and how it will be used is part of the game. Companies that exist long enough will always reach a point where they fail to do this to some degree. Intel's P4 was a failure in a lot of ways because Intel made a gamble that didn't pay off. AMD did the same thing in almost the exact same way. I don't think this is lawsuit worthy as people had a choice whether they liked it or not. No one had to buy Bulldozer. What it was and how many FPU's it had were well known at the time the first reviews came out.

 

[IdiotInCharge]

I also want to say, I liked AMD's idea with Bulldozer. Essentially a quad-core CPU with hardware symmetric multithreading, it meant that actual multithreaded work, so long as it wasn't extremely float heavy (compute heavy, really), could run with less limitations than a quad-core design that had 'soft' symmetric multithreading support.

Their quad-module Dozers still had the same number of FPUs as every Intel i7 up till the 7700K with the same number of addressable hardware threads.

A great idea that they just couldn't make really work.

 

[CAD4466HK]

Intel's P4 was a failure in a lot of ways because Intel made a gamble that didn't pay off. AMD did the same thing in almost the exact same way. I don't think this is lawsuit worthy as people had a choice whether they liked it or not. No one had to buy Bulldozer. What it was and how many FPU's it had were well known at the time the first reviews came out.

The P4 might of been a failure in the tech world, but in the "real world", it was a marketing success. Church going swap meeting grandmothers knew what a P4 was, even if they only knew it was some whachyoumacallit in a PC. And they wanted one thanks to Intel's viral marketing on all the TV stations and magazines, showing UFO's and aliens proclaiming that Intel was the future. Folks wanted that "Intel Inside" P4 case badge no matter what I told them. It was so bad, that those case badges were sold for $20 a pop at computer shows.

Hell I had gone through a Northwood and a Prescott in my personal box (3.0C & 3.2E), before I switched to an A64. I worked on over a hundred of them in the wild, and maybe a handful of FX's. Shit the A:XP was way more popular than FX.

My point is, if AMD had had a marketing team like Intel did during the P4 era, shit might of just hit the fan instead of clogging it.

 

[XoR_]

@Dan_D
I was not even suggesting Bulldozer did not have 8 cores but that it was risky to call it that, especially given it's performance levels and needed OS tweaks which suggested otherwise. AMD knew all this well but with their financial situation could not afford calling it 4 core CPU anyway... unfortunate bad situation for AMD....

Back in NetBurst days one could give the same argument for Pentium 3 architecture to be no go for Intel and in the end Intel tweaked it to beat AMD's Athlon 64 and even Phenom cores while still with using FSB design. It is rather hard imagine AMD could not do redesign instruction decoder (also add new instructions), make it wider (more execution units), increase HyperTransport speeds, improve branch prediction, etc. and build unambiguous 8 core CPU's that would be not only faster than Bulldozer but also than 4c/8t Core i7 in productivity software. Mind you that Bulldozer is on 32nm node and Phenom2 on 45nm and that change would yield better clocks.

It was not that AMD could not improve Phenom2 but rather that they invested in new architecture and it was hard sell to not follow it after they realized performance is subpar. Unlike Intel they are money limited and cannot work on few arch once and even Intel once they invested in NetBurst they did follow it. Phenom2 didn't have new instructions and by the time AMD realized Bulldozer will be new NetBurst they were simply stuck with it. It is broken because it did not deliver expected performance levels that were promised by its designers in the beginning. It is not like AMD knew this, if they did they would not have build in the first place. AMD expected IPC to be even higher than Phenom2 and module design requiring less transistors. Phenom 2 x6 in 32nm would have been better processor even without any optimizations in the core itself or new instructions which is hard to imagine would ever happen they would go this route. They went ambitious way but failed. In IT industry this is normal, especially when you do not have money to have plan B... or rather plan P XD

 

[XoR_]

The P4 might of been a failure in the tech world, but in the "real world", it was a marketing success. Church going swap meeting grandmothers knew what a P4 was, even if they only knew it was some whachyoumacallit in a PC. And they wanted one thanks to Intel's viral marketing on all the TV stations and magazines, showing UFO's and aliens proclaiming that Intel was the future. Folks wanted that "Intel Inside" P4 case badge no matter what I told them. It was so bad, that those case badges were sold for $20 a pop at computer shows.

Hell I had gone through a Northwood and a Prescott in my personal box (3.0C & 3.2E), before I switched to an A64. I worked on over a hundred of them in the wild, and maybe a handful of FX's. Shit the A:XP was way more popular than FX.

NetBurst had one interesting feature: HyperThreading which improved user experience when using many programs at once. Single core computers had issues with some IO operations requiring interrupts which hogged one core and HT improved on that by allowing programs to run on second thread.

Athlon X2 had issues with Windows XP wh

Of course when Athlon X2 came up they were much better product compared to Pentium D's... except for issues with the design which made programs go wonky on Windows XP. AMD released Dual core optimizer software to alleviate issues which would schedule programs on single core when it detected program is single core only. I personally got Pentium D after using Athlon K7 for this reason and also Pentium D 805 being a sweet deal. I ran mine at 3.8GHz. Cheapest Athlon X2 3600+ was more expensive than Pentium D 805 with motherboard and had zero issues with it being multi-threaded. One could say it was XP that was broken not K8 X2 but somehow they fixed these issues with Phenom and in the end XP users were affected so for end user it was what it was. I even stumbled on this myself when developing program myself. Everything worked perfectly fine on Phenom X4 and of course other computers in the company I worked at the time but on Athlon X2 it would have different clock reading for each core. I also saw issues in games on these processors eg. Painkiller would have augmented physics where objects would randomly fly off to the sky XD Scheduling game to single core would fix it or installing AMD dc optimizer fix which would do it automatically. I am not that much knowledgeable about this issue however and if some BIOS tweaks could fix it (did try fixing in BIOS without success though) or not cause I have never owned X2 processor myself. I only know the issue did exist and made these CPU's less attractive. Of course not many people knew about it and some even thought it is general dual-core processor thing not knowing Pentium D did not have any issues.

Interresting is fact Intel had better CPU's at the time of NetBurst but kept them in mobile only. Pentium M was especially fast in games where it could beat Athlon 64 and at lower TDP. Intel also had option to install L3 case into Pentium 4 which they did in just one instance which was found in Pentium 4 Extreme Edition based on Northwood. Similarly they can now install L4 cache and one CPU with it, Broadwell can at 4.2GHz outperform Skylake at 5GHz despite lower clock speed and IPC difference. They like to keep something "just in case" and not needlessly spend money rising on manufacturing costs even if their products cost a lot compared to how much they cost to make. In that regard Intel might seem greedy but this also shows they are do not come unprepared, at least in this market.

At least AMD with their Zen do seems to be increasingly competitive and push technology forward - eg. forced Intel to increase core count - which bring back memory of Athlon K7 which was pretty much like like Zen is today. Maybe Zen 2 will be like Athlon 64? =)

 

[Dan_D]

@Dan_D
I was not even suggesting Bulldozer did not have 8 cores but that it was risky to call it that, especially given it's performance levels and needed OS tweaks which suggested otherwise. AMD knew all this well but with their financial situation could not afford calling it 4 core CPU anyway... unfortunate bad situation for AMD....

Back in NetBurst days one could give the same argument for Pentium 3 architecture to be no go for Intel and in the end Intel tweaked it to beat AMD's Athlon 64 and even Phenom cores while still with using FSB design. It is rather hard imagine AMD could not do redesign instruction decoder (also add new instructions), make it wider (more execution units), increase HyperTransport speeds, improve branch prediction, etc. and build unambiguous 8 core CPU's that would be not only faster than Bulldozer but also than 4c/8t Core i7 in productivity software. Mind you that Bulldozer is on 32nm node and Phenom2 on 45nm and that change would yield better clocks.

It was not that AMD could not improve Phenom2 but rather that they invested in new architecture and it was hard sell to not follow it after they realized performance is subpar. Unlike Intel they are money limited and cannot work on few arch once and even Intel once they invested in NetBurst they did follow it. Phenom2 didn't have new instructions and by the time AMD realized Bulldozer will be new NetBurst they were simply stuck with it. It is broken because it did not deliver expected performance levels that were promised by its designers in the beginning. It is not like AMD knew this, if they did they would not have build in the first place. AMD expected IPC to be even higher than Phenom2 and module design requiring less transistors. Phenom 2 x6 in 32nm would have been better processor even without any optimizations in the core itself or new instructions which is hard to imagine would ever happen they would go this route. They went ambitious way but failed. In IT industry this is normal, especially when you do not have money to have plan B... or rather plan P XD

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.

I don't think it was a simple matter of AMD being stuck with Bulldozer. It could have returned to a variant of Phenom II had they wished to. I think AMD was at the end of the road for Phenom by that point. Sure, a die shrink would have improved things but AMD knew that Phenom wouldn't be able to compete long term. Again, AMD took several risks with Bulldozer which simply didn't pay off. I think if Phenom II had been a good base to build on we'd see variants of that vs. Zen. Because once again, AMD chose to start over with Zen after their last ground up architecture was deemed an utter failure. It would have been less risky to redevelop Phenom than to go an entirely new direction again with Zen. All that said, I think Bulldozer and AMD's Radeon group's performance are symptoms of a larger problem which is mismanagement. They've let talent walk out the door time and time again and many people have reported less than favorable conditions in the CPU division. This is a company that's sold off profitable divisions of itself to fund an ailing CPU business which AMD has had more bad years with than good.

 

[CAD4466HK]

NetBurst had one interesting feature: HyperThreading which improved user experience when using many programs at once. Single core computers had issues with some IO operations requiring interrupts which hogged one core and HT improved on that by allowing programs to run on second thread.

Athlon X2 had issues with Windows XP wh

Of course when Athlon X2 came up they were much better product compared to Pentium D's... except for issues with the design which made programs go wonky on Windows XP. AMD released Dual core optimizer software to alleviate issues which would schedule programs on single core when it detected program is single core only. I personally got Pentium D after using Athlon K7 for this reason and also Pentium D 805 being a sweet deal. I ran mine at 3.8GHz. Cheapest Athlon X2 3600+ was more expensive than Pentium D 805 with motherboard and had zero issues with it being multi-threaded. One could say it was XP that was broken not K8 X2 but somehow they fixed these issues with Phenom and in the end XP users were affected so for end user it was what it was. I even stumbled on this myself when developing program myself. Everything worked perfectly fine on Phenom X4 and of course other computers in the company I worked at the time but on Athlon X2 it would have different clock reading for each core. I also saw issues in games on these processors eg. Painkiller would have augmented physics where objects would randomly fly off to the sky XD Scheduling game to single core would fix it or installing AMD dc optimizer fix which would do it automatically. I am not that much knowledgeable about this issue however and if some BIOS tweaks could fix it (did try fixing in BIOS without success though) or not cause I have never owned X2 processor myself. I only know the issue did exist and made these CPU's less attractive. Of course not many people knew about it and some even thought it is general dual-core processor thing not knowing Pentium D did not have any issues.

Interresting is fact Intel had better CPU's at the time of NetBurst but kept them in mobile only. Pentium M was especially fast in games where it could beat Athlon 64 and at lower TDP. Intel also had option to install L3 case into Pentium 4 which they did in just one instance which was found in Pentium 4 Extreme Edition based on Northwood. Similarly they can now install L4 cache and one CPU with it, Broadwell can at 4.2GHz outperform Skylake at 5GHz despite lower clock speed and IPC difference. They like to keep something "just in case" and not needlessly spend money rising on manufacturing costs even if their products cost a lot compared to how much they cost to make. In that regard Intel might seem greedy but this also shows they are do not come unprepared, at least in this market.

At least AMD with their Zen do seems to be increasingly competitive and push technology forward - eg. forced Intel to increase core count - which bring back memory of Athlon K7 which was pretty much like like Zen is today. Maybe Zen 2 will be like Athlon 64? =)

Smithfield & Presler were a no go for me, not while I had a San Diego 3700+ @ 2750, 2833 suicide paired with a K8N Diamond Plus and DDR500 in the house.

I later went with a Toledo 4200+ X2 on the same mobo. AMD was kicking ass and taking names back then, multi-threading issues be damned.

 

[Nobu]

Bottom line is it was a different architecture. The 8 core design was different, and favored different things than modern processors traditionally (whoo oxymoron!) favored, which meant it needed software optimizations to perform it's best. The problem is it needed more work than they thought in order for it to perform to users' expectations, who for some strange reason thought double cores means double performance regardless of the underlying architecture (and maybe regardless of usecase, I dunno...haven't read the plaintiff's claims so I don't know the level of naivety, or else ignorance, they have). In any case, I thought AMD did a decent job of laying out the architecture details, though maybe they didn't explain the consequences of their design too well, I don't feel like they should have to do that when there are plenty of knowledgeable people who you could ask if you were a journalist, who could have told you.