Ex-AMD Engineer explains Bulldozer fiasco

[aigomorla]

Maier saying that if AMD gets bought out it loses its x86 license, which is non-transferrable. Unless I misunderstood what he was saying but I don't think I did.

yeah i heard this as well....

The license is non transferable, and in a buyout new negotiations would need to commence.

And thats basically intel going.. so hello mr(s). rich person, how much are you willing to pay?

 

[bryanW1995]

I seem to remember having an indepth discussion with CTho when this originally came out, where he was saying that current tools have shown that they are usually more efficient compared to hand designs, and are also usually more compact. He showed examples, and gave the reasoning (since there have been years upon years of designing circuitry, the tools are able to use multiple tricks found by people throughout that time to reduce size, latency, interference, and power.) Since he is the only actual chip designer I know, I took his word for it, and still don't think that this argument holds a lot of water.

If I can find the string of comments, I will link them here.

EDIT: I spent about an hour looking for the comments, and I really don't have anymore time to spend on that task (I shouldn't have spent that much time in the first place, but I got caught up in my task.) I remember the conversation, but I am starting to think that it may have been Hardball and not CTho who it was with. Regardless, most of Intel's designs have been automated for years, with some hand tweaking, and I would expect AMD to do something similar. Hell, Brazos was marketed as the first fully automated design, and it did extremely well for the size and power budget.

I remember reading that thread, it was extremely interesting. I think that it was CTho. I'll PM him if he doesn't show up on this thread.

 

[EightySix Four]

On the topic of AMD being bought, I found it interesting, when looking over the thread containing Cliff Maier's comments about AMD's general suckitude, was Maier saying that if AMD gets bought out it loses its x86 license, which is non-transferrable. Unless I misunderstood what he was saying but I don't think I did.

No, you're correct. AMD has to remain it's own corporation, but I do believe investment firms could "purchase" AMD as long as it remained AMD and was not merged with another tech company. To me, this is the most likely course of action, purchase AMD, replace the management and hire back some of the old engineers they have lost over time (if possible).

I can also imagine that the U.S. government would get pretty involved if AMD were about to go under. Intel having a de facto monopoly would be bad.

 

[Vdubchaos]

No, you're correct. AMD has to remain it's own corporation, but I do believe investment firms could "purchase" AMD as long as it remained AMD and was not merged with another tech company. To me, this is the most likely course of action, purchase AMD, replace the management and hire back some of the old engineers they have lost over time (if possible).

You are assuming that these companies that buy others have common sense and do above sort of thing.

they DON'T

Usually when companies merge you see most of the people of the swallowed up company go....QUICK

 

[frostedflakes]

On the topic of AMD being bought, I found it interesting, when looking over the thread containing Cliff Maier's comments about AMD's general suckitude, was Maier saying that if AMD gets bought out it loses its x86 license, which is non-transferrable. Unless I misunderstood what he was saying but I don't think I did.

I'm sure they would work something out. Intel has become kind of dependent on AMD because they cross-license some technologies like x86-64. Until the patents on that runs out, I don't think Intel could afford to pull the x86 license if someone bought them out, because AMD could just as easily tell them to stop manufacturing x86-64 compatible CPUs. They'd waste a year or two fighting in court and eventually come to an agreement that is mutually beneficial to both parties and allows both to continue making CPUs, that's what almost always happens with these patent disputes between large companies. Just like what happened with the GlobalFoundries spinoff a few years ago, Intel threatened to pull their x86 license but of course that never happened. Seems like it was just a ploy to extend the cross-licensing agreement with AMD, which was about to expire.

 

[EightySix Four]

You are assuming that these companies that buy others have common sense and do above sort of thing.

they DON'T

Usually when companies merge you see most of the people of the swallowed up company go....QUICK

Oh I understand that's not how it normally works. Generally when looking to purchase a failing company investment firms look at strength of management first. If the management is not strong then there has to be some other extremely valuable peice in play. In this case, it's AMD's x86 license.

 

[aigomorla]

Itanium~

Oh god.. no... please no...

I like A64....

 

[bryanW1995]

I don't see how the company is worth even $4 billion or whatever the market cap is these days. Regardless of "why", AMD's fortunes have been on the decline since the day they bought ATI. I like their gpus a lot, but considering the value of ATI (or the amd graphics division if you prefer), I don't see how the it can be worth very much money when they spend Billions in research to make ~ $50 million a year, and that's an improvement over what it was doing a few years ago. The business model just doesn't make sense.

 

[intangir]

So if BD was largely a synthesized design, why was it delayed so much?

Synthesis only speeds the physical design. It doesn't help with writing the RTL (the input to the synthesis tools), nor does it help to validate that the RTL model of the microarchitecture is functionally correct and achieves the performance needed.

On the other hand, signs don't really point to functional bugs. Bulldozer does fall short in terms of clockspeed, which I attribute mainly to the process, with the use of design automation tools a relatively minor contributor. After all, the design does clock to 8.4 GHz; it just requires an impractical amount of wattage to get there.

In terms of IPC, there's doesn't appear to be a single part of the design that can be pinpointed as choking the rest, which is what would appear if parts of the chip were disabled because of bugs, and that seems to be reinforced by AMD's expectation of only ~5% IPC uplift in Piledriver, with the other ~5-10% performance from power improvements that let them clock higher.

It seems to me the overall microarchitecture is just badly designed, which suggests that something is fundamentally broken in AMD's performance modeling, and caused them to fall ~10-15% short from their IPC goal of matching the Phenom II.

 

[pm]

Damn, just damn. I'm sorry I tried to ever make a little joke, which IMO is being indirectly done ALL OVER the CPU thread. Honestly, I think the talks about Bulldozer sucking or whatnot became old after a day or so, but it rages on.

Sheesh.

Sorry, I'm perhaps too quick to jump to my friend's defense, and perhaps I lack basic reading comprehension skills as well. I didn't see the humor of it - it has been said that I lack a sense of humor. My over-reaction is for a good cause - CTho is a talented engineer and a good friend of mine.

 

[EightySix Four]

I'm sure they would work something out. Intel has become kind of dependent on AMD because they cross-license some technologies like x86-64. Until the patents on that runs out, I don't think Intel could afford to pull the x86 license if someone bought them out, because AMD could just as easily tell them to stop manufacturing x86-64 compatible CPUs. They'd waste a year or two fighting in court and eventually come to an agreement that is mutually beneficial to both parties and allows both to continue making CPUs, that's what almost always happens with these patent disputes between large companies. Just like what happened with the GlobalFoundries spinoff a few years ago, Intel threatened to pull their x86 license but of course that never happened. Seems like it was just a ploy to extend the cross-licensing agreement with AMD, which was about to expire.

The following quote holds true for x86 and AMD64:

However, the agreement provides that if one party breaches the agreement, it loses all rights to the other party's technology, while the other party receives perpetual rights to all licensed technology.

If AMD was purchased by another company, depending on the licensing agreements details, that could be a breach of agreement and Intel would have perpetual rights.

 

[frostedflakes]

This holds true for x86 and AMD64:

Is there really a clause like that in the contract? D:

Can't believe AMD would have agreed to that. Really limits their options.

edit: Looks like their 2000 cross-licensing agreement had a similar clause, but the legal disputes between Intel and AMD in 2009 didn't result in AMD losing x86 and Intel getting x86-64. Legalese makes my brain hurt, though, so maybe I'm misinterpreting it.

http://contracts.corporate.findlaw.com/operations/ip/802.html

 

[alyarb]

that could be a breach of agreement and Intel would have perpetual rights.

If, one day years from now, Intel was run by foreign investors instead of domestic engineers, the decline would be similar to AMD's.

 

[bryanW1995]

Oh I understand that's not how it normally works. Generally when looking to purchase a failing company investment firms look at strength of management first. If the management is not strong then there has to be some other extremely valuable peice in play. In this case, it's AMD's x86 license.

To whom is that license most valuable? IBM? Samsung? Maybe. What about JHH? If he could somehow get the gpu combination past the regulators then NV would own high end gaming indefinitely. Even if he had to spin off/sell ATI, he could make sure that it went to somebody that wouldn't give it to intel. I would bet that NV has a strike plan in place for AMD if it gets cheap enough.

 

[bryanW1995]

The following quote holds true for x86 and AMD64:



If AMD was purchased by another company, depending on the licensing agreements details, that could be a breach of agreement and Intel would have perpetual rights.

That is pure speculation on your part, there is no way that AMD would have agreed to that "licensing agreement detail". Their board has to have seen the writing on the wall for years, why would they prevent themselves from taking some sucker's money in the future?

 

[quest55720]

If you divide 2 billion transistors by 315 mm^2, you get about 6.35. If you divide 450 million transistors by 75, you get about 6. So Zacate is actually less dense than Zambezi (TSMC vs GloFo). This means if this "20% bigger" accusation turns out to be remotely true, then it will have some implications for Zacate and presumably Llano as well.

You would have to compare a pure bobcat with out the GPU for a fair comparison. But I can't imagine the bobcat being 20% smaller and more powerful. I honestly think it is an engineer trying to cover up his failure by blaming the bean counters. The bobcat is the complete opposite of the bulldozer. It is a winner in every single way and beats its direct competition in basically everything.

 

[EightySix Four]

To whom is that license most valuable? IBM? Samsung? Maybe. What about JHH? If he could somehow get the gpu combination past the regulators then NV would own high end gaming indefinitely. Even if he had to spin off/sell ATI, he could make sure that it went to somebody that wouldn't give it to intel. I would bet that NV has a strike plan in place for AMD if it gets cheap enough.

IBM has a really crazy x86 license. They can fab any x86 chip they want, and a third party can come to them and fab an x86 chip even if they don't have an x86 license. My understanding is IBM's fabs are more expensive than most other fabs though so it hasn't been worth it. I'm a tiny bit suprised that IBM hasn't had interest in buying out Global foundries yet though.

nVidia would make for an interesting acquisition, but Intel would need to become a serious player in the graphics industry before that could ever pass regulation unless ATi could be spun off (which is highly unlikely considering every CPU under development by AMD use ATi GPUs in them).

Most likely you'd see a large cap investment firm begin to take interest in a takeover/buyout as AMD's market cap continues to plummet. This would solve a lot of licensing issues as AMD would still remain it's own entity. Unfortunately most firms that would have interest right now are foreign and again the regulatory issues would be huge.

AMD is a really really nasty position right now, they need to hit one out of the park in the next 2-3 years, if they can hold on that long. The good news is, AMD is actually well positioned in the PC segments which are going to grow in the near future with Llano/Bobcat and their future upgrades. If they can get the power envelope down enough that they can take on the ARM based tablets with Windows 8 next year, I feel strongly they can limp along.

We also really need to see Bulldozer performance on server and cloud workloads. Some of the workstation benchmarks have actually been pretty impressive.

That is pure speculation on your part, there is no way that AMD would have agreed to that "licensing agreement detail". Their board has to have seen the writing on the wall for years, why would they prevent themselves from taking some sucker's money in the future?

Yes, it was speculation, but look at the position the clause stating the x86 license can't be transferred if AMD is sold puts them in. Do you really think AMD has been making great business decisions lately? I'm actually really curious about this, and if anyone knows the agreement in-depth I'd like to hear about it. If not, I'll look into it myself on Sunday.

Honestly, when the agreement with Intel was signed allowing Intel to use x86-64, AMD needed Intel to support it so it would gain traction. I'd really like to see those two both granted perpetual rights to both, because they market needs it.

If, one day years from now, Intel was run by foreign investors instead of domestic engineers, the decline would be similar to AMD's.

If some of the rumors about the way AMD is run are true, then AMD has a severe disconnect between management and engineering. This disconnect is incredibly dangerous and you can see the results of a similar issue within the auto industry. AMD needs a fantastic CEO with a great engineering sense to run the company right now, I don't believe they have that. The shortsightedness is painful, I know they needed cash but selling off the mobile division at the time that they did was just a complete misread of the market.

 

[EightySix Four]

Any lawyers want to go over section 6.2? It seems like the transference of AMD property to a receiver is cause for agreement termination, meaning Intel perpetually has rights to x86-64 and AMD loses rights to X86.

http://contracts.corporate.findlaw.com/operations/ip/802.html

 

[HNNstyle]

cmaier in this thread is former AMD engineer Cliff Maier

http://forums.macrumors.com/showthread.php?t=898409

Looks like all good engineers at AMD are gone.

Very interesting read. Seems like cmaier was right on! That's just too bad though, we consumers need better competition.

 

[Idontcare]

Any lawyers want to go over section 6.2? It seems like the transference of AMD property to a receiver is cause for agreement termination, meaning Intel perpetually has rights to x86-64 and AMD loses rights to X86.

http://contracts.corporate.findlaw.com/operations/ip/802.html

I'm no lawyer, but this is as I understand it to be.

Synthesis only speeds the physical design. It doesn't help with writing the RTL (the input to the synthesis tools), nor does it help to validate that the RTL model of the microarchitecture is functionally correct and achieves the performance needed.

On the other hand, signs don't really point to functional bugs. Bulldozer does fall short in terms of clockspeed, which I attribute mainly to the process, with the use of design automation tools a relatively minor contributor. After all, the design does clock to 8.4 GHz; it just requires an impractical amount of wattage to get there.

In terms of IPC, there's doesn't appear to be a single part of the design that can be pinpointed as choking the rest, which is what would appear if parts of the chip were disabled because of bugs, and that seems to be reinforced by AMD's expectation of only ~5% IPC uplift in Piledriver, with the other ~5-10% performance from power improvements that let them clock higher.

It seems to me the overall microarchitecture is just badly designed, which suggests that something is fundamentally broken in AMD's performance modeling, and caused them to fall ~10-15% short from their IPC goal of matching the Phenom II.

According to DrWho (Francois) on XS, the powerpoint slide claims regarding the architecture features of Bulldozer don't match up to reality.

http://www.xtremesystems.org/forums...nally-tested&p=4972103&viewfull=1#post4972103

http://www.xtremesystems.org/forums...nally-tested&p=4972367&viewfull=1#post4972367

http://www.xtremesystems.org/forums...nally-tested&p=4972442&viewfull=1#post4972442

This is relevant when we are attempting to lazy-boy our design fixes, we can't assume the architecture is as has been presented in powerpoint.

for the tl;dr crowd - Francois is basically saying he analyzed the functionality of bulldozer with code designed to tease out the details of the microarchitecture and the integer cores are infact decoder limited, they aren't effectively sharing decoders.

So if AMD wanted to address the IPC issue it would seem they need to address the decoders.

Does anyone know if the bobcat was auto designed or hand tuned? Because if it was auto designed he is dead wrong and making excuses that it was not the engineers but the bean counters.

Remember this:

You get that kind of a floor plan layout when using synthesis tools, or so I am told. (I'm not a design guy)

Also, unless I am mistaken, Intel's iGPU is also heavily synthesized.

 

[IntelUser2000]

You get that kind of a floor plan layout when using synthesis tools, or so I am told. (I'm not a design guy)

Also, unless I am mistaken, Intel's iGPU is also heavily synthesized.

Intel started using automated tools with Prescott. Ok, bad example because that chip underperformed too, but it doesn't mean Intel doesn't do it.

I think what the synthesis tools give in terms of layout is more of a symmetrical layout rather than being more random. Makes sense. That's why GPUs work well on it because its based on largely repeated structures.

 

[RampantAndroid]

yeah i heard this as well....

The license is non transferable, and in a buyout new negotiations would need to commence.

And thats basically intel going.. so hello mr(s). rich person, how much are you willing to pay?

Not totally true. Intel licenses AMD64...so Intel has something to lose.

 

[dmens]

Synthesis only speeds the physical design. It doesn't help with writing the RTL (the input to the synthesis tools), nor does it help to validate that the RTL model of the microarchitecture is functionally correct and achieves the performance needed.

On the other hand, signs don't really point to functional bugs. Bulldozer does fall short in terms of clockspeed, which I attribute mainly to the process, with the use of design automation tools a relatively minor contributor. After all, the design does clock to 8.4 GHz; it just requires an impractical amount of wattage to get there.

In terms of IPC, there's doesn't appear to be a single part of the design that can be pinpointed as choking the rest, which is what would appear if parts of the chip were disabled because of bugs, and that seems to be reinforced by AMD's expectation of only ~5% IPC uplift in Piledriver, with the other ~5-10% performance from power improvements that let them clock higher.

It seems to me the overall microarchitecture is just badly designed, which suggests that something is fundamentally broken in AMD's performance modeling, and caused them to fall ~10-15% short from their IPC goal of matching the Phenom II.

That's my point, blaming synthesized design for lack of performance is not right if the uarch is bottlenecked or poorly designed.

Besides, I find it hard to believe that AMD did not slice up the design and stick with hand design on the most critical portions.

 

[dmens]

You get that kind of a floor plan layout when using synthesis tools, or so I am told. (I'm not a design guy)

Also, unless I am mistaken, Intel's iGPU is also heavily synthesized.

You are correct on both.

 

[IlllI]

hopefully that guy amd hired from lenovo can help turn things around