Samsung Nintendo AMD semicustom ARM-x86 Showdown Scenario

[The Alias]

And I am sure they said the same thing about Bulldozer.

Assuming a big improvement is just as big a stretch as the naysayers who say it has to be terrible because of AMDs lack of resources. Bottom line is nobody really knows, and nobody even knows if it will come out on time.

I felt with bulldozer they were trying to design for a future that wasn't really going to come. With this one I feel they will be more accurate in terms of the applications they're designing for, but you're right. No one really knows

 

[The Alias]

$19 billion, how cute! Intel spent, what? Nearly $12 billion in R&D last year?

And don't think for a second that the Chinese Academy of Sciences will hand AMD its entire $19 billion "war-chest" to build server chips.

Intel is really different . they have fab tech which absolutely devours tech. Especially now with nodes coming at a hell of a premium. They have their cellphone chips which they're playing catch up with along with their modems. They have ssd's they have to fund. They have the desktop architecture, and they have GPUs they have to develop.

Amd only has zen,cat cores(maybe), arm, and graphics. And they all bleed into eachother.

 

[DrMrLordX]

$19 billion, how cute! Intel spent, what? Nearly $12 billion in R&D last year?

Intel spends a boatload on fab-related R&D. AMD needn't concern itself with that. Besides, the R&D budget of some other company is not germane to the discussion at hand: the claim has been made that AMD's R&D budget is shrinking. The Chicoms are changing that situation, for better or worse.

And don't think for a second that the Chinese Academy of Sciences will hand AMD its entire $19 billion "war-chest" to build server chips.

Honestly I don't think AMD would spend $19 billion on R&D in one year or in one go. If they got that kind of money in their hands all at once, I'll bet the first thing they'd do would be retire debt.

 

[Idontcare]

Intel spends a boatload on fab-related R&D. AMD needn't concern itself with that. Besides, the R&D budget of some other company is not germane to the discussion at hand: the claim has been made that AMD's R&D budget is shrinking.

Its not just a "claim," it is established fact per AMD's own financial filings.

The Chicoms are changing that situation...

That is merely a claim, credited to a rumor site.

China hasn't confirmed, and AMD has yet to mention anything about it (which they would need to as it is a materially changing event).

I know you aren't the one making the claim, but follow the purported chain of evidence and you quickly end up in click-bait rumor-article territory.

Honestly I don't think AMD would spend $19 billion on R&D in one year or in one go. If they got that kind of money in their hands all at once, I'll bet the first thing they'd do would be retire debt.

No one smart enough to amass $19B is going to be daft enough to hand it over to AMD. Think about it.

 

[frozentundra123456]

It doesnt say how much money is targeted to amd, just that the "war chest" is 19 billion. If you read the entire article, it also says the deal is not even finalized yet. At this stage of development it may not have much effect on zen.

 

[mrmt]

Why the hell not? That's exactly what people do for GPU's -- If you want to play games well in 4K at high detail, good luck doing it with a single video card. SLI or Crossfire is pretty much the only way to get there.

Well, people do that for GPU to get more performance than they get with single GPUs. AMD gave people more cores and more transistors but didn't take them to new levels of performance. That's the difference.

 

[MiddleOfTheRoad]

Well, people do that for GPU to get more performance than they get with single GPUs. AMD gave people more cores and more transistors but didn't take them to new levels of performance. That's the difference.

But again -- it was the execution that failed them. The design concept is sound. It's a lot like Dodge putting a V-10 in the Viper to make more horsepower than the Corvette -- the Corvette has never been able to generate as much power as a Viper with its naturally aspirated V8.... GM needed to supercharge (akin to overclocking) to get it to make similar horsepower to the Viper.

It also depends on the application. In heavily multi-threaded applications for example (video rendering or integer calculations), AMD did successfully give people new levels of performance with Bulldozer. But most apps rely on fast single threaded performance which is exactly why Bulldozer earned it poor reputation.

 

[mrmt]

But again -- it was the execution that failed them. The design concept is sound. It's a lot like Dodge putting a V-10 in the Viper to make more horsepower than the Corvette -- the Corvette has never been able to generate as much power as a Viper with its naturally aspirated V8.... GM needed to supercharge (akin to overclocking) to get it to make similar horsepower to the Viper.

How is bulldozer a sound concept? Is there any other successful implementation of the concept? What successful CMT products exists on the market?

It also depends on the application. In heavily multi-threaded applications for example (video rendering or integer calculations), AMD did successfully give people new levels of performance with Bulldozer. But most apps rely on fast single threaded performance which is exactly why Bulldozer earned it poor reputation.

It didn't, Intel was fielding 6C and 8C chips that mopped the floor with Bulldozer chips in these multithreaded apps by the time around its launch. The fact that AMD had to sell its chips for penauts is more a symptom of Bulldozer failure, not an engineering feat from AMD.

 

[Enigmoid]

But again -- it was the execution that failed them. The design concept is sound.

It also depends on the application. In heavily multi-threaded applications for example (video rendering or integer calculations), AMD did successfully give people new levels of performance with Bulldozer. But most apps rely on fast single threaded performance which is exactly why Bulldozer earned it poor reputation.

One could argue that a design concept that cannot be implemented as conceived is a terrible design concept.

I have this design concept for giant floating cities a la Bioshock. I have meticulously planned out every detail.

It really doesn't matter how well its planned out. Its implementation is unfeasible.

A design that you cannot execute is not a good design from a practical point of view.

Bulldozer did no better than an 8 core die shrunk PII would have done (with AVX and newer features). On servers it was noncompetitive with Mangy-Cours; similar performance but higher power consumption.

 

[Gikaseixas]

We are way off topic now. Not worth beating that horse any further

So who might be that customer? Apple, Samsung, Mediatek or Nintendo??? Or is this utter nonsense?

 

[DrMrLordX]

Its not just a "claim," it is established fact per AMD's own financial filings.

That was then. This is now . . .

That is merely a claim, credited to a rumor site.

We'll find out in a year or two whether or not AMD's palms are getting greased by the Chinese gub'ment depending on the when (or if) there is start to the movement of cash. It seems credible, but yeah, technically, nothing has changed until money changes hands.

China hasn't confirmed, and AMD has yet to mention anything about it (which they would need to as it is a materially changing event).

Right, it'd show up in a quarterly report, wouldn't it?

I know you aren't the one making the claim, but follow the purported chain of evidence and you quickly end up in click-bait rumor-article territory.

No one smart enough to amass $19B is going to be daft enough to hand it over to AMD. Think about it.

Look at how much money the Chinese gubment has thrown at their MIPs knock-offs, though. What has that availed them? Not much. Compared to what they have now, making a grab at AMD's IP might actually make sense, not that it says much good about AMD's future. Being a puppet of the Chicoms is not always a net positive.

Besides, Mubadala already threw a bunch of money at AMD. It's happened before, and it may well be happening again . . . not that I expect AMD to wind up with a guarantee of $19B or anything of the sort, even bled out over a period of time. As I indicated before, I do not fully expect the Chicoms to just hand over the entire "warchest". They probably have a number of acquisitions and other manipulations to fund from that cash pile.

So who might be that customer? Apple, Samsung, Mediatek or Nintendo??? Or is this utter nonsense?

Kumar may be referring to multiple contracts equal to "over one billion" in total value. It could be several customers. Mediatek and Nintendo seem probable candidates, especially since we've heard other news about Mediatek licensing GCN tech.

 

[ARMD Supremacy]

Assuming the other unnamed semicustom is x86, I imagine it will probably be for a server client.

I honestly think Apple is a more likely semi-custom customer than Samsung. Samsung already exited the laptop business, so there's not much reason to get a x86 semicustom. Apple, I don't really see it unless they're expecting AMD's Zen core to be powerful and crazy efficient, but still more likely than Samsung at least. They have more than enough money, and more importantly, they have the infrastructure to justify commissioning a semi-custom chip.

Ask yourself about reasons why OEMs like Samsung are rethinking mobile and PC business?

One reason is differentiation and with mainly Intel providing reference solutions based on their platform, 99% of mobile and PC are like the same fast foods from Mcdonalds!

Another reason that Samsung is asking is why pay Intel so much for their CPU/GPU especially Iris Pros if it is now possible to make superior 14nm semicustom solutions that can be made in Samsungs factories?

Plus add new and superior performance boosts and form factors made possible with HBM technology (also from Samsung), then it answers why Samsung will follow Microsoft, Sony, Nintendo and other major companies for a semicustom solution from AMD.

 

[geoxile]

Ask yourself about reasons why OEMs like Samsung are rethinking mobile and PC business?

One reason is differentiation and with mainly Intel providing reference solutions based on their platform, 99% of mobile and PC are like the same fast foods from Mcdonalds!

Another reason that Samsung is asking is why pay Intel so much for their CPU/GPU especially Iris Pros if it is now possible to make superior 14nm semicustom solutions that can be made in Samsungs factories?

Plus add new and superior performance boosts and form factors made possible with HBM technology (also from Samsung), then it answers why Samsung will follow Microsoft, Sony, Nintendo and other major companies for a semicustom solution from AMD.

Hey, maybe it's Microsoft. My point is, Samsung already got rid of their infrastructure. It's unlikely to be them.

For a Surface Pro tablet any recent CPU performance is probably good enough, so all it needs is higher efficiency and better GPU performance.

 

[mrmt]

Right, it'd show up in a quarterly report, wouldn't it?

A small deal yes, it will appear in the quarterly Q&A. A big deal would need a separate filling and Q&A.

Besides, Mubadala already threw a bunch of money at AMD.

Mubalada threw enough money to keep AMD alive, but not to prop up the company. They are not a valid source of funding for AMD.

Another reason that Samsung is asking is why pay Intel so much for their CPU/GPU especially Iris Pros if it is now possible to make superior 14nm semicustom solutions that can be made in Samsungs factories?

What superior semi-custom solutions? AMD subpar solutions?

 

[Gikaseixas]

Question to all: Samsung's 14nm + AMD expertise could offer a proper alternative to leading arch's no? After all, GF is to blame for many missteps lately.

Are scenarios such as Samsung opening up it's factories for someone like AMD who's clearly in great need, impossible?

 

[mrmt]

Question to all: Samsung's 14nm + AMD expertise could offer a proper alternative to leading arch's no? After all, GF is to blame for many missteps lately.

AMD designs are still subpar, the lagging node is just the icing on the cake. Using Samsung node might make AMD products better in terms of performance and costs, but it is still a dog in terms of design.

 

[ShintaiDK]

Samsungs 14nm is low power. And you cant blame GF for all of AMDs bad design mistakes and limited R&D.

 

[NTMBK]

AMD designs are still subpar, the lagging node is just the icing on the cake. Using Samsung node might make AMD products better in terms of performance and costs, but it is still a dog in terms of design.

Oh, has there been some in depth disclosure of the Zen architecture that I missed? Or are you just pulling that fact out of thin air?

Yes, Bulldozer was a poor design. But he was discussing 14nm designs, and AMD would have to be insane to port a Bulldozer derivative down to 14nm.

 

[mrmt]

Oh, has there been some in depth disclosure of the Zen architecture that I missed? Or are you just pulling that fact out of thin air?

No, there was a huge R&D budget drop in the last few years that you seem to have missed and this alone will make impossible for AMD to develop a competitive, high performance architecture.

 

[NTMBK]

No, there was a huge R&D budget drop in the last few years that you seem to have missed and this alone will make impossible for AMD to develop a competitive, high performance architecture.

There you go again talking in absolutes. 'Impossible'? No. Highly unlikely, yes. But hey, maybe if K12 is cancelled then more resources went into Zen. And if it is a replacement for both Cat and Construction cores, it's one less way to split their R&D funds.

 

[mrmt]

There you go again talking in absolutes. 'Impossible'? No. Highly unlikely, yes. But hey, maybe if K12 is cancelled then more resources went into Zen. And if it is a replacement for both Cat and Construction cores, it's one less way to split their R&D funds.

We are talking about a company that spends less than Nvidia spends on their GPU line and Tegra. And you expect them to field a competitive x86 uarch, a competitive ARM architecture *and* a competitive GPU architecture? Yes, this is impossible in my book, especially for a company with the track record of incompetent management AMD has.


Ed: Given the R&D crunch of the last four years, I wouldn't be surprised if AMD ends up with a subpar x86 architecture, a subpar ARM architecture *and* a subpar GPU architecture. 2 out of three is pretty much warranted.

 

[DrMrLordX]

A small deal yes, it will appear in the quarterly Q&A. A big deal would need a separate filling and Q&A.

Okay. So how big of a deal would it need to be to require a separate filing?

Mubalada threw enough money to keep AMD alive, but not to prop up the company. They are not a valid source of funding for AMD.

Not my point though . . . Mubadala kept them alive. What makes you think the Chicoms will do anything different?

Samsungs 14nm is low power. And you cant blame GF for all of AMDs bad design mistakes and limited R&D.

Samsung's current 14nm process is advertised as being low-power, not that any of us really know anything about it because they haven't sold any commercial products based on it yet (that I know of). Isn't Samsung in the habit of producing two processes based on the same basic node, usually one low-power and one high(er) performance?

AMD designs are still subpar, the lagging node is just the icing on the cake. Using Samsung node might make AMD products better in terms of performance and costs, but it is still a dog in terms of design.

Designs for what? Their GCN designs are actually quite good. It's not like Samsung has to license Steamroller or Excavator, it could be they want graphics IP for products. Besides, people are constantly bashing Steamroller for I-don't-know-what reason. It's pretty good, and would have been much better had AMD had a suitable process and socket to make a 4M chip out of it. Steamroller on IBM 22nm SOI would have been pretty badass.

 

[ShintaiDK]

Samsung's current 14nm process is advertised as being low-power, not that any of us really know anything about it because they haven't sold any commercial products based on it yet (that I know of). Isn't Samsung in the habit of producing two processes based on the same basic node, usually one low-power and one high(er) performance?

Can you tell me any high performance logic parts Samsung have manufactored in recent times?

 

[mrmt]

Okay. So how big of a deal would it need to be to require a separate filing?

There's no formal threshold, executives must gauge what is a relevant fact is and properly address the market.

Not my point though . . . Mubadala kept them alive. What makes you think the Chicoms will do anything different?

I think it would be foolish to keep AMD alive if you don't have any stake in AMD-related ventures like Mubadala. The chinese companies might fund AMD R&D, but that's not the same as providing working capital or buying shares as Mubadala is doing.

Designs for what? Their GCN designs are actually quite good. It's not like Samsung has to license Steamroller or Excavator, it could be they want graphics IP for products. Besides, people are constantly bashing Steamroller for I-don't-know-what reason. It's pretty good, and would have been much better had AMD had a suitable process and socket to make a 4M chip out of it. Steamroller on IBM 22nm SOI would have been pretty badass.

Market disagrees with you. NVidia is mopping the floor with GCN and Steamroller was yet another failed AMD product.

But the point I want to make is that GLF has access to IBM 22nm SOI process that they could make available for AMD. They could even outsource production for IBM and still charge AMD if the commercial conditions were right, but instead both AMD and GLF decided to not pursue this path. Regardless of how badass Steamroller is now, it is not badass enough to make feasible the porting to 22nm SOI. AMD could have a suitable process, it was their choice to not go down that rabbit hole, not GLF limitations

I think this is related to the crazy densities of the GPU part, I don't think IBM process is suited for that, much less to HDLs they are employing with Carrizo.

 

[frozentundra123456]

Question to all: Samsung's 14nm + AMD expertise could offer a proper alternative to leading arch's no? After all, GF is to blame for many missteps lately.

Are scenarios such as Samsung opening up it's factories for someone like AMD who's clearly in great need, impossible?

If Samsungs 14nm is in fact as great as you say, which we don't really know yet, why would they open it up to AMD, who is trying to break into ARM, making them a competitor?