AMD presents future plans at investor meetings

[Idontcare]

Well I like the "modern UI," so that might be why I like it.

Me too! That's why I noticed it, it looks sophisticated to me which is not what we'd ordinarily expect from AMD marketing slides.

Why are you discussing a number taken from a retail store when talking about embedded CPU? When you talk about embedded, and talk about oem worth something, you are talking about thousands, or even millions of units spread in a spam of years. You don't buy in 1000 batches, but enter in agreements of years of duration with the cpu manufacturer himself.

This is why embedded is low margin. You don't need bleeding edge, you don't need huge quantities upfront, it is always like "slow but reliable". And this always commanded low margins.

What Atenra is arguing is akin to expect that margins for a given component in a contract to supply a car manufacturer are huge because he only finds exorbitant prices in his local store.

When I worked at TI we had a huge embedded product portfolio spanning both digital as well as analog parts. It was huge, like 20,000 products huge.

In the lobby of the TI headquarters (the forest lane bldg) there was a Ford Explorer in the middle of the foyer area that was enclosed in plexiglass because it had orange (or was it pink?) stickers all over the vehicle pointing out where there was a TI embedded product located within the vehicle.

The tally was something absurdly ridiculous, like >300 or some such. It really made for great marketing-by-eye because of course they held as many contract negotiations as possible in the forest lane building.

The contracts were long (10-15yrs), used the same node and were high margin because of it. Once the node depreciated in 4yrs your margins went sky-high because the fab and process tech tools were already written off. You basically pay for labor and consumables.

And because you are built into a product that was qualified by a safety body (anything that involves risk of life must be, medical/auto/air) the customer has no leverage to replace you or negotiate down prices in the coming decade.

Where this model of high margins falls apart is when you go fabless. You don't own the asset that is depreciating, so you don't scoop up high margins once the first four years have passed. The foundry does.

I don't know what kind of margins AMD makes but if they are buying the chips from GloFo in order to resell them to embedded markets then they are already giving up a hefty percentage of their embedded margins to GloFo, that is the nature of the business since GloFo is not a charity.

This was the one reason TI did not go fabless. They kept their existing fabs and just decided to stop developing new digital CMOS nodes (they still develop new analog nodes). All the new digital nodes are at the foundries, but they kept their existing depreciated fabs with all those 10-15 yr contracts because of the margins.

It will be interesting to see what kind of revenue model AMD ends up with in say 3-5 yrs. If they were "just another fabless business" then it would be rather straightforward to see the lay of their land. But being a fabless company with the GloFo contracts in place really makes for an uninspiring prognosis

 

[MeldarthX]

IDC - that was issue with dell; I've been building system since forever - worked at Falcon NW - *yes that was a fun job and also so damn thankless; the hours sucked*

I've seen the tactics Intel pulled to get us not to use AMD - *as their chips were faster at the time; and when X2 came out there was no comparison*

I've had to pull apart dells - hell I had several dells shipped to me at my old job a few years ago *mix of AMD and Intel* that didn't work out of the box; *reason they didn't bother to plug in the power into the motherboard; on ALL OF THEM*

Most prebuilts have their own issues....and some issues windows had with some drivers - tended to be issues with Via; and Nvidia; before AMD got into the chipset business - *which I believe was the main reason they did go in was to control the quality of the chipsets*

 

[Idontcare]

IDC - that was issue with dell; I've been building system since forever - worked at Falcon NW - *yes that was a fun job and also so damn thankless; the hours sucked*

I've seen the tactics Intel pulled to get us not to use AMD - *as their chips were faster at the time; and when X2 came out there was no comparison*

I've had to pull apart dells - hell I had several dells shipped to me at my old job a few years ago *mix of AMD and Intel* that didn't work out of the box; *reason they didn't bother to plug in the power into the motherboard; on ALL OF THEM*

Most prebuilts have their own issues....and some issues windows had with some drivers - tended to be issues with Via; and Nvidia; before AMD got into the chipset business - *which I believe was the main reason they did go in was to control the quality of the chipsets*

Yeah, very true, I think it can't be stressed that there is a distinction between "the problem was because of my AMD CPU" versus "the problem was because of my AMD-powered computer system".

I never had issues with AMD's cpus, but the issues were there with the maturity of the rest of the components.

As an enthusiast the distinction had merits, but for the vast majority of consumers the distinction was irrelevant. In the end they are left to conclude had they bought an Intel rig then they could have avoided all the trouble.

We see it now with Netflix and Microsoft not supporting HD on AMD bobcat boxes. It doesn't matter who is at fault, but if you want to avoid being a victim of the problem then you best not buy the AMD kit if that is your intended use.

 

[Phynaz]

Me too! That's why I noticed it, it looks sophisticated to me which is not what we'd ordinarily expect from AMD marketing slides.



When I worked at TI we had a huge embedded product portfolio spanning both digital as well as analog parts. It was huge, like 20,000 products huge.

In the lobby of the TI headquarters (the forest lane bldg) there was a Ford Explorer in the middle of the foyer area that was enclosed in plexiglass because it had orange (or was it pink?) stickers all over the vehicle pointing out where there was a TI embedded product located within the vehicle.

The tally was something absurdly ridiculous, like >300 or some such. It really made for great marketing-by-eye because of course they held as many contract negotiations as possible in the forest lane building.

The contracts were long (10-15yrs), used the same node and were high margin because of it. Once the node depreciated in 4yrs your margins went sky-high because the fab and process tech tools were already written off. You basically pay for labor and consumables.

And because you are built into a product that was qualified by a safety body (anything that involves risk of life must be, medical/auto/air) the customer has no leverage to replace you or negotiate down prices in the coming decade.

Those margins weren't as high as you are thinking.

First, my qualifications to speak about it. I spent my second career in auto parts manufacturing. (my first was in sales and now I'm in healthcare as you know).

The autos negotiate very hard. Here's some of the provisions they put in their contracts:

They own the design of everything they are buying.

If any kind of tooling is needed they own that too. I know that in semiconductors this may be different. But go on a tour of a parts manufacturer and you will the tooling stamped "Property of Ford Motor Company".

Contracts are have very stringent non-performance clauses - see the first points above. They own the designs and the tooling and if you aren't performing they just take the designs and tooling and give it to somebody else to manufacture.

Annual price decreases are part of the contract. The supplier is expected to gain economies as they get better at production. Trust me, Ford knew the litho equipment was off depreciation and that was written into the contract.

Suppliers hope that their production efficiencies out pace the price reductions in the contracts, so as a rule bids are below cost for the first couple of years of a contract.

Those long term contracts can come to bite you. One company I worked for had made a great contract for the coat hanger hooks in automobiles. 11 cents each was a great deal in 1975. By 2000 they cost 25 cents each to make, and the customer wasn't budging on the ~15M a year we were obligated to supply.

The thing is, automobiles are one of most complex things to manufacture from a logistics standpoint. When you are assembling 70K parts from 30K suppliers you get very, very good at negotiating contracts.

 

[Dribble]

You did a nice job ignoring the 'dense server' and 'ultraportable/low-power' dot points within that 50% figure.

Here's the full PDF of the presentation if anyone wants to take a look:
http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9MTc1OTU5fENoaWxkSUQ9LTF8VHlwZT0z&t=1

That's my point - they don't really get anything from the other dots right now. Basically they have consoles and that will magically turn into 50% revenues mostly from 'dense servers' and 'ultraportable/low-power'. So what the slides should say is:
-consoles 5%
-consoles 20%
-consoles 20%

-the rest 0%
-the rest 1%
-the rest 30%

or something...those other markets aren't going to magically grow like that anywhere but an AMD marketing slide.

 

[SocketF]

^ No room for Excavator unless AMD wants to stay on the left-side. Doing "all of the above" is not financially feasible.

IF there is a new server platform with DDR4, then they have to launch Excavator, because the platform wont pay of otherwise. Who will invest in a platform that is outdated after 2 years? Server platforms at AMD are always planned for 2 generations (sometimes even three if OEMs demand it, e.g. Istanbul). I don't expect too much of Excavator, probably just a 20nm shrink of Steamroller. That shouldn't cost too much to design, but will provide a sufficient performance/watt benefit.

 

[Homeles]

IF there is a new server platform with DDR4, then they have to launch Excavator, because the platform wont pay of otherwise. Who will invest in a platform that is outdated after 2 years? Server platforms at AMD are always planned for 2 generations (sometimes even three if OEMs demand it, e.g. Istanbul). I don't expect too much of Excavator, probably just a 20nm shrink of Steamroller. That shouldn't cost too much to design, but will provide a sufficient performance/watt benefit.

Excavator will be bringing high density cell libraries, and using an automated design favoring density over the hand tuned high speed designs that we're used to. It will also fix the terrible L3 latency.

 

[Idontcare]

IF there is a new server platform with DDR4, then they have to launch Excavator, because the platform wont pay of otherwise. Who will invest in a platform that is outdated after 2 years? Server platforms at AMD are always planned for 2 generations (sometimes even three if OEMs demand it, e.g. Istanbul). I don't expect too much of Excavator, probably just a 20nm shrink of Steamroller. That shouldn't cost too much to design, but will provide a sufficient performance/watt benefit.

Excavator will be bringing high density cell libraries, and using an automated design favoring density over the hand tuned high speed designs that we're used to. It will also fix the terrible L3 latency.

Realistically speaking, looking at how long it has taken for AMD and GloFo to get 28nm products to the market, 20nm won't be a reality for what? Another 3 or 4 yrs?

Unless AMD does it at TSMC, which I don't think they can afford.

 

[Ajay]

Realistically speaking, looking at how long it has taken for AMD and GloFo to get 28nm products to the market, 20nm won't be a reality for what? Another 3 or 4 yrs?

Unless AMD does it at TSMC, which I don't think they can afford.

Since 28nm was a couple of years late; applying the same metric would indicate a 2015/2016 timeframe. Supposedly, GF has been increasing investment in their 20/14nm nodes - so maybe they'll remain behind TSMC at the same lag time as they are now. And they can't afford to build CPUs @ TSMC with the WSA in effect. AMD has to build it's CPUs, according to the terms of the WSA, or they are just throwing money out the window and won't hit their cost reduction targets.

Sucks being AMD.

 

[ams23]

But being a fabless company with the GloFo contracts in place really makes for an uninspiring prognosis

This gets to the root of the problem. Cash balance already went down by about 20% from Q3 to Q4 2012 in part due to an $80 million penalty payment to GF. And between Q2 2013 and Q1 2014, AMD owes another $240 million to GF (with $40 million due in Q2 2013 and $200 million due in Q1 2014). So of course AMD expects to be cash flow positive in 2H 2013 when no penalty payments are due to GF, but then they will be cash flow negative again in Q1 2014 due to the heavy rear-loading of the GF penalty payment.

 

[SocketF]

Realistically speaking, looking at how long it has taken for AMD and GloFo to get 28nm products to the market, 20nm won't be a reality for what? Another 3 or 4 yrs?

That's the big question. Depends what went wrong. If gate-first was the big problem, then it could go smoother. But I have no clue how complicated the double-pattering technique is. Intel is doing it already quite long, but GF?

Anyhow, if I remember correctly, then the NY fab will be first to go to 20nm. IBM is a partner there too, i.e. I assume more experts are available.

But I would not be surprised, if the Excavator Opterons would be announced for ~2016 and then postponed to 2017. Then we would have the 3 years you have mentioned :biggrin:

Probably a very realistic scenario.

The fun fact is: Officially, GF has 20nm on the roadmap already for 2013 (!).

So we could use ARM chips @20nm as indicators. If there will be some launches in 2014 then everything is fine. If not, then it's like 28nm all over again.

 

[VirtualLarry]

We see it now with Netflix and Microsoft not supporting HD on AMD bobcat boxes. It doesn't matter who is at fault, but if you want to avoid being a victim of the problem then you best not buy the AMD kit if that is your intended use.

According to wirednuts, that problem has been fixed. I'm not a Netflix subscriber, so I cannot test the veracity of his claims.

 

[VirtualLarry]

You can also look at what Intel is doing with the NUC, $299.00 for a dual core+HT i3 1.8GHz and a motherboard, huge margins.

Is that a sign of things to come, if Intel switches to BGA-only CPUs? Will mobo makers attempt to use the market change to "reset" margins in the industry?

 

[jpiniero]

This gets to the root of the problem. Cash balance already went down by about 20% from Q3 to Q4 2012 in part due to an $80 million penalty payment to GF. And between Q2 2013 and Q1 2014, AMD owes another $240 million to GF (with $40 million due in Q2 2013 and $200 million due in Q1 2014). So of course AMD expects to be cash flow positive in 2H 2013 when no penalty payments are due to GF, but then they will be cash flow negative again in Q1 2014 due to the heavy rear-loading of the GF penalty payment.

Is that part of the $2B debt that AMD has or is that technically separate?

 

[Ajay]

Is that a sign of things to come, if Intel switches to BGA-only CPUs? Will mobo makers attempt to use the market change to "reset" margins in the industry?

IIRC, it's LGA through Skylake (for desktops).

 

[Idontcare]

Is that part of the $2B debt that AMD has or is that technically separate?

Seperate. Technically it should be booked as a short-term loan but AMD has been doing some really dodgy stuff with its accounting in how it deals with GloFo billings.

It serves the shareholders right though that they let the BoD get away with what they do, everyone there is getting what they deserve IMO.

 

[Homeles]

Realistically speaking, looking at how long it has taken for AMD and GloFo to get 28nm products to the market, 20nm won't be a reality for what? Another 3 or 4 yrs?

Unless AMD does it at TSMC, which I don't think they can afford.

Well supposedly GloFo has gotten their act together, though we certainly aren't seeing it. When you have two incompetent companies having issues, it's hard to tell which is the bottleneck at the current point in time.

 

[Arachnotronic]

Excavator will be bringing high density cell libraries, and using an automated design favoring density over the hand tuned high speed designs that we're used to. It will also fix the terrible L3 latency.

There goes competitiveness against Intel...oh, wait.

It's just going to get worse for AMD at the high end.

 

[Homeles]

There goes competitiveness against Intel...oh, wait.

It's just going to get worse for AMD at the high end.

Yeah, going from high speed designs to low cost, power efficient designs does not sound like enthusiasts will be having a good time. At least then the AMD people will see the light and realize that AMD has stalled progression as much as Intel has. Ah, who am I kidding...

 

[cytg111]

What could be working in amd's favor here is the diminishing returns on process nodes ahead, being a node behind is not as devastating now as it was a few back. But it looks like they really gotta get out of glofo's grip or for glofo to step it up. It's a pipe dream, but hey, it is a possibility.
Even if AMD were to yield a 20% IPC + clock increase with steamroller and put out a 6 module/12 core part, how long would it take Intel to respond to that? (guess they allready have with 2011). It is kinda sad, wish they'd just semi catch up though.

 

[mrmt]

Is that part of the $2B debt that AMD has or is that technically separate?

Separate. long term liabilities have more than 12 months before maturity, and debt implies purely financial obligations, not manufacturing related liabilities.

 

[AtenRa]

Where this model of high margins falls apart is when you go fabless. You don't own the asset that is depreciating, so you don't scoop up high margins once the first four years have passed. The foundry does.

I don't know what kind of margins AMD makes but if they are buying the chips from GloFo in order to resell them to embedded markets then they are already giving up a hefty percentage of their embedded margins to GloFo, that is the nature of the business since GloFo is not a charity.

Over time, wafer prices starting to fall. For example, that is one reason why current and past AMD/NVIDIA GPU prices fall after all. GloFo/TSMC will eventually sell wafers at a lower price and that’s translates in to lower retail price for the products. If TSMC/GloFo or any other Fab would not lower its wafer price then retail product prices would stay flat.


There is a reason why AMD targets the x86 embedded market and that is high margins, much higher that what AMD has on desktop products.

 

[jpiniero]

IIRC, it's LGA through Skylake (for desktops).

Well, the rumor is that there will be only BGA for Broadwell but future versions it is up in the air. Keep in mind Skylake will probably end up coming out in early 2016 so there's time to see what the market looks like before making a decision.

Even if AMD were to yield a 20% IPC + clock increase with steamroller and put out a 6 module/12 core part, how long would it take Intel to respond to that?

I don't think they would honestly. The power draw would be like 180W and OEMs would laugh at that.

As for the PS4, if they really are doing per-chip pricing, it's likely that GloFo will in the end get most of the money; and as the wafer costs decrease, so will Sony's cost.

 

[Idontcare]

Over time, wafer prices starting to fall. For example, that is one reason why current and past AMD/NVIDIA GPU prices fall after all. GloFo/TSMC will eventually sell wafers at a lower price and that’s translates in to lower retail price for the products. If TSMC/GloFo or any other Fab would not lower its wafer price then retail product prices would stay flat.


There is a reason why AMD targets the x86 embedded market and that is high margins, much higher that what AMD has on desktop products.

I hope for AMD's sake you are right.

Their financials paint a picture of them not having been right about much of anything in the past 8 yrs.

Looking at their contractual obligations with GloFo it is rather sad to think that knowledgeable educated individuals at AMD thought that such contractual obligations were in AMD's best interests and in their shareholders best interests (even though they kept it from their shareholders)...if they were silly enough to tie their hands with such obligations it leaves the question just how competent they are in their plans for just about everything else.

But even a broken clock is right twice a day, so hopefully AMD gets this one right and it doesn't become another ATI adventure.

 

[ShintaiDK]

Well, the rumor is that there will be only BGA for Broadwell but future versions it is up in the air. Keep in mind Skylake will probably end up coming out in early 2016 so there's time to see what the market looks like before making a decision.

That part is wrong. Broadwell is LGA too.