AMD 2017 Financial Analyst Day Discussion

[Daniel Nenni]

The AMD Analyst Day materials are up and there are some interesting nuggets especially on the process side. I can say with all certainty that AMD is in full-on 7nm node mode but the question is which foundry will they use for 7nm and why? Also, will AMD products make a dent in Intel and NVIDIA revenue? AMD stock has dropped 9%+ thus far today, any thoughts as to why? Feel free to email me privately if you prefer.

[table] style="width: 100%"
|-
| style="width: 19%" | Presenters
| style="width: 41%" | Titles
| style="width: 40%" | Presentations
|- style="border: 0px; font-style: inherit"
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Lisa Su
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | President and Chief Executive Officer
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Corporate Strategy
|- style="border: 0px; font-style: inherit"
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Mark Papermaster
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Senior Vice President and Chief Technology Officer
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Back in High Performance...To Stay
|- style="border: 0px; font-style: inherit"
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Jim Anderson
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Senior Vice President and General Manager, Computing and Graphics Business Group
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | High-Performance PCs
|- style="border: 0px; font-style: inherit"
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Raja Koduri
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Senior Vice President and Chief Architect, Radeon Technologies Group
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Radeon Rising
|- style="border: 0px; font-style: inherit"
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Forrest Norrod
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Senior Vice President and General Manager Enterprise, Embedded, and Semi-Custom Business Group
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | The Dawning of a New Era in the Datacenter
|- style="border: 0px; font-style: inherit"
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Devinder Kumar
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Senior Vice President, Chief Financial Officer and Treasurer
| style="padding: 10px; border-width: 0px 0px 1px; border-top-style: initial; border-right-style: initial; border-bottom-style: solid; border-left-style: initial; border-top-color: initial; border-right-color: initial; border-bottom-color: rgb(210, 217, 227); border-left-color: initial; font-style: inherit; vertical-align: middle" | Financial Roadmap
|-
[/table]

 

[Jozo035]

Their 7nm plan is pretty agresive. AMD should finaly get clear process advantage over Intel (Intel is planning at least 2 future generations on 14nm - Coffelake and Cascadelake - since their 10nm and 10nm+ will be failure in terms of performance).

Threadripper is interesting too. 16 cores (32 threads), more than 3GHz base + turbo... If they will price it at $1000-$1300 then it will be hard hit for Intels's HEDT (and small Xeons).

On the other hand, their MCM solution is not best... ...without interposer there will be big cache latencies and they are limiting it to just 2S servers... (which might not be major problem, since 2x32 cores in single system is a lot).

 

[lefty]

All tech stocks dropped today, so AMD's drop was just part of that and maybe also, the fact that RX Vega has been delayed didn't do down to well with the PC enthusiasts.
AMD will go with Globalfoundries all the way, because of the wafer agreement.
AMD say Zen2 is on 7nm in 2018 (but knowing Globalfoundries execution that could really be 2019).
Before that there will be a "Zen refresh" on 14nm+.
I predict that AMD will make a huge dent in Intel's revenue (mainly in the server market) and very little in NVidia's.

 

[Daniel Nenni]

The AMD GPUs were manufactured by TSMC up to 28nm and at 14nm they were moved to GF. CPUs were always manufactured by AMD then GF at 28nm and now at GF 14nm. The AMD CPU/GPU chips for the consoles are manufactured at TSMC on 16nm. Here is what Lisa Hsu said on analyst day, your guess is as good as mine:

Lisa T. Su, Advanced Micro Devices, Inc. - CEO
So look, on the technology question, we are a multi-foundry -- we have a multi-foundry strategy. This generation, we ramped a number of products in 16-nanometer and we ramped a number of products in 14-nanometer. And for all of the products that we had going, it was very, very important, actually, for us to have both of our foundry partners very actively engaged. In 7-nanometer, Mark showed some very, very aggressive road maps. We'll be one of the first to adopt a 7-nanometer for high performance, and we will similarly use 2 foundries. So we will use TSMC as well as GLOBALFOUNDRIES for 7-nanometer. And the key is -- our goal is to use the best that process technology has to offer so that we can innovate, design architecture, all of those other things. And so with our modified Wafer Supply Agreement that we did last year, it does give us the flexibility to use 7-nanometer at both TSMC and GLOBALFOUNDRIES.

Advanced Micro Devices' (AMD) CEO Lisa Su hosts 2017 Financial Analyst Day (Transcript) | Seeking Alpha

 

[Daniel Nenni]

And this just in, gonna be another rough day for AMD:

Intel Refutes Rumor of Licensing AMD Graphics Technology
Intel refuted a rumor spread by a technology Web site saying it is licensing graphics chip technology from AMD. The rumor had pushed up AMD shares by 12% on Tuesday, only to see the stock come crashing down by that amount on Wednesday after AMD management did not offer any mention of such a deal.

Intel Refutes Rumor of Licensing AMD Graphics Technology - Barron's

 

[lefty]

The AMD GPUs were manufactured by TSMC up to 28nm and at 14nm they were moved to GF. CPUs were always manufactured by AMD then GF at 28nm and now at GF 14nm. The AMD CPU/GPU chips for the consoles are manufactured at TSMC on 16nm. Here is what Lisa Hsu said on analyst day, your guess is as good as mine:

Lisa T. Su, Advanced Micro Devices, Inc. - CEO
So look, on the technology question, we are a multi-foundry -- we have a multi-foundry strategy. This generation, we ramped a number of products in 16-nanometer and we ramped a number of products in 14-nanometer. And for all of the products that we had going, it was very, very important, actually, for us to have both of our foundry partners very actively engaged. In 7-nanometer, Mark showed some very, very aggressive road maps. We'll be one of the first to adopt a 7-nanometer for high performance, and we will similarly use 2 foundries. So we will use TSMC as well as GLOBALFOUNDRIES for 7-nanometer. And the key is -- our goal is to use the best that process technology has to offer so that we can innovate, design architecture, all of those other things. And so with our modified Wafer Supply Agreement that we did last year, it does give us the flexibility to use 7-nanometer at both TSMC and GLOBALFOUNDRIES.

Advanced Micro Devices' (AMD) CEO Lisa Su hosts 2017 Financial Analyst Day (Transcript) | Seeking Alpha

Yeah, that's interesting. It sounds like AMD don't expect Globalfoundries to have 7nm ready on time. AMD had a quota to meet with the wafer agreement and before this their Carrizo CPUs sold too little to make the quota. So they had to switch manufacture of GPUs and consoles to Globalfoundries to make up for the shortfall. Since, Ryzen seems to be selling much better, they have more flexibility to use other foundries.

 

[Daniel Nenni]

Here are some random notes from the investor call:

Gross Margins:
No one talks about when (if?) AMD will be sustainably profitable any
longer. There was sporadic profitability when they first sold the
game console chips to Microsoft XboxOne and Sony PS/4, but have not
been in the black since. As a result, the analysts are looking at
revenue growth and gross margins as an indicator of the return to
profits.

With the introduction of a refresh of the GPU line (RX500 series) and
the new Ryzen CPU (and soon APU and server-centric) designs, the
analysts were expecting gross margins to move from the low 30% to
"38-40%".

When AMD said, "Well, in the 2nd quarter, GM will be 32%.", the
analysts were definitely perplexed (that's even lower than the 33% in
the first quarter).

There is a definite tug-of-war between the low margin game console
chips (Microsoft and Sony are both talking about a product refresh
later this year), and the newer GPU/CPU designs. So, for the next
quarter, at least, the lower margin designs will curtail the overall
financials. You could draw the conclusion that the forecast for
Ryzen CPU's will be muted, and not bring the GM up.

Inventory is up significantly:
I'm not sure the analysts bought into the argument that the
significant rise in inventory quarter-over-quarter was due to AMD's
readiness for "additional OEM's building desktops for back-to-school"
(see below), or whether they over-anticipated sales of Ryzen and GPU's.


Ryzen:
So, when will Ryzen contribute significantly to profitability?
Apparently, not in Q2. In Q1, AMD admitted that the goal was to
announce the product refresh, and let the DIY market buy parts from
Amazon and Newegg. "Traditional desktop manufacturers will announce
in Q2.", according to Lisa Su on the conf call.

Well, we're halfway through Q2, and the only desktops with Ryzen are
available from three off-brands... "CyberPowerPC", "CybertronPC", and
"iBUYPOWER". (Ironically, almost all of these three manufacturers
desktops with Ryzen CPU's use Nvidia GPU cards.

Supposedly, Asus and Lenovo have upcoming products to be available
with Ryzen, but I couldn't find any indication that Dell and HP have
desktops coming.

Q2 does not appear to be any more promising, in terms of Ryzen
contributing significantly to GM and profitability.


APU's and Naples server parts:
There were positive comments about 2H2017 releases of Zen core-based
APU designs for mobile laptops and Naples server CPU's.

Usually, AMD is aggressive releasing benchmark data vis-a-vis Intel.
I haven't seen anything about Naples yet, but will keep an eye out for
an announcement. In the desktop market, the strategy is "comparable
to mid-range i7, with an aggressive ASP". In the server space, the
mid-range performance with lower ASP approach doesn't really apply.


Vega GPU's:
The AMD high-end Vega GPU announcement later this year will be
interesting, although Nvidia appears to be on a roll.

The minor blip in Nvidia momentum when the AMD RX500 series was first
announced appears to have returned to normalcy. I haven't seen GPU
market share data lately, but I'm guessing that AMD has not
significantly moved the needle over the past year... I could be
wrong. Even so, GPU sales over the past year have not moved AMD to
higher margins and profitability, so it's really up to Ryzen and Naples.


Financial monkey business:
And, finally, although the $$$ amounts aren't major, AMD did pull a
quick bit of financial trickery... and, their explanation was kinda
confusing...


(1) "We have shifted our accounting of production mask sets from
OpEx to capital."

The analysts were all over this one... "Why?" "Shouldn't we see
better OpEx improvements compared to previous quarters, then?"
"Should that have accelerated reaching profitability?"

It wasn't clear to me whether all mask sets are now capital or just
"production" (last revision) mask costs. Capitalizing masks is OK,
but their explanation of how it will be recorded wasn't clear.


(2) lots of outstanding share warrants

One analyst smartly noted that AMD has issued a bunch of share
warrants to Abu Dhabi (which are able to be cashed in at any time, if
I recall). He asked whether AMD should be using a different "total
number of shares" in EPS calculations in the future.

With the total number of shares to increase significantly in the
future, that path to profitability will continue to be a difficult one.


(3) GAAP vs non-GAAP

And, lastly, I'm still amazed that companies (not just AMD) are able
to quote non-GAAP results, removing stock option expense from the GAAP
financials. If it's an expense that is incurred each quarter, it's
not really a non-recurring item, and shouldn't be part of the non-GAAP
data.

Hopefully some of this will be cleared up at the analyst day....

 

[lefty]

There's a good chance AMD will be profitable in the second quarter. Margins go down probably because there is more low margin console business with the launch of Scorpio, but also profits go up. Revenue will be up 17%.
IMHO the killer product is Naples. They showed how weak Intel's products are (too few cores and not enough PCIe lanes). A single socket Naples machine out performs a mid range dual socket Xeon. Dropbox seems to already have switched to AMD, with more cloud providers to follow. There are 30 server products to be released this quarter and AMD have seeded 5000 CPUs to OEMs and customers. Customers hate Intel's monopoly in the server market and they hate Intel's arbitrary pricing. This is a quote from a semiaccurate article:

Customers were not happy about this progressive squeeze and the ones SemiAccurate talked to all claimed to have vocally complained over the years.

Intel's new Scalable Xeon branding is just a price increase - SemiAccurate

 

[smeyer0028]

There is an interesting argument on Seeking Alpha I do not understand. The argument
is that AMD can't ramp up volume very fast because they do not have enough cash.
I don't understand how the foundary model accounting works, i.e. when are the
various payments and revenue recognitions in say a million IC order. Who pays for
chip failures or defects and when. Can anyone explain how this works.

 

[Daniel Nenni]

There is an interesting argument on Seeking Alpha I do not understand. The argument
is that AMD can't ramp up volume very fast because they do not have enough cash.
I don't understand how the foundry model accounting works, i.e. when are the
various payments and revenue recognitions in say a million IC order. Who pays for
chip failures or defects and when. Can anyone explain how this works.

I have done wafer agreements in the past but I do not have direct knowledge of this one. It is complicated by the fact that AMD's major investor (Mubadala Development Company) also owns GlobalFoundries so it is probably not a normal wafer agreement. Given that, I highly doubt GF would withhold wafers from AMD for cash flow reasons.

In regards to yield, if it is a process yield issue the foundry is responsible, if it is a design yield issue then it is on the fabless company. In regards to Samsung 10nm for example, Samsung had horrible process yield so instead of selling wafers they sold known good die to QCOM.

I do know that GF 14nm, which is what AMD currently uses, is a copy exact of Samsung 14nm and is yielding quite well for other companies so I highly doubt it is a process yield issue.

 

[smeyer0028]

Thanks. What happens with CPUs where I assume a Ryzen 5 is just a die that
will not pass the highest clock frequency tests? Also what happens with die's
that are not totally functional so they need to be sorted for different products.
I am just asking how things work in general. In the old days I wrote system go/nogo
tests. Method was to put the IC in a system and try it. That has changed now right?

 

[astilo]

Thanks. What happens with CPUs where I assume a Ryzen 5 is just a die that
will not pass the highest clock frequency tests? Also what happens with die's
that are not totally functional so they need to be sorted for different products.
I am just asking how things work in general. In the old days I wrote system go/nogo
tests. Method was to put the IC in a system and try it. That has changed now right?

Ryzen 5 and 7 are indeed the same chip. Every die has 8 physical cores (separated in two blocks, so a 4+4) and 16MB L3 Cache, also configured in a 2x 8MB. A prime die makes a Ryzen 7 CPU (1800X, 1700X and 1700), while a defective die, where 1 to 4 cores are not fully functional, makes a Ryzen 5 CPU (either a 6 or a 4 cores). In case of exceptional high prime yield, AMD could also decide to deactivate some good cores on purpose, in order to feed the Ryzen 5 CPU high demand. Voltage and leakage current, should instead be the way to separate the CPU by frequency range offerings.