AMD Migrates 20nm Chips In Development to FinFET – Pays $33 Million Charge

[Phynaz]

The article doesn't say it's a write off, nor specifies it implicitly, so we don't know that for sure. And even if it was, it would indicate that the 20 nm process tech related design costs were only $33M. Corresponding 14 nm may be more than that, but not that much, let's say $50M. So in the end, the whole idea that designs would be so expensive on 14 nm that AMD could not afford it is debunked.

PS. You're very keen on telling everyone else on this forum that they don't understand this and that. It's kind of funny. You should step down from your high towers and direct that towards yourself, where it's more correctly aimed at since you're usually the one on this forum that gets things wrong.

Holy crap dude. I don't care what the article doesn't say, that's what a one time charge is. I'm now convinced that you are just trolling me.

Where in the world did I say that it only cost $33M to design for 20nm? Nowhere does the article "indicate" such a thing. The $33M is a charge for what isn't usable anymore. For example they may have had 20nm masks made. Those masks are a company asset that are now worthless. Therefore there is a one time charge to reflect that.

Another place the charge could be coming from could be a GloFo agreement. If AMD had placed orders for 20nm test vehicles that they will no longer be using they may still owe GloFo for those.

How about you skip the articles and go strait to AMD :

Additionally the company anticipates GAAP gross margin to be further impacted by a one-time charge of approximately $33 million associated with a technology node transition from 20 nanometer (nm) to FinFET. The company started several product designs in 20nm that will instead transition to the leading-edge FinFET node.

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=2064993
The non-usable 20nm work is what the one time charge is.

Now you can go use your favorite search engine and look up "one-time charge". I'll even help you with that: http://www.investopedia.com/terms/o/one-time-charge.asp


Finance 101. Classes are available at your local community college.

 

[Fjodor2001]

The $33M was for 20 nm process tech specific costs related to their designs that are now not useable. Do you envision *any* of the 20 nm process tech specific work to be usable? If so what? Otherwise that sum covers all 20 nm specific costs.

PS. Ofcourse there are other costs associated with their designs. But if the result of that work can be reused on 14 nm it is not 20 nm specific.

 

[dark zero]

If Bristol Ridge were made on that nm size...

 

[mrmt]

The article doesn't say it's a write off, nor specifies it implicitly, so we don't know that for sure. And even if it was, it would indicate that the 20 nm process tech related design costs were only $33M. Corresponding 14 nm may be more than that, but not that much, let's say $50M. So in the end, the whole idea that designs would be so expensive on 14 nm that AMD could not afford it is debunked.

For this kind of expense to impact Gross Margin it *has* to be a write off, but it has to be a write off from very specific accounts like inventory or gross revenue they had already booked, or both. It is not an ordinary expense, as migrating to 14nm would be just plain vanilla OPEX, not subject of a direct charge impacting gross margin.

 

[pablo87]

I don't really care but I'm assuming they spent some NRE's and intended them to be amortized into the COGS of 20nm product, probably following previous practice. More Global Foundries' stuff that probably wouldn't happen with TSMC.

 

[Thala]

The $33M was for 20 nm process tech specific costs related to their designs that are now not useable.

You are describing a write down. A one time charge is not neccesarily a write down of an asset. We only know that it was one time charge, but not if it was a cache charge or non-cache charge. Do you have more information than anyone else?
The $33M one time charge could for instance be the additional investment for moving the design to 14nm if not already covered by OPEX.

 

[Dresdenboy]

Yes. They just missed yet another sales forecast this quarter.

Sorry, but you're asking for it.

I remember the EPS forecast contests (covering INTC and AMD) and I did even won sometimes a T-shirt. But if you learned about system theory and other stuff (closely related to stock trading) - esp. in the book "Black Swan", you'll know how difficult it is for anyone of us, the whole bunch of experts on this forum, Intel's market experts, or AMD's, to exactly predict markets. The bigger the unforeseen random events in the world, the greater the usual deviation from any prediction based on historical facts and assumptions about the future will be. That also fits well to the car OEM posting in another thread..

And lucky shots don't count as success, btw. But this is what most fund managers or EPS estimators do. Otherwise their ranking wouldn't fluctuate that much per iteration.

 

[mrmt]

Sorry, but you're asking for it.

I remember the EPS forecast contests (covering INTC and AMD) and I did even won sometimes a T-shirt. But if you learned about system theory and other stuff (closely related to stock trading) - esp. in the book "Black Swan", you'll know how difficult it is for anyone of us, the whole bunch of experts on this forum, Intel's market experts, or AMD's, to exactly predict markets. The bigger the unforeseen random events in the world, the greater the usual deviation from any prediction based on historical facts and assumptions about the future will be. That also fits well to the car OEM posting in another thread..

And lucky shots don't count as success, btw. But this is what most fund managers or EPS estimators do. Otherwise their ranking wouldn't fluctuate that much per iteration.

So your reply to "why AMD cannot hit a 90 days forecast" is... to show that others cannot hit multi-year forecasts. Very scientific.

 

[Dresdenboy]

So your reply to "why AMD cannot hit a 90 days forecast" is... to show that others cannot hit multi-year forecasts. Very scientific.

Ok, it's your right to demand what you want from any forecast out there. But what is observed reality telling you about your model about how forecasts are done? Especially in difficult to predict markets (e.g. single digit market shares), other than the forecast of "sell the mix to nearly anyone out there without much interferring of others".

Those Itanium forecasts are an interesting case study for difficult to predict markets.

 

[Phynaz]

So your reply to "why AMD cannot hit a 90 days forecast" is... to show that others cannot hit multi-year forecasts. Very scientific.

From nine years ago

 

[mrmt]

Ok, it's your right to demand what you want from any forecast out there. But what is observed reality telling you about your model about how forecasts are done? Especially in difficult to predict markets (e.g. single digit market shares), other than the forecast of "sell the mix to nearly anyone out there without much interferring of others".

I build and tune forecasting models, and not for environments where a correct prediction get me a t-shirt for not missing a forecast. But since you want to go down this rabbit hole, I do not have the right to demand a higher level of precision of a 90 days prediction against a 5 years business plan of a new, unproved product, that should be a given despite the reduced market share of AMD.

The degree of uncertainty of both forecasts are completely different, the amount of variables is also far smaller too, and some of the variables on the 90 days forecast are also under control of the said companies, but you can't have the nowhere the same level of control for a 3-10 years business plan. That's why you never saw Intel actually committing itself to these huge Itanium numbers on their quarterly or even annual outlooks, and nobody expected them to do it.

A 90-days sales forecast is a different beast. The companies usually commit to it, they are a signal to investors of how things are done in terms of execution of the aforementioned business plans, and the drivers are far easier to deal with, as there are multi-quarter supply contracts that must be honored, and consequently fulfill part of the sales forecast, there are the marketing guys and sales teams that can give a significant "feel" of the market direction. I'm sure you now can get a picture of the "small" difference between modeling the two scenarios.

AMD situation is a little harder, because they are in the bottom in terms of margins they can get in every contract, they can't go any lower otherwise it's better to not sell at all, so they do not have control of what contracts they get, they only get what Intel doesn't want on the PC market, and their capital structure is wrecked, so every quarter counts for survival and two or three missed forecasts means cuts deleterious for the business plan. They basically have less controls over the drivers' model than a healthy company of any size would.

What should AMD do then? Well, being a bit more honest and being a lot more conservative would help, but they can't do that otherwise Kumar and Lisa won't be able to tap dance around the numbers as they do every Q&A, they would have to explain investors that the numbers don't add up and that their current business plans rely on pie-in-the-sky premises that they don't see many chances of being sustained, hence my comment about AMD fantasy numbers, and hence Stacy Rasgon comments about AMD being unable to give a 60 days forecast.

 

[Dresdenboy]

From nine years ago

Do you think, the predictive capabilities of the human mind given incomplete information somehow changed during the last years?

Ok, I don't want to hurt someones feelings here, so I bring a more generic example of the capability of experts even with the usually better method of taking the average guesses than looking at individual divergences:

Source: http://greenbackd.com/2013/03/21/sp...e-too-optimistic-and-the-market-is-expensive/

 

[Dresdenboy]

I build and tune forecasting models, and not for environments where a correct prediction get me a t-shirt for not missing a forecast. But since you want to go down this rabbit hole, I do not have the right to demand a higher level of precision of a 90 days prediction against a 5 years business plan of a new, unproved product, that should be a given despite the reduced market share of AMD.

That's interesting to hear. In some abstract view, any situation prediction used in ADAS systems also contains models. And no matter how exact a model is (despite ressource constraints, etc.), the quality of its output heavily depends on that of the input. Otherwise it would be: garbage in, garbage out. There's some similarity: If a financial model or some other model predicts the outcome of a given timespan, it still works better if there is no significant event happening during that timespan, while it passes in reality.

Those contests were more than 10 years ago (2002 - 2003 or '04 I think), covering a much different AMD, and a less different Intel. Now I'm more to applying EA (esp. LGP, CGP) to trading models. A colleague is trying NNs to predict stock prices and forex rates only for the next day. Even that is no simple task due to the events.

The degree of uncertainty of both forecasts are completely different, the amount of variables is also far smaller too, and some of the variables on the 90 days forecast are also under control of the said companies, but you can't have the nowhere the same level of control for a 3-10 years business plan. That's why you never saw Intel actually committing itself to these huge Itanium numbers on their quarterly or even annual outlooks, and nobody expected them to do it.

A 90-days sales forecast is a different beast. The companies usually commit to it, they are a signal to investors of how things are done in terms of execution of the aforementioned business plans, and the drivers are far easier to deal with, as there are multi-quarter supply contracts that must be honored, and consequently fulfill part of the sales forecast, there are the marketing guys and sales teams that can give a significant "feel" of the market direction. I'm sure you now can get a picture of the "small" difference between modeling the two scenarios.

AMD situation is a little harder, because they are in the bottom in terms of margins they can get in every contract, they can't go any lower otherwise it's better to not sell at all, so they do not have control of what contracts they get, they only get what Intel doesn't want on the PC market, and their capital structure is wrecked, so every quarter counts for survival and two or three missed forecasts means cuts deleterious for the business plan. They basically have less controls over the drivers' model than a healthy company of any size would.

What should AMD do then? Well, being a bit more honest and being a lot more conservative would help, but they can't do that otherwise Kumar and Lisa won't be able to tap dance around the numbers as they do every Q&A, they would have to explain investors that the numbers don't add up and that their current business plans rely on pie-in-the-sky premises that they don't see many chances of being sustained, hence my comment about AMD fantasy numbers, and hence Stacy Rasgon comments about AMD being unable to give a 60 days forecast.

I think, this is a good summary of the situation, way better than the often seen ranting in this forum.

AMD's situation is not a comfortable one and market fluctuations usually hit them harder due to some market dynamics inherently helping the bigger participants. And with their small product portfolio they are much more prone to negative events happening during their forecasted quarter. This could be any missed design win, or sales contract.

BTW all this is why I always traded this stuff short term and never thought about doing some investing on my own.

 

[mrmt]

That's interesting to hear. In some abstract view, any situation prediction used in ADAS systems also contains models. And no matter how exact a model is (despite ressource constraints, etc.), the quality of its output heavily depends on that of the input. Otherwise it would be: garbage in, garbage out. There's some similarity: If a financial model or some other model predicts the outcome of a given timespan, it still works better if there is no significant event happening during that timespan, while it passes in reality.

If you notice the questions on the Q&A in the last few quarters and in the FAD some analysts have been questioning AMD business premises with variable degree of aggressiveness, and it seems that those questions aren't a matter of lucky shots only as their questioning have been spot on in more than a handful of cases. How can AMD not to get these drivers, and how can external analysts get them more than once?

I don't think it's a matter of coincidence, the only reason I can think of is a hidden agenda by AMD management, not necessarily in bad faith, that makes them not show the bad, but accurate forecasts that they should if they could show them. But in any case I don't think AMD management is being remotely forthcoming in their forecasts.