The AMD Execution Thread [2007 - 2017]

[Alexko]

[sorry, accidental double post]

 

[Alexko]

I'm not going to tell you to stop it with the noise again, but rather will stop it myself (again).

function: really likely is difficult to establish, as the design and context are not without risk. It's opaque to us how the 14nm process they intend to use is faring in practice, and how adequate it is for the sort of MPU they want to bring. Furthermore, this should be AMD's first SMT implementation, so validation there is an unknown. They also appear to be changing their cache hierarchy from the one that they have been using for ages (in no small part because they were confident in their ability to validate it), which adds further murkiness. Historically, AMD's woes tended to came as they were moving into validation, as opposed to when they were presenting slides, and the former might be in pretty early stages at best at this moment. We are likely to start hearing about how things are going later on during this year, one assumes.

If the validation process uncovers issues that can be circumvented by disabling certain features (at the expense of some performance/power-efficiency, of course) I would expect AMD to do just that, even with SMT. They're in such a dreadful competitive situation that they have no choice, and even a half-broken Zen CPU ought to be better than Vishera.

 

[Deleted member 13524]

Zen will not trickle down into the APU lineups, and that those will see some XV refresh being brought about, still on 28nm. So to some extent they should have a "tried & true" part that that ships without much pain ready, although it is difficult to work out much enthusiasm for what can only be regarded as rather long in the tooth.

If the Excavator refresh is practically Carrizo + DDR4 (rumors point to it being a dual-module CPU + 8 CU GCN1.2) and made using a rather old process, why isn't it coming much sooner?

 

[Razor1]

http://seekingalpha.com/article/3283275-amd-fails-again-keeping-it-in-perspective

This kinda sums everything up.

 

[UniversalTruth]

....

 

[UniversalTruth]

If the Excavator refresh is practically Carrizo + DDR4 (rumors point to it being a dual-module CPU + 8 CU GCN1.2) and made using a rather old process, why isn't it coming much sooner?

Remember that Carrizo is mainly optimised for certain power ranges, mainly around 15 W, where it shows best efficiency ?

I would think they need time to bring similar heavy optimisations in the desktop suited power envelopes.

 

[DavidGraham]

What is the impact of Fury's underwhelming launch on the company's earnings through 2015? Could they ever recover now?

 

[silent_guy]

I don't think it's very important: they will still sell out because, crappy marketing aside, it's a decent performers for its price.

I think that the 3xx rebranding at a higher price will result in increased sales at a higher ASP. That may have a bigger impact on earnings than FuryX.

 

[UniversalTruth]

Oh, no, quadruple facepalm

Microsoft might be interested in buying AMD, according to source

 

[Rikimaru]

Oh, no, quadruple facepalm

Microsoft might be interested in buying AMD, according to source

Analysts estimate that Microsoft pays around $100 for every Xbox One system-on-chip to AMD. Life-to-date sales of Xbox One are around are around 12.6 million units, which means that Microsoft has already paid AMD around $1.26 billion for Xbox One chips. The acquisition of AMD could save it around a billion per year on Xbox One chips alone. It the company develops appropriate chips for smartphones and tablets, Microsoft’s savings could be even higher.

Very good analytic article.

 

[Alexko]

Someone on the Tech Report suggested an AMD acquisition drinking game with the following rules:

1) If the merged companies can form a new acronym (DAMMIT, for example), drink.
2) If the news announcement for the merger mentions "synergies," drink.
3) If the company purchasing AMD is not based in the USA, drink.
4) If the purchasing company is based in the middle east, two drinks.
5) If the purchasing company is based in Taiwan, two drinks.
6) If the price paid for AMD is less than what AMD paid for ATI, drink - must be hard liquor.
7) If the buyout is blocked by parties that own AMD debt, two drinks.
8) If the company purchasing AMD was ever owned by AMD (Global Foundaries, Qualcomm, etc) finish your drink and spend your life savings investing in the purchasing companies stock.

That last part is probably not a good idea.

It looks life fun, but at this point I'd be really concerned about my liver.

 

[Grall]

5) If the purchasing company is based in Taiwan China, two three drinks.

Microsoft buying AMD wouldn't be so bad maybe. Dunno what would happen with discrete GPU development tho - MS has traditionally wanted to prop up Xbox on the expense of the PC due to one of these earning them royalties on software while the other doesn't - we would have to watch out about that, I think...

 

[AlexV]

Beyond demonic possession or losing a terrible bet while drunk, what could motivate MS to do this? I understand that after (apparently? that's what the press said) offering 55 bln. for Salesforce it is expected that no ridiculous purchase is a bridge too far for MS, but still...

Full redisclosure: I work for MS, however not in any area related to any potential acquisition, therefore what I write is merely my 2c.

 

[McHuj]

I guess savings on Xbox chips maybe enough. Right now AMD's market cap is 1.92B. Obviously the price would go up for a sale, but that's less than what MS paid for Minecraft. If they buy AMD for 3B, but save a 1B in the long term, it would probably be worth it, especially if they're planning another Xbox. Perhaps there's some savings on servers for Azure this way as well.

I would be surprised in this acquisition, but with all the cash MS has, they have to do something with it.

 

[silent_guy]

I find it highly unlikely that Microsoft pays $100 to AMD per Xbox.

 

[BRiT]

I find it highly unlikely that Microsoft pays $100 to AMD per Xbox.

Indeed.

 

[Psycho]

I find it more unlikely that AMD pays $0 to TSMC/GF per xbox SoC wafer

 

[Razor1]

well the cost of the Xbox chip is ~105 bucks to produce

 

[Rys]

Got a source for that information that's reliable?

 

[Gubbi]

Got a source for that information that's reliable?

I concur, >$100 seems excessive.

A 300mm 28nm HP wafer cost around $3000 in 2013 (from memory, can't find the source). At ~360mm²/die that would yield around 150 dies per wafer or around $20 per die, - not discounting yield.

Yield can't be worse than 50% (redundancy for regular structures as SRAM and CUs), with packaging I'd expect a per die cost below $50 a pop.

*Edit: I can see 28nm wafer prices were around $5800 in early 2014 (crunch?)

Cheers