The AMD Execution Thread [2007 - 2017]

[lanek]

Ugh. AMD's marketing people are unbearable.

Well you need to separate AMD marketing in India and rumors, hype spreaded by EU/US sites and forums.

It seems in India they offtly do separate launch ( maybe when the hardware will launch on shop later of the US release ). They like to do live presentation for press, reviewers etc. this was this time ( again ) in Goa. ( Same for China now that i think to it ).

On this sense, AMD India do a great job for create separated and specific evenement for India peoples.

The guy do a tweet for this show intended for Indi peoples. Then you have full bunch of *** who start spreads rumors, hype.. question in US and EU sites / forums. Just read that it was from AMD India PR should have close the discussion if they had a bit investigated about it. ( like maybe send a mail to an hardware site there lol or just look if the R9 285 have been allready released there.).

 

[eastmen]

http://www.extremetech.com/gaming/190026-amd-ceo-just-laid-out-companys-two-year-roadmap

There is nothing here that fills me with confidance .

Their flagship cpu is still on 32nm. Now they want to scrap it and replace it with something new that also isn't a prority for them.

It seems like Intel will be the only game in town for higher end chips at this point.

 

[3dilettante]

Read is no longer CEO, effective immediately.

http://www.cnbc.com/id/102071900

 

[Kaotik]

http://www.extremetech.com/gaming/190026-amd-ceo-just-laid-out-companys-two-year-roadmap

Make that ex-CEO
http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=1975570

Lisa Su is the new CEO

 

[3dilettante]

That release is more clear in that Read has ceased being CEO, president, and a member of the board of directors.
That's a lot of jobs to lose in one day.

edit:
Come to think of it, perhaps the order of precedence is his dropping from the board. If it stems from some kind of disagreement with them, he would not serve their purposes as an executive any longer.

It could be a new job, although I'm not clear on when that requires leaving the board.

 

[Raqia]

Read is still a young guy and although his tenure at AMD wasn't very long, it must have been extremely stressful. I'd like to know what he has to say about his departure since the company isn't exactly running on all cylinders yet.

Edit: Just listened to the CC, he said it was the right time. Lisa Su's technical expertise is deeper than his and perhaps he was just brought on for the initial turn-around and the connections he had. This is still awkward and sudden timing though and there must be more to the story.

 

[eastmen]

I hope she is more aggressive than he was.

 

[ninelven]

I really like this move from AMD. However, it seems to be a bit make or break point for the company. I feel like they now have the right people in the right places focusing on the right things, so if things don't go well for them now I'm not sure they ever will.

 

[AlexV]

I hope she is more aggressive than he was.

It is somewhat difficult to be aggressive in a pitched artillery battle when all you have is your trusty squirt gun, and your opponent sports Tsar Bombardas. Attempting to survive is a tall enough order to begin with.

 

[eastmen]

It is somewhat difficult to be aggressive in a pitched artillery battle when all you have is your trusty squirt gun, and your opponent sports Tsar Bombardas. Attempting to survive is a tall enough order to begin with.

I dunno , everyone knocks the bulldozer cpu line but it isn't bad its the fact that its still on 32nm that's the problem.

 

[trinibwoy]

I dunno , everyone knocks the bulldozer cpu line but it isn't bad its the fact that its still on 32nm that's the problem.

How would a different manufacturing node have helped? The architecture is just unsuitable for current workloads.

 

[3dilettante]

It is somewhat difficult to be aggressive in a pitched artillery battle when all you have is your trusty squirt gun, and your opponent sports Tsar Bombardas. Attempting to survive is a tall enough order to begin with.

Being outgunned doesn't make it harder to be aggressive, see the ending to Butch Cassidy and the Sundance Kid.
Frankly, some of the moves like semicustom are AMD's aggressivly making itself submissive to any client that can foot the cash up front.

The lead time of these architectures is such that we probably haven't seen the full outcome of Read's tenure, not that AMD is promising much in the next few years. The period of time before we see an opportunity for a change is at once somewhat understandable and in many ways not good enough.

I dunno , everyone knocks the bulldozer cpu line but it isn't bad its the fact that its still on 32nm that's the problem.

So was Sandy Bridge.
The transition to 28nm for Steamroller and the loss of the upper range of the Bulldozer lineage's clock envelope shows that it is out of step with the physical realities going forward, possibly more so than even Bulldozer was for the process it debuted on.
It's only going to get worse with the upcoming processes that are even more highly optimized for density and power at the expense of the design corner Bulldozer targeted--one that the now-important GPUs in the APUs seriously dislike.
Kaveri's 28nm is still partially optimized to cater to AMD, whose influence in this regard is vastly reduced, and it was customized to help the GPU, which hurt the Bulldozer-derived core further.

Intel's early forays with Broadwell and its foundry overtures show process tuning industry-wide that is avoiding Bulldozer's landing zone.

 

[silent_guy]

Yes, it's just too early to judge Read's tenure. How many new chips can you really think up, design, and bring to market in such a short time?

It's still a very open-ended question if, eg, the ARM based server chips will be a major source of actual revenue.

 

[pMax]

Kaveri's 28nm is still partially optimized to cater to AMD [...] and it was customized to help the GPU, which hurt the Bulldozer-derived core further.

Would you mind to extend this? Northbridge apart, I do not see them.

 

[digitalwanderer]

I really like this move from AMD. However, it seems to be a bit make or break point for the company. I feel like they now have the right people in the right places focusing on the right things, so if things don't go well for them now I'm not sure they ever will.

You summed up my feelings pretty well. They seem to have all the pieces, now what can they put together with 'em?

 

[3dilettante]

Would you mind to extend this? Northbridge apart, I do not see them.

The process was modified and the transistor mix was changed. Trinity to Kaveri brought a change from SOI to bulk, per AMD's description--a shift in the metal stack away from a V-type stack with finely pitched low metal layers and increasing size of the wires in upper metal layers to a more ASIC-like stack with more layers at a finer pitch and more gradual growth in size, and a change in transistor mix with a larger share of normal and higher-threshold transistors and a big drop in the leakier fast transistors.
(edit: Correction, my memory got it backwards:
http://pc.watch.impress.co.jp/img/pcw/docs/635/132/html/09.png.html

Per my thinking here in another thread, the shift hinted at a global shift in the characteristics of the process, such that the use of nominal and long-length transistors could acheive similar results at 28nm as the share of high-Vt at 32nm.

The following slide shows the shift in metal layer density and distribution, and the change in transistor leakage that penalizes the fastest ones:
http://pc.watch.impress.co.jp/img/pcw/docs/635/132/html/10.png.html

(end edit)

I'm trying to hunt down the slide that showed the shift in transistor mix, but here is the slide on Extremetech that include's AMD's description of the metal layer changes:

http://www.extremetech.com/wp-content/uploads/2014/01/28nmSHP.jpg

 

[eastmen]

Its still sad to me that they are talking about cuzorio or whatever that has the evacuator core and yet there is no real refresh to the FX line. It be nice if they threw a bone out to socket 939 users and put 4 evacuator modules / 8 cores onto a chip and released it for those users to buy.

 

[Alexko]

Its still sad to me that they are talking about cuzorio or whatever that has the evacuator core and yet there is no real refresh to the FX line. It be nice if they threw a bone out to socket 939 users and put 4 evacuator modules / 8 cores onto a chip and released it for those users to buy.

There likely won't be anything until 2016, when the new architecture shows up with FinFETs, and possibly HBM.

 

[eastmen]

they might as well not release anything at that point , everyone left will be on intel. I don't think they will be able to reclaim the high end market from intel as they should be on 9nm by 2016

 

[Alexko]

Why not, if they have a good product?