[AT] AMD Reorganizes Business Units - no more high performance x86 cores?

[NostaSeronx]

There is no such thing on any roadmap, sorry :/.

Well that explains the 40h-4Fh models.

https://www.google.com/patents/US20120258611

It's been on the roadmap since 2012.

Don't act like you know if you don't have proof that you know.

 

[Homeles]

It's been on the roadmap since 2012.

It's been omitted from recent roadmaps, no?

I really wish AMD was more forthcoming with their high end plans.

 

[jpiniero]

IIRC, the tweaked Piledriver Opertons that were released earlier in the year is it for the non-APU Constructor Equipment line.

 

[NostaSeronx]

It's been omitted from recent roadmaps, no?

The roadmap doesn't go into 2015 nor is it time accurate. Berlin and Seattle are both Q4 2014 products while Warsaw is a Q1 2014 product.

For 2015;
15h 40h-4Fh = 28nm FDSOI with Medium Performance(High Perf+High Density) Libraries
15h 60h-6Fh = 20nm LPM with High Performance Libraries
16h 30h-3Fh = 20nm LPM with High Density Libaries

I really wish AMD was more forthcoming with their high end plans.

If you look at linkedin profiles all day they are pretty forthcoming.

AMD's Kaveri started development in 2012 while 40h-4Fh didn't start till late 2013. Both designs were on paper way before that. While the silicon designing phase didn't start till way after.

--
Off-topic but relevant;
https://i.imgur.com/vzjNZpm.jpg
^-- Cheat sheet for GlobalFoundries metal stack.

 

[mrmt]

I really wish AMD was more forthcoming with their high end plans.

The management doesn't even bother to talk about the high end in every Q&A they have the chance to speak. How can they be more forthcoming?

 

[mrmt]

IIRC, the tweaked Piledriver Opertons that were released earlier in the year is it for the non-APU Constructor Equipment line.

It's not even tweaked, it's probably errata.

 

[pablo87]

P4 was developed and marketed under Barret. Otelini took the ceo post in 2005.

Otellini was COO. His team pushed P4 hard well past expiry.

 

[mrmt]

Otellini was COO. His team pushed P4 hard well past expiry.

Otelini started his COO term in 2002, when P4 was a well established reality. P4 development started circa 1996. In fact, if you look at the dates, Conroe started to be designed in... 2002, right after Otelini assumed the COO position.

 

[NTMBK]

The roadmap doesn't go into 2015 nor is it time accurate. Berlin and Seattle are both Q4 2014 products while Warsaw is a Q1 2014 product.

Honestly, I hope that AMD would be focusing their resources on getting the K12 and next generation x86 processor right, instead of putting out another Bulldozer based server SKU which will fail to compete against Broadwell-EP.

 

[NostaSeronx]

Honestly, I hope that AMD would be focusing their resources on getting the K12 and next generation x86 processor right, instead of putting out another Bulldozer based server SKU which will fail to compete against Broadwell-EP.

AMD roadmaps don't show everything unless you go back in time.

There is another reason to launch a 28nm CPU with integrated Freedom Fabric in 2015; http://lmgtfy.com/?q=site%3Alinkedin.com+AMD+Project+%22Exascale%22

That and a semi-steady track record of yearly releases;
Magny-Cours - March 29, 2010
Interlagos - November 14, 2011
Abu Dhabi - November 5, 2012
Warsaw - January 20, 2014

Warsaw is a stop-gap but for what reason is it a stop-gap. We are getting the 28nm Server APUs via the Opteron X1100-2200 series. Why hide that next generation server CPU that replaces G34 and C32.

http://lmgtfy.com/?q=site:linkedin.com+"AMD"+Next+Generation+Server

Is the delay for 20nm? That would explain a shift from Q3-Q4 2014 to Q3-Q4 2015.
http://in.linkedin.com/pub/amit-sharma/5/a4b/641
http://www.linkedin.com/pub/huong-an/35/509/851
http://www.linkedin.com/in/jasontylerscott
http://www.linkedin.com/pub/russell-mcmullan/4/a4a/847
http://www.linkedin.com/pub/russell-schreiber/20/668/2b7
http://uk.linkedin.com/pub/seng-oon-toh/36/20/583
http://www.linkedin.com/pub/sivanandh-ramadass/25/364/5b0
http://ca.linkedin.com/pub/manpriya-singh/43/178/b03
http://www.linkedin.com/in/naveennandishkubsad
http://in.linkedin.com/in/viveksabnis

It doesn't show a delay for 20nm though as there are some profiles very late into the game. That state 28nm high performance CPUs way beyond the Kaveri threshold. The 28nm mention could be a typo since 8 and 0 is separated by only 9.

---
K12 and its sister x86-64 core are using Cluster Multithreading just in a 4-way pattern. The big change to the ISA is the addition to true 3-operand and 4-operand to general purpose instructions. Zen/Mountain Peaks are using the building blocks of core cluster multithreading from Dozer and processor cluster multithreading from Cat. Each core in the Mountain Peak module has full local coherency in the shared L1D cache. While full global coherency with each unshared L2 partition between all cores, homogeneous and heterogeneous.

 

[mrmt]

Honestly, I hope that AMD would be focusing their resources on getting the K12 and next generation x86 processor right, instead of putting out another Bulldozer based server SKU which will fail to compete against Broadwell-EP.

Before bringing Broadwell-EP to the table, what scenarios Kaveri and Carrizo would make sense for a server box? You know, except if AMD decided to give it away for free and I had the need for a very simple web-hosting box, I can't think of any other. The reality is that Xeon-EP is far beyond APU capabilities, Competition for them comes from the Xeon "12" series based, and for servers they mop the floor with the APU and eat them for breakfast afterwards.

I think Berlin and Toronto were PR stunts and won't sell in any meaningful quantities. They can't even match raw performance of the 2 years old 4300 series and they aren't such a big perf/watt improvement. Berlin and Toronto are just PR stunts for AMD to say they launched something. It's AMD saying "hey, we're still here, don't forget us".

Wanna see a funny thing?

http://seamicro.com/SM15000

It seems that someone shot the wrong point in the performance curve, and that the sweet spot for server performance isn't really around Atom/Jaguar.

 

[chrisjames61]

I disagree. Otellini had StrongArm and basically killed it - lacked vision. Pushed for P4 too IIRC (it's only Israel mobile team saved the day) still hung on to P4 way too long.

IIRC Apple gave them first option to come up with the processor for the iPhone. I guess unless you had a crystal ball to know they would sell 600 million to date. But these guys are paid mega millions to figure these things out.

 

[jpiniero]

Hey don't forget that AMD had Snapdragon.

IIRC Apple gave them first option to come up with the processor for the iPhone. I guess unless you had a crystal ball to know they would sell 600 million to date. But these guys are paid mega millions to figure these things out.

Apple sure lucked out on that. Buying PA Semi turned out to be a pretty good decision.

 

[frozentundra123456]

Yea, Intel didnt really have a suitable chip at the time anyway did they, (for a phone I mean)? Even now they dont have an integrated modem in a shipping chip.

I agree hindsight is always 20/20, but I also agree that the executives are getting paid huge salaries to make forward reaching, intutitive decisions, which intel didnt do.

 

[Idontcare]

Otellini was COO. His team pushed P4 hard well past expiry.

Otelini started his COO term in 2002, when P4 was a well established reality. P4 development started circa 1996. In fact, if you look at the dates, Conroe started to be designed in... 2002, right after Otelini assumed the COO position.

The Chief Operations Officer should not be making any decisions that impact the specific product offerings made by their employer in the present or the future.

The COO's job is to see to it that the products (which someone else decided the company would make) are being made as efficiently as possible. That's all.

As to who chose to pursue P4 and Core - that responsibility and decision ought to have been left to the CEO, the CTO and the CMO.

 

[witeken]

I agree hindsight is always 20/20, but I also agree that the executives are getting paid huge salaries to make forward reaching, intutitive decisions, which intel didnt do.

Putting a cheap Intel chip inside Apple's iPhone was absolutely not intuitive. But they had to know phones were the next big thing and don't follow the clear money but the innovation. Otellini knew this decently but didn't do it.

 

[TreVader]

No, Intel didn't need ARM to figure this out.

Yes they did.

 

[TreVader]

AMD needs to focus on power targets and I think they are smart to get into SoC design considering they have far better GPU design capabilities than intel and nvidia obv isn't going to be succeeding at CPU design anytime soon.

 

[Arachnotronic]

Hey don't forget that AMD had Snapdragon.

No, they had low power GPU technology that Qualcomm bought, nurtured, and built up with hundreds of millions in R&D and smart business decisions.

Qualcomm didn't buy Snapdragon ready-made, despite what people want to say.

 

[TreVader]

No, they had low power GPU technology that Qualcomm bought, nurtured, and built up with hundreds of millions in R&D and smart business decisions.



Qualcomm didn't buy Snapdragon ready-made, despite what people want to say.

Where is the evidence for this? Do you think Qualcomm buys tech so they can just go and spend "hundreds of millions" aka a large percentage of their R&D budget on reinventing the wheel?


Your understanding of how tech acquisition works is... Interesting.

 

[ShintaiDK]

Where is the evidence for this? Do you think Qualcomm buys tech so they can just go and spend "hundreds of millions" aka a large percentage of their R&D budget on reinventing the wheel?


Your understanding of how tech acquisition works is... Interesting.

I didnt know you just bought finished products you could make billions on it right away. I always thought you bought IP and and knowhow instead.

All these tech companies seems to have gotten it wrong all the time then.

 

[Arachnotronic]

Where is the evidence for this? Do you think Qualcomm buys tech so they can just go and spend "hundreds of millions" aka a large percentage of their R&D budget on reinventing the wheel?


Your understanding of how tech acquisition works is... Interesting.

Do you think Qualcomm doesn't spend billions each year on building new GPU technology, new CPUs, DSPs, SoCs?

You think that the AMD assets gave them ready-made the SoCs they're shipping today?

 

[TreVader]

Do you think Qualcomm doesn't spend billions each year on building new GPU technology, new CPUs, DSPs, SoCs?



You think that the AMD assets gave them ready-made the SoCs they're shipping today?



LOL!

Usually when you buy technology you don't then go invent the same technology yourself. That's the whole point.

 

[Arachnotronic]

Usually when you buy technology you don't then go invent the same technology yourself. That's the whole point.

So, because Apple bought PA Semi, it doesn't need to pay engineers to develop future products?

 

[sirroman]

But I think the shift is much deeper than what Anand's article is showing. The division that Lisa Su will be leading is composed by CPU, GPU and professional GPUs.

AMD’s client, consumer graphics, and professional graphics groups will now be combined under a single group, the Computing and Graphics Business Group.

Meanwhile AMD’s second group will be the Enterprise, Embedded and Semi-Custom Business Group. This group consolidates the server, embedded, and semi-custom groups under one roof. This structure does mean CPUs are essentially split – an Opteron sale is now an Enterprise sale rather than being kept with the Computing group CPU sales – but otherwise this marks the combining of AMD’s “fringe” groups such as SeaMicro and the semi-custom groups, which in contrast to the core technology focused Computing group are focused on building designs and applications around AMD’s core technologies.

Heading up these groups both directly and indirectly will be Dr. Lisa Su, who is getting a promotion from Senior VP and GM of Global Business Units to the C-level position of Chief Operating Officer (COO). AMD has not had a COO for a few years now, so this marks the return of that position to AMD’s executive organization and arguably makes Lisa AMD's second-in-command. Meanwhile AMD’s Chief Sales Officer, John Byrne, will also be getting a promotion of his own, which will see him move up to SVP and GM of the Computing group.

In regards to AMD’s new structure, Lisa will be taking direct control of the Enterprise group on an interim basis. Meanwhile Lisa will have indirect oversight of the Computing group, with John serving as GM of that group and reporting to Lisa. Lisa in turn will now report directly to CEO Rory Read.

Lisa will be COO and interim over the Enterprise, Embedded and Semi-Custom Business Group. It's Mr. Byrne that will oversee the Computing and Graphics Business Group that you mentioned.