AMD Blasts Intel Atom Chips for Micro Servers

http://www.xbitlabs.com/news/other/display/20121211234004_AMD_Blasts_Intel_Atom_Chips_for_Micro_Servers.html

Truth is stranger than fiction. You just can't make up s*** this funny.

-VBG

red colored layer
Thick gate for high voltage I/O
http://www.eetimes.com/electronics-news/4403075/IBM--Intel-face-off-in-22-nm-process-at-IEDM?pageNumber=2

Nothing is up with the double fins. It's just representative.
To increase current on a planar transistor, you make it wider.
On a tri-gate, you add fins.

-VBG

Larger gate widths => Less subthreshold conductance

-VBG

Considering this is from two years ago, they may be sold out...

-VBG

We'll see.

Maybe Valleyview?

http://www.anandtech.com/show/6203/details-on-intels-valleyview-soc-emerge-22nm-atom-with-ivy-bridge-graphics

-VBG

http://blogs.amd.com/amd-blog-site-back-online/

Not sure if anyone posted this yet. From AMD's blog site:

AMD Blog Site Back Online, No External Users Exposed
As you probably noticed, our AMD blogs were down for approximately 24 hours. Our blog site was the target of an attack on August 19th. We believe that the attackers posted fewer than 200 AMD employee usernames and salted password hashes to a hacker website.

AMD only uses salted password hashes, which is an industry best practice for encryption and extremely difficult to crack. We immediately took the blog site offline and changed all passwords.

We do not store any customer or personal user information, or any AMD employment data, on the blog site. Again, the only data exposed were less than 200 AMD employee email addresses and their accompanying encrypted passwords. Please note that no AMD customer information was retrieved or affected – meaning no external users were exposed in any way. AMD remains committed to data security and user privacy and has launched an investigation into this matter.

We don’t anticipate any other issues with our blog site and expect to have new content up shortly. In the meantime, feel free to visit our Facebook page.

Elmer is a moderator and there are those who appreciate his efforts to keep the off-topic posting to a minimum. Were I a moderator, my first act would be to moderate your last post.

-VBG

Do you happen to have a link?

Thanks!

-VBG

The real question is why did MSFT chose IvyBridge over Cloverview for the high-end and Tegra over Cloverview for the low end.

I guess MSFT had to have a WinRT option, that was the point of Surface. Did Cloverview not provide enough of of an uplift over Tegra, that they had to use IvyBridge?

-VBG

http://en.wikipedia.org/wiki/Netronome

In November 2007, Netronome announced a technology licensing and sales and marketing agreement with Intel Corporation focused on the extension of the Intel IXP28XX product line of network processors. Under the terms of the agreement, Netronome is developing a next-generation line of IXP-compatible, high-end network processors that combine the Intel IXP28XX technology with Netronome’s architecture.

My take:
Intel has another small foundry customer who does not complete with them in any way directly. The are using technology they bought from Intel and are run by an ex-Intel exec.

In February 2011, Netronome announced Howard Bubb as Chief Executive Officer who brings more than 20 years of experience in leading growth for a number of semiconductor and communications companies including Dialogic and Intel’s Communications Infrastructure Group

BWDIK?

-VBG

Indeed. They can't even get their facts straight what process is used to make the processor for Romley (Jaketown 32nm). But they are going to directly contradict an Intel Fellow.

youtube video around 20:00.
Kelin Kuhn talking about how tri-gate improves RDF in the channel vs. planar and how multiple theshold voltages are possible.

I guess the real take-away here is that the so-called "industry experts" that Piper analysts are talking to have no clue how to make this stuff work well. Maybe Intel is further ahead than people think.

Edit: had wrong time (28 vs. 20)

-VBG

Dual core. Google intel atom centerton

ARM analyst day slides:

http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9MTQxMzA2fENoaWxkSUQ9LTF8VHlwZT0z&t=1

-VBG

http://blogs.barrons.com/techtraderdaily/2012/05/25/armh-ubs-ups-to-buy-notes-reasons-intel-wont-crush-them/?mod=yahoobarrons

ARM’s ecosystem support is substantial with 75% of the top smart phone applications being ARM specific (with the exception of simple compute apps like Sudoku, apps have to be re-written/re-compiled which is Intel currently in the process of doing to port these to an x86 environment).

ARM is aiming to move the goal posts with products like ARM Cortex- A7 and big.LITTLE

ARM maintains a process advantage today, with customers shipping out products based on 28nm versus Intel at 32nm currently.

Intel joins a crowded market with 15+ apps processor companies and the two biggest OEMs with in-house capabilities (Samsung/Apple).

Had to laugh at the process advantage one.
What analysis! 28<32, ergo ARM is ahead. Idiot.

-VBG

This Noonen guy is either confused or blowing smoke.

The 250K wafers was over 5Q. If yields were averaging 60% ramping linearly, and 32nm was ramping start linearly, then
250K wafers * 256 die/wafer * 60% yield is 38.4M die. about half that volume would be in the last Q, or 19M die. That number seems a bit absurd.

Most likely die yields increased from 20% to 40% and a factor of two got mixed in there somewhere. That would be more in line with historical AMD/GF yields.

-VBG

No one can
Someone can, but I don't think it is GF on 32nm in 2012.

-VBG

GLOBALFOUNDRIES Dresden Fab Ships 250,000th 32nm HKMG Wafer

http://www.marketwatch.com/story/globalfoundries-dresden-fab-ships-250000th-32nm-hkmg-wafer-2012-03-21

Wonder if 32nm yields can be inferred from this?

BD is 315 mm2
Llano is 228 mm2

AMD shipped how many BDs?
560k maybe? (9M TAM in units/yr * 5% share * 5Qs)
How many Llanos?
30M maybe?? (350M TAM/yr * blended 17% share desktop/mobile * fudge for mix Llano v.s. Thuban, Brazos, etc.)

Ignoring BD and dedicating all starts to Llano, assuming 256 die candidates/wafer would mean ~46% yield on average.

Prob too much uncertainty to get a good handle on it without better mix info. But yield probably aren't fantastic, or everything they shipped in last 5Q was 32nm, which is unlikely considering they blamed Q3 numbers on one-time bad 45nm yields.

-VBG

AMD to buy SeaMicro...

http://online.wsj.com/article/SB10001424052970203986604577253501384154044.html

This is a negative in my opinion.

-VBG

See Wouter's post.
I believe it has been stated publicly that Saltwell is a compaction of Bonnell and Silvermont is a new arch.

http://www.anandtech.com/show/4333/intels-silvermont-a-new-atom-architecture

-VBG

I doubt it.

-VBG

You guys are giving marketing waaaay too much credit.

-VBG

Meet the analyst...

Bobby Burleson
Research Analyst, Technology
bburleson@canaccordgenuity.com
1.415.229.7163

Bobby Burleson joined Canaccord’s San Francisco office as a Managing Director in the US Equity Research department. Mr. Burleson covers growing opportunities in small and mid-cap semiconductor devices and related technologies, an area which he has focused on for many years. Prior, he worked at ThinkEquity Partners. Earlier in his career, he was in the Semiconductor Research group at Thomas Weisel Partners and also worked as an investment banking analyst with Banc of America Securities. His experience enables him to tap into and leverage an extensive network of contacts in North America, Europe and Asia in carrying out the research in this diverse field. Mr. Burleson holds a BA from the University of California at Berkeley.

Covered Companies: 3D Systems; Advanced Analogic Technologies; Advanced Micro Devices; Altera Corporation; Analog Devices; Atmel Corporation; BE Semiconductor Industries N.V.; Cascade Microtech; Fairchild Semiconductor International; Integrated Device Technology; Intel Corporation; Intersil Corporation; Magma Design Automation; Mentor Graphics; MKS Instruments; NVIDIA Corporation; OmniVision Technologies; ON Semiconductor Corporation; SMSC; Texas Instruments; Xilinx

Larrabee uses P54C Pentium-based cores

Link?

I'm not saying you are wrong, but curious to see confirmation.

Found a link here: http://arstechnica.com/hardware/news/2008/07/intels-larrabee-gpu-based-on-secret-pentagon-tech-sorta.ars
but this sounds far-fetched to me.

Obviously if they use whatever Atom chip is current die size will balloon to something unmanageable

Not obvious at all.
Silverthorne is 26mm2 in 45nm.
32 cores is 832mm2 which is bigger than FOV on a scanner, but not by that much. Tukwila is over 700mm2. If you consider that much of the Silverthorne die is I/O



then I don't think it unreasonable at all.

Larabee is a pretty big die, and each core is vaguely similar to SVT.



Silverthorne was what, 40M transistors?
P54C was maybe 3M?
If KNF is P54C-based, why is it so big?

-VBG

The TCN article claims 125W TDP for the SCC.
It doesn't say anything about TDP for other products.
SCC is neither KNF nor KNC.
SCC is 48 cores, KNF is 32 cores, KNC is 50+ cores.

The SCC die shot in the picture is codenamed "Rock Creek".
The SC'11 presentation by Dr. Hazra has mulitple die shots of "Aubrey Isle", AKA Larrabee, AKA Knight's Ferry or KNF.
No die shot of KNC, of course, or that blow the mystery around 50+

BTW, I believe Rock Creek was Pentium based, but I don't think Aubrey Isle was. Look at the cores in Dr. Hazra's presentation. Looks a lot like 32 Bonnell type cores to me.

-VBG

I was thinking that about a dividend increase too, but there was just an increase in May. Is Nov too soon?

-VBG

This is just the first step in becoming the "Predator"....

-VBG

This Galen Gruman guy is smoking crack.
Intel has minimum requirements for Ultrabooks.
Too bad Infoworld doesn't have minimum requirements for their reporters. But then this is the entity that harbored Tom Yager for a while...


-VBG

(Is that a slip?)

No.

-VBG

So many errors in this but amusing I suppose.

I enjoyed this one:

When Turbo Core kicks in, the standard clock speed of the FX-8150, the highest-end version of the FX, can speed from 3.6GHz to 3.6GHz

-VBG

OT

Not to nitpick your grammatical reference, but I think you meant gerund, not infinitive.

-VBG

So basically PSO said $2.05 EPS this year and FY rev up 10% YOY and this guy is projecting 2011 EPS at $1.94 and $1.81 for 2012? I guess we will see who is right...

I think you're wrong.

I know you are wrong, because I asked someone last month who works in F12/F32 and is in a position to know. As a former employee, I would think you would accept the word of a current one. But believe what you will. If you really need a link try this:
azcentral.com
Check the sidebar on the right side.

In 2009, Fabs 22 and 32 were converted into one mega-factory called Fab 32.
Source: Intel

-VBG

No. Why?

It has been upgraded.

-VBG

Link to Barclays Capital Global Technology Conference

http://www.intc.com/eventdetail.cfm?EventID=88918

-VBG

Laptop nightmares: Scary systems with fatal flaws

http://news.cnet.com/2300-1041_3-10005374.html

3 of the 6 systems are AMD laptops with poor performance cited as one of the flaws. A Dell, a Compaq and an Acer.

No wonder OEMs are "excited" about Bobcat...

-VBG

Shhhh....

Don't tell Mr. Smith.
Intel is already late to the cloud party.

-VBG

Please just sell your Intel stock.
You put the N in SNR.