Breaking News
News & Analysis

Data Centers Tap ARM, 100GE

2/9/2016 10:00 AM EST
16 comments
NO RATINGS
1 saves
More Related Links
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 2   >   >>
Bruzzer
User Rank
Author
Re: Competitive Environment
Bruzzer   2/12/2016 1:27:39 AM
NO RATINGS
@nick_rb

Any specific thoughts or inquiries including on vernacular I'll answer as straight forward and simply as possible.

Mike Bruzzone, Camp Marketing

Bruzzer
User Rank
Author
Raspberry Pi, NUC Boards, 4xxx dual mobile SIP
Bruzzer   2/12/2016 1:13:16 AM
NO RATINGS
Raspberry Pi daisy chained into server, NUC Bay Trail and Cherry Trail single boards daisy chained into a server. Yes I've seen them all up to eight blades wide.  Ok at least operational as a network appliance . . .

The really interesting conundrum is what some plan for all that ECC less dual mobile system in package salvage . . . daisy chained into a server, who would have thought?

Yes, I said dual mobile SIP salvage from a notebook.  Hope those design build producers and their manufacturers screen out the melted ones.

Where the next platform opportunity might see are a row of PC sticks on USB plane interfacing with their dedicated 128 GB flash drives, off to a PCI NIC. and you've got yourself a  data center in a shoe box. 

Mike Bruzzone, Camp Marketing

perl_geek
User Rank
Author
Power consumption is key
perl_geek   2/11/2016 1:20:48 PM
NO RATINGS
I was at a meeting the other night where a service provider described replacing 2 X86 servers with Raspberry Pi units. The X86's 200-watt 24/7 power bills are roughly $200/year. The $50 Pi consumes about 1 watt. CAPEX paid back in six months. (Numbers not necessarily spot-on, but close enough for government work.)

It's true that there's a major difference in the computing power available, but CPU grunt isn't really key in most server applications.

(See the article about LinkkedIn, too.)

 

moronda
User Rank
Author
Re: More to come
moronda   2/11/2016 12:18:48 PM
NO RATINGS
Are you an AMCC employee or board member?


They are morons. They have a fiduciary responsibility to spend company money wisely.


All they have done is flush millions of dollars down the toilet.


This ridiculous initiative started years ago. He joined in 2009. That is 7 years ago.
 

It's been a massive failure.

nick_rb
User Rank
Author
Re: Competitive Environment
nick_rb   2/11/2016 4:50:02 AM
NO RATINGS
@ Bruzzer

As with all your participations on EETimes (and recently on Electronics Weekly), I recognise the words you use as English, but your use of syntax and argumentation make it difficult (at least for me!) to understand what you are attempting to express. My not being a native speaker might of course be the cause. Would you be able to reformulate your comment in a different way, so that I can benefit from your inputs? I would also welcome the help of any other kind souls on that forum, if one would like to participate...

Thanks in advance!

GSMD
User Rank
Author
ARM server CPUs
GSMD   2/11/2016 2:25:22 AM
NO RATINGS
The key problem is the lack of a credible ARM server processor. If a good 16 core 64 bit part with   single thread performance of about 1.5 IPC and sufficienttly high  IO/DDR bandwdth exists, you will see ARM get a 30 plus market share. X86 legacy compatibility or sw compatibility is a non issue in the Linux world.

Markets like India are slowly becoming Linux only, in fact in Govt. purchases Linux is mandatory. You have to prove that open source does not cut it to justify buying Windows or Oracle. Or you could potentially get hauled up on corruption charges ! These days OS means linux, java server means jboss or tomcat and rdbms means postgres. This with an esb is the core of most enterprise systems , maybe with apache spark/hadoop thrown in. Whether it runs on an x86 or a monkey with an abacus does not matter.  As an example, all large Indian core banking db servers are IBM power. The reason being oracle licensing costs are lower compared to x86. So the sole determinant is cost.

In fact if IBM power costs were on par with x86, you would have seen IBM get a 30 % share. RH now supports ARM pretty well and barring the odd pacakge, most standard enterprise SW has supported ARM builds. Bottom line, in the low and mid level dual and quad socket Linux market, the only thing that matters is cost, as long as it is a Tier1 supplying the box. In fact for security sensitive servers, ARM trustzone is way better than the new Intel extension. The market is for ARM to lose. x86 compatibility or legacy issues are purely an American or European phenomenon. Japan too I guess.

DMcCunney
User Rank
Author
Re: What a laugh
DMcCunney   2/10/2016 11:49:04 AM
NO RATINGS
@moronda: I don't think ARM servers are going anywhere. It's all hype and no reality.

Well, Intel certainly hopes so.  But ARM is getting serious investigation in the server space.  The biggest driver is probably power efficiency.  ARM CPUs use less power than Intel, and as data centers proliferate and continually add servers to meet demand, power costs are a major factor.  And the more power you consume, the more heat you generate as a by product, and the greater your cooling requirements become. The additional cooling equipment and the power needed to run it only drives costs up higher.

Intel's attempts in the low power space have not matched ARM.  And while X86 servers may be more powerful on an individual basis, ARM based machines may be powerful enough to meet most needs.  The main roadblock I saw to ARM adoption was lack of a clear 64bit upgrade path, but we are beginning to see that picture clarify.

If I'm someone at Google or Facebook looking and building Yet Another Datacenter, with thousands of servers in racks, I'm watching ARM with high interest.  I meet demand by throwing more servers at it, but the servers I throw don't have to be the fastest and most powerful to perform their function, and factors like power requirements and cooling become far more important in the purchase decision.

>Dennis

Wilco1
User Rank
Author
Re: More to come
Wilco1   2/10/2016 2:54:20 AM
NO RATINGS
Samsung achieved 14nm production at the same time as Intel and has higher volume. Intel delayed 10nm to 2nd half of 2017 while Samsung/GF and TSMC accelerate 10nm rollout.

AMCC has been shipping ARM servers for some time now. The experience from this will help next generation XGene-3 compete with designs from Broadcom, QC and others. Being early is risky but can also lead to great success. Time will tell.

>I'm not an Intel employee or fan. I am just a realist about trying to compete with chipzilla.

Calling AMCC CEO/directors morons is not being a realist.

moronda
User Rank
Author
Re: More to come
moronda   2/10/2016 1:05:41 AM
NO RATINGS
If anyone could gain some traction it would be Broadcom or Qualcomm. They have the resources to try and gain some traction. I still don't think there is much hope to take a large bite out of Intel any time soon. They are not standing still. They will have 10 nm processors coming soon. Intel has their own dedicated fabs and therefore is able to beat the fabless model.


Broadcom and Qualcomm will come out with theirs. Then the idea may gain more traction. They will have the software engineer resources to make it half way usable.


AMCC has been at this for years with no pay off.


I'm not an Intel employee or fan. I am just a realist about trying to compete with chipzilla.

 

 

Bruzzer
User Rank
Author
Competitive Environment
Bruzzer   2/9/2016 9:00:44 PM
NO RATINGS
Thunder places Cavium nicely within the Intel Xeon high margin E5 and E7 thin volume zones. And up's the data processing ante on network function off load, accelerated packet inspection, codec and encryption.

Applied Micro despite my early 2012 observations as loose lips, followed by 2013 hype of the moment playing the stock price pumper, into 2014 began showing maturity in the best practices of innovative whole platform development, into 2015 introduces statesmanship exemplifying coordination, decision and control.

Always playing the first mover advantage, on the science of industrial management best practice to commercialize a novel technology, AMCC is doing everything right. Even as ARM itself implemented on its unique forms of assistance that enable processor catch up and leap frog. Along with stalls in software development primarily organic, sometimes sabotaged, and never some how in parity with Intel speed through this nascent period of developments.

Applied is operating in the same 18 month window of opportunity as all player's, melding data, communication and understanding to implement a new type of phone system, which is a very old, tried and true method of industry, business and capitalist growth; SDN and NFV.

AMCC has demonstrations in network switch, HPC and data processing and appliances that will drive business intelligence on analytics. If there is a product platform opportunity at this inflection point of competitive platform development, at process saturation, Applied has dabbled in all of them.

AMCC on X-Gene generation 1 needed to sell 3 million $100 units to break even on assessment of research and development cost that is stayed, perhaps lessened on a higher per unit price game achieved exhibiting competitive performance in the Intel Xeon slim volume zones.

QCOM the following ARM server entrant appears to have two strategic thrusts; 1) ARM server SOC, 2) ARM server processor core others can wrap their unique system logic around.

This analyst originally conceived a disintegrated platform design regardless of bridge power and cost impact as an industrial enabler to rebuild compliment industry, set back technologically following financial devastation of that design compliment associated with the Intel blitz. I pressed for kit. QCOM likely sees a licensing model.

This analyst for some time has been an advocate of the second strategy on necessity to coordinate to build a compliment with sufficient mass and leverage, to tactically, reconfigure the Intel platform development playing field for competitive mass and leverage.

Where constant coordinated disruption of the Intel status quo becomes a necessity for gaining parity, and subsequently from a parity position, the ability to implement design enhancement's ahead, not necessarily to leap frog, but to stretch platform design possibilities and subsequently utility benefits that now makes Intel chase.

When that happens competitive design squad's are traditionally targeted by Intel associate trade guild for break up and industrial elimination. Take out and make a bad trade example of the leadership is one of the primary methods. When this happens Intel is losing on some front.

The analyst sees majority of ARM server entrants gaining traction through their network wedges aiming for Intel Xeon slim volume zones in excess of $900 IDM 1K volume price. The disadvantage competing in Xeon slim volumes zones, in excess of $900 IDM, is development visibility to Intel associate trade guild sharp shooters and saboteurs. Placement of confidence agent's for nefarious purpose is the greatest threat; stall, dismantle, poor counsel sent down wrong path and into the waiting grips of Goliath.

Below the $900 IDM price rung, now competing in the Xeon volume zones including setting in and among Xeon surplus volumes, development opportunities can be systematically picked off, primarily on guerrilla offensives associated with leap frog over Intel surplus capabilities. If indeed there is an ARM low power advantage, it can be relied here, because from an Intel mass volume and price perspective there are no competitive openings for platforms in parity. Under $300 IDM on the mass of surplus there are no opportunities what so ever lacking new use values.

Heterogeneous compute across a coherent bus is the new design hurdle. If requisite platform understanding exists within the cloud providers, I suggest it gets down to upstream development tools, libraries, and system modeling including managing instances of virtual compute into the memory tiers and flash arrays. Interconnect is ever more important. And of course the custom algorithms and refined applications support.

It would not hurt to have some enterprise regional channel footings as well.

This is the 29,504,144 million unit E5 2600 EP DP market;

26xx 4 core = 1,453,309 units or 5.07% of total

IDM average weighed price at 35% discount $470

26xx 6 core = 9,981,361 units or 33.87% of total

IDM average weighed price at 35% discount $284 which is an ARM block.

26xx 8 core = 6,205,161 units or 21.06% of total

IDM average weighed price at 35% discount $428

26xx 10 core = 5,379,118 units or 18.25% of total

IDM average weighed price at 35% discount $801

26xx 12 core = 2,881,729 units or 9.78% of total

IDM average weighed price at 35% discount $1188

26xx 14 core = 2,318,126 units or 7.87% of total

IDM average weighed price at 35% discount $1548

26xx 16 core = 574,051 units or 1.95% of total

IDM average weighed price at 35% discount $2097

26xx 18 core = 635,308 units of 2.16% of total

IDM average weighed price at 35% discount $2632

On DDR3 to DDR4 transition Haswell Xeon at 42.8% channel inventories is close to crossing over Ivy Xeon channel inventories at 57%.

On cluster appliance; load balancing and analytics IVY 4600 v2 MP has found new uses:

46xx 4 core = 2,264,545 units or 8.39% of total.

IDM average weighed price at 50% discount $275.50

46xx 6 core = 6,274,703 units or 23.25% of total.

IDM average weighed price at 50% discount $442.50

46xx 8 core = 10,234,843 units or 37.93% of total.

IDM average weighed price at 50% discount $748

46xx 10 core = 8,109,097 units or 30.05% of total.

IDM average weighed price at 50% discount $1491.33

46xx 12 core = 103,786 units or 0.38% of total.

IDM average weighed price at 50% discount $2197

On the volume of E5 46xx v2 produced there is very little E5 46xx v3

Average of all E7 v2 8 way grades IDM at 50% discount $2793.18

Average of all E7 v2 4 way grades IDM at 50% discount $1610.91

Average of all E7 v3  8 way grades IDM at 40% discount $3494.34

Average of all E7 v3 4 way grades IDM at 40% discount $1156.90 which is an ARM block that says wonders.

The opportunity currently in high volume mass surplus zones, is continuation of the Broadwell lag on Skylake Osborne. If so, Broadwell might be jinxed enabling this opening into the cover of clouds to continue. Otherwise is new use models in the Xeon slim volume zones.

On difficulties of Broadwell large dice to yield, Haswell will be refreshed. Competitors pursing the democratic capitalist interest of an industrial ideal might also gain time.

Mike Bruzzone, Camp Marketing

Page 1 / 2   >   >>

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed