http://fudzilla.com/news/43223-optane-could-be-hd-s-swan-song

Intel’s Optane Memory is headed to be one of the most disappointing bits of tech in the world of storage.

NOT WHAT WE WERE PROMISED    ..... Optane is just a cache memory,  one that has OS loading issues too.

The non-volatile memory promised “1,000X” the performance of today’s NAND-based SSDs with far higher density and lower cost than DRAM.    

THIS ISN'T THE CASE. REALLY.

Some even thought that if the tech worked like that you would not need RAM and a hard drive.  Not so, both RAM and a hard drive are still required on any older system.

While the tech is not bad and can find a niche it is not the blistering game-changer we had been promised. Instead we are getting a better version of the Smart Response Technology, an older tech which is easier to use and not so prone to pairing problems.

Intel Optane Memory keeps parts of the OS on the Optane drive to speed up performance and the majority of the OS remains on the platter drive. This means if you want to pull the hard drive from the system, you’ll have to unpair it first.

The current Intel Optane Memory implementation is also limited to a single hard drive. If you run two hard drives, the second one will see no caching improvement. Very much like SRT, Intel Optane Memory increases responsiveness overall. Of course, anything is an improvement over a hard drive.

The other issue is the price difference is not that hot. You can get a 128GB SATA SSD for about 60 euro so why bother?   Your entire OS and all  your data can go on a cheaper 128GB SATA SSD with room to spare.


Point to all of this ya ya blah blah is that Intel is touting the heck out of a high cost proprietary memory technology implying it can do a lot of wonderful things that IT CURRENTLY CANNOT DO.  

Next, Intel saying that Intel can stick at 14nm lithography first part of this year because "all improvements this year will come from Optane" is just slightly wrong headed.  "All performance improvements this year will come Optane" is just some more brown vapor speak from Intel trying to keep their stock price from going even lower.

All REAL 2017 performance improvements will actually come from AMD Ryzen chipsets at 10nm and from as yet unshipped new 10nm 3-D memory products coming from other memory vendors using a STANDARD 3-D NON-VOLATILE MEMORY INTERFACE that has yet to be declared by MS and the other OS vendors.   (4/2/17 the new generation persistent module NVDIMM-P standards (for the new Optane types) have now been declared by the JEDEC industry standards group)

Optane is just another over priced, overly touted, slower than expected,  overly expensive way to implement Intel "proprietary groundbreaking innovation" yet again.

Point being that you have to be able to put your whole OS into this new style memory system for it to be useful and right now you cannot do this because Wintel's MS part simply hasn't caught up with the tech wave at all yet.

Visual reality check on Intel claims to being smaller and denser than the "others" ......


   this is now for ARM products

Qualcomm's current 10nm chipset, the Snapdragon 835, a full function SOC with memory and all functions present inside the penny sized SOC.






Samsung's current 10nm chipset, the Exynos 8895, a full function SOC with memory and all functions inside the smaller than a penny sized SOC.


vs


Intel's "small and light M class" CPU  (still missing lots of various functions and it HAS NO BULK MEMORY as shown)    Yes, Intel's two larger chipsets sitting on a daughter board still don't make up a full function SOC like the Qualcomm 835 and the Samsung Exynos 8895 shown above

   this is now for Intel x86 products





Intel, your misleading Power Points and Press releases cannot change what is real.   Your "technical leadership" is all past tense and getting further and further behind as the months roll on past into whole full years of "behind".


=======================================


News Flash    Intel says they plan to go to 10nm, 7nm and 5nm using their existing FinFET processes.   Time line stated by Intel during their lithography tech presentations this past week is 2018 for 10nm, 2020 for 7nm and 2022 for 5nm.   These brown vapor announcements are running two full years behind the others at this current time.

TSMC and Samsung have announced (and have built Apple sample runs on) a refined multi-mask EUV FinFET for 7nm for this upcoming year --- with Gate All Around (electron laser) being their production pick for 5nm and 3nm in 2019-2020.  

Global Foundries announces a complete skipping of later generation 7nm FinFET processes and an early adaptation of a 5nm electron laser based Gate All Around process at 5nm and 3nm (skipping out on the 7nm generation completely as wasted capital and wasted time).    This is a wise plan for Global as Apple has already let the main contracts for 7nm to TSMC with Samsung as a production backup "safety".

ASML (the company that makes all the lithography equipment) was up on the stage with the NON-Intel groups, silently supporting the reality of what was being said as two (2) 5nm electron laser production machines are being built, with one of each slated to be installed at both TSMC and at Samsung sometimes later this year ---  with Global Foundries getting their first one early of next year.

Yep, Apple invests in their product lines 3 years out with basic production process research and Apple funds one future current ASML production line at each pet vendor for their dedicated production refinement use for the next 2-3 years as they build the "at risk" early lots of iPhone chipsets.

"Read between the lines" time --- Intel is using their brown vapor cannon very freely yet again to cover the fact they HAVE NO REAL PRODUCTION LITHOGRAPHY "INNOVATION" PLANS PAST 10nm and that their current 10nm plan is shot full of nagging low yield production issues at the moment, issues that will likely take a projected 12 months to work through Intel's crop of 10nm issues.  

Intel is also noted to be at the foot of the line to get the new ASML 5nm electron laser lithography equipment -- a fact that explains why Intel is sticking to finFET at 7nm, 5nm and 3nm supposedly.    That puts them yet another year behind the pack when they finally get real about having to use electron laser technology.

Intel now states publicly their tick-tock cycle now takes 3-4 years to complete on a single lithography level .....      





......  and yes, Intel, that is your brand new non-competitive 14nm Optane stuff shown over on the left in the pic below.

(example comes from 14nm Intel Optane vs the Besang 10nm Gate All Around system)

That Gate All Around is some powerful stuff, really ..... this comparison holds true at any future lithography level, btw ..... this comparison is made at 14nm which is the current Intel reality.    10nm, 5nm and 3nm will be even worse for Intel when that point in time rolls around.  

Please remember, Intel is two full years behind in lithography as of right now, so Intel's 14nm goes vs everybody else's 10nm, 10 Intel goes vs everybody else's 7 ..... by then Intel should be pretty much defunct or close to it and simply be unable to buy the new ASML technology and do the needed development to even try to catch up.




========================================


Voltage downsize plays out against aging Intel PC CMOS technology

Intel announces they are forming a "task force" to redo all PC CMOS requirements to allow all devices to run on 2 volts or less.

Intel can't buy readily buy really cheap pieces for their PC stuff any more since there is now a ever larger voltage gap between the cheaper, smaller, more efficient stuff the phone industry uses (in very large volumes) vs the old stuff what Intel x86 PC products still have to use (legacy production in much smaller volumes).

This need has arisen very recently since oncoming even more modern lithography levels run the entire shebang on 1.5 volts or less.   The voltage gap has become a Grand Canyon of late .... with Intel sitting on the wrong side of the gap.

Intel still produces legacy products that run at 5 volts or more --- and Intel HAS to get with the sub 2 volt program ASAP or lose the market for those items completely to the voltage decrease which has already arrived.

http://www.wired.co.uk/article/samsung-s8-dex

Samsung's DeX dock makes most old home PCs look anachronistic     the only cords I see in this picture are the monitor power cord and the DeX dock power cord.   Bluetooth 5.0 is quite capable for running the mouse, monitor, exterior speaker and the keyboard, as was promised to us months ago.



"WIRED was perhaps most impressed with a new dock accessory, almost casually introduced after the reveal of the handset itself.

This dock effectively turns your mobile into a home PC that can run Windows Word, PowerPoint and Excel. After you lower the S8 into the DeX, it couples the smartphone to an HDMI compatible monitor and connects to any Bluetooth-enabled, USB or RF-type keyboard and mouse.

The screen wakes up on docking and quick as a flash you are looking at not a mirror of the handset’s display but what appears to be a regular run-of-the-mill PC desktop.

The reason for this is Samsung has completely redesigned the Android UI so it is optimised for use with keyboard and mouse. This pimped UI offers up those familiar multiple resizable windows, contextual menus and a desktop web browser. In short, you immediately feel you are working on a normal home computer.

The DeX Station itself not only houses and fast charges your S8, it has some ports you’d expect from a PC: two USB 2.0, ethernet and USB-C. For the security conscious, while connected to the DeX, the smartphone is protected by the Samsung Knox security platform and no mobile data is transferred from the device to the “desktop”.

Collaborations with Microsoft and Adobe mean Samsung DeX is compatible with Microsoft Office and Adobe mobile apps, including Adobe Acrobat Reader mobile and Lightroom Mobile. The apps themselves have been altered, too, as part of that pimping of the UI. Microsoft Word, for example, looks more like the full-fat home PC version and not the mobile one. However, DeX also allows users to remotely access virtual desktops through offerings including Citrix, VMware and Amazon Web Services should you need access to Windows OS-based apps."

Google and Sammy and Mickysoft did it right, boys and girls, and it smells like a hit "standard feature" now doesn't it ????

Ubuntu seems to think so, as they just tossed in the towel on their Unity convergence system as their efforts to date are not nearly as nice or as good looking.    Collaborations with Microsoft and Adobe mean Samsung can apparently now do good enough mods to Mickysoft Word and Adobe products to make it "much prettier looking" than Micky could do alone on their own super expensive Windows10 phones.

Probably MickySoft is actually pretty happy about this development as a lot of copies of MS Word and MS Excel will be sold with these docking station software sets.

AND IT IS CLEAR TO ALL THAT A QUALCOMM 835 CAN BE A DESKTOP CHIP SET, ditto for a Samsung Exynos 8895, and ditto for one of the new 7nm Mediatek 12 core chipsets when they get here  ......  ditto for an Apple A-11x chipset ......   and it is also clear that any one of these phones can do it a lot cheaper than an old style Wintel desktop or laptop can do it.    Please remember, you will be able to do this with mid-line phones next year, and with all phones within another 2 years .....

MICROSOFT COOPERATING WITH THE LICENSING OF THEIR OFFICE SOFTWARES, including making display fit modifications to these same MS products so they fit well into the DeX package really really breaks some new "flexibility" ground for MS.    It was a smart move for MS to do as it opens up an entire new market segment with MS getting into the ground floor by doing this modification package deal with Samsung.

Intel isn't so happy, they were left standing out in the rain way way out in the parking lot and Intel flat didn't even get invited inside the building to attend this dance at all.  

Look to see Intel to try to partner with somebody to get into this Phone / PC action space ASAP ......   each one of these units sold means one PC chipset Intel cannot sell and Intel cannot afford that volume loss very well right now.


=======================================


Intel hits the air raid siren, the big fog horn and all the brown vapor cannons all at once yesterday -- something has really really upset them over in Automotive/Server world.   One item was ARM saying their Big Little replacement task organizer along with Google's new neuron net style Tensor Flow AI processor unit allowed a paltry 28nm lithography Google custom built Tensor Flow unit (as shown) to run at 30x greater throughput on any properly learned task set compared to Intel's very best x86 based core i-7 14nm product running a standard Win 10 OS or a distro Linux (i.e. using a deluxe normal processor and normal OS programming).  

This shows the very first example of what will come after PC on the big job "power desktop" space.   This thing isn't state of the art lithography and the processor hasn't got much in mhz speed, but it handles 200 threads at once instead of 4 or 6 like Intel does.  

And this is the first design effort actually, all refinements are yet to come.  A prototype that worked so well Google built it all last year to replace a lot of their rack farm hardware (it uses much much less power and is 30x faster to boot).   The Gen 2 is on the way, btw.





This new Tensor Processor puts Intel so far down in the ground they likely need to jest go ahead and dig the rest of the way to China ......

                                   

                                               ....... as right now Intel's great white hope that they have been hooting so much about lately is Optane which only gives 2x to 6x performance boosts and that only to some very memory recall intensive operations  --- while using VERY VERY expensive x86 i7 chipsets and very large expensive Optane arrays.   That is only 194 times less throughput than this palm sized Tensor Processor unit running the new Google in-house 28nm chip design and Google's Tensor Flow operating system.

Google just blew up Intel's outhouse.     BOOM !!!

Intel is now suddenly planning to be breaking their chipsets up into smaller component pieces (now how is that again, since they aren't really SOCs right now anyway ???) and Intel has announced they are coming up with their own organizing AI software to allow the broken up pieces to be mixed and matched together as needed into an increased throughput product that learns by doing.

A classic "me too" knee jerk reaction from Intel on saying they are going to be doing something they really haven't got a clue how to go about doing.

       Intel's technical leadership must have been a sitting in the outhouse when the dynamite went off .....

CHANGE, she comes ........

   yup, this is a whole motherboard

https://www.nextplatform.com/2016/05/19/google-takes-unconventional-route-hom...

....... this means the very basic ideas that stand behind the old PC world have started whirling around at the center of the toilet bowl swirl
and are actually beginning the dropping away to never never land ......

https://www.extremetech.com/computing/247199-googles-dedicated-tensorflow-pro...
 
New terms for you to absorb .....

Neural Net, Machine Learning, Inference, Massively Parallel, AI

These explanations are not technically complete, but just a way for you to grasp what they are about in general.

  A single Tensor unit       4,000 racked Tensor units .....

Tensor Flow softwares deal with whole integers (whole numbers) and they do this round up/down as step #1 --- this helps a lot as 8 bits can contain most integer data, 16 bits can get almost all of it and 32 bits does hold it all.  64 bits is not needed, ever, except to hold memory addresses on Intel x86 based machines.  This round up/down trick simplifies all subsequent calculations TREMENDOUSLY and the end result comes out the same, with some very minor ending results variation which will be mentioned later on in this post.

The new machines are by basic design kinda like the neurons in your brain, they can connect in lots of different ways and can do a vast number and array of different things simultaneously.   (Neural Net, get it ??)    Alpha Go (show to the right above) the AI that always beats the best humans at Go, the most complex game we own has 1.5 times the potential "neural connects" than a human brain has protein neuron connection pathways ....

Like you, the AI have to learn the ins and outs a new task --- which takes them about as long as our old familiar mainframe computing tasks used to take.  This is called Tensor flow learning. (Machine Learning, get it ??)   It too is a massively parallel activity, and a Tensor Processor can do both machine learning and then do the reflex "inference" execution using that learning.  

The same card can learn new things, then do them inside itself or do them across a bunch of networked cards -- by sharing what was learned with other cards.

Once learned, these simplified and tuned execution tasks (inference) can become "reflex" fast as no learning or modification is needed.   Many many many reflex type "inference" tasks can go on at the same time ......  200 things at a time can be done per card right now with this first very basic Tensor Processor.  (Massively Parallel, get it ??)  When you rack 4,000 of them the power goes up by the total combination possibilities = 4,000²⁰⁰ so the available compute range is HUGELY HUGE.   What used to be considered large tasks like graphics and video become reflex (inference) fast and execution speed is limited only by how fast bus speeds and how fast dynamic RAM memory speeds are on your equipment.  (Google spent the $$$ there as that was where the bang for the Tensor bucks really was).

And here is where Skynet creeps in ..... given enough repetitions, the slight natural ending variations in execution caused by the up/down rounding determine which path of execution is quicker/better, and that "better" execution pathway then becomes the norm.  (AI tuning, get it ??)  Artificial Intelligence learning and on the fly inference tuning means AI can get better and better at tasks as time goes on.      This process takes whole 100's of milliseconds per iteration it does, like forever and a day in computer time, huh ?   It also takes many many repetitions to sense those finer improvements.  So, AI surely CAN learn from its failures and its poorer results, and it never repeats them again as it has perfect memory.



...... and yes, preventing Skynet and HAL are real topics that these guys actually do talk about at AI Conferences as AI grows and becomes more complex.

DUNE had AI too -- AI took over for millennia and the Butlerian Jihad that fought the Machine Minds resulted in "Thou shall make no machine in the image of a human mind" as Dune Universe Rule #1.

Never ever even consider building a "thinking machine" inside my Empire .... ever.
I will immediately burn your world with atomics and stone burners if you do, such that the "thinking machine" infection cannot leave your world to infect others.
Edicts of Empire     Emperor Paul Muad dib

..... and so, what do we go do right off the bat ???   We make a goal out of building an AI that can beat our very best Go players in the world, learning from each contest and game until it is now arguably unbeatable by any human intellect.  

And the idiots in our Military see this wonderful campaign planning tool when they see Alpha Go beating humans at Go ..... so they want to go have their very own top secret AI running both wheeled and 4 footed warbots (machine gun & rocket equipped) and small efficient unmanned automated tanks and automated missile firing hover drones to go play their silly war strategy games with.

Arrugh, are you guys complete arsehole idiots or what ??   SKYNET anyone  ???

https://www.extremetech.com/computing/247199-googles-dedicated-tensorflow-pro...

https://www.cnet.com/news/elon-musk-worries-skynet-is-only-five-years-off/

http://www.huffingtonpost.com/2015/06/22/skynet-real_n_7042808.html

  This entire old style rack farm has throughput like the 4,000 card Tensor rack shown up above.

Skynet being born .....
when the repair guy on the ladder goes away ....  
and the machine begins to repair itself directing a robotic fix-it machine .....
which then installs some self-designed improvement modifications that the machine made up for itself .....
hey, then you got you some real potential Skynet type action headed your way .....
because you, Human, you were both too lazy AND too stupid .....
and you didn't watch out to see what your brainchild was doing.



=================================================================



You personally have a little bitty tensor flow based AI "potentially always sitting" in a rack at a Google processing facility.  This AI that comes into being (gets loaded) whenever you tap the mike button on your android phone or give the phone a voice based command.   This AI knows your voice patterns, your common commands and recent destinations and has a big leg up on anything new you might choose to command it to do.   If you give your AI a unique name and command it by that name using your voice, you can even use somebody else's phone to access your AI.

Get used to it, it is a new brave world out there ......

Far Eastern cell phone suppliers are all reporting a declining market this spring.   They are all now attempting to diversify way from Consumer Electronics as that market is now TOTALLY SATURATED world wide.    

It is now a mid-range pricing "upgrade or replacement" market world wide.   Premium displays and processors have moved down to mid-range as the new 10nm Qualcomms and Samsung chipsets define the new premium zone now.

Qualcomm and the others have taken to suing each other right and left in a feeble attempt to milk the last bit of juice out of the no longer rapidly growing Phone industry.

It is interesting to see all the brand new crop (first time ever) of Chromebooks and Android laptops popping up all over the Far East as the phone boys go looking someplace else to make a buck in 2017.

https://liliputing.com/2017/04/intel-cancels-entire-developer-forum-program-ahea
d-summers-idf17.html



IDF 2016 --- the very last developer conference ever on Intel processors and support bits and pieces.
Chicken and egg time again, who actually quit first, Intel or their developer base?


Intel cancels the entire Developer Forum program ahead of this summer’s IDF17

Intel says the move comes as the company is moving away from a focus primarily on PCs to one that is producing a wider range of products including automotive technologies, Internet of Things products, wireless communications, artificial intelligence, and storage, among other things.



AI -- invented it all by yourself last week, huh?    Oh, you went and bought yourself a little AI company last week.  I guess that means you invented it all by yourself six month ago when they say they invented it ....... right .......      

Storage, does that mean Optane ???   As invented by Micron, that 3D-XPoint memory stuff, is that what you mean?          

What was it again Intel that YOU actually invented in the last 5 years or so ???????

Oh yes, the infamous "Microprocessor Density Metric" PowerPoint slide.    I had forgotten all about that little gem.

Google tests second Quantum Computing chipset

https://www.technologyreview.com/s/604242/googles-new-chip-is-a-stepping-ston...


Read the article if you have any interest.  Quantum Computing doesn't use silicone gates, it uses quantum states of matter (real atomic level sorta stuff).   Google and IBM are working in the field and have made great strides making it real inside of the last year.

Smart long money says we will begin to depart "traditional" silicone inside the next 3 years as better stuff (Tensor Flow and now Quantum Computing) is showing practical progress towards real commercialization.

Google is building and using Tensor Flow Processors inside its own server farms right now (for over a year now) so that tech is fully real right now.  

Google also has a strong R&D effort going on Quantum Computing.   This year they plan to make up a Qputer that is more powerful than ALL the supercomputers in the world added together.

That is 12 high speed diode laser connectors that are needed to just move data in and out of the current Google 6-Qbit Quantum Computing chip.    Next one being built is a 50 Q-bit computer.  

The one after that is Unity, the 100 Q-bit capacity rig that is supposedly equal to all the rest of the computers that ever were built all stacked up together .....

Change, she comes ......