Lisa Su's dirty little "secret sandbagging" during her CES presentation.   Lisa held up a TSMC 7nm lithography Vega 10 unit APU (complete with enlarged AI block and wider data paths) matched up with a TSMC 7nm lithography 8 core CPU block and then simply didn't say one more word about it.

All she spoke to was a mild one upping of the rather disappointing 11-12nm Intel "10nm" performance as promised by Intel in the "end of 2019 in the far distant future".   Lisa Su was simply doing some classical sandbagging, in other words .......

One potential reason for this sandbagging reticence is large existing stocks of older AMD units that are not moving as quickly as were originally hoped ---- upgrading your own inventory for the 3rd time in two years might tend to hurt your own financials if that is done in a precipitous fashion.

AMD needs to sell off their channel full of old stuff before replacing it whole hog with much more desirable items.     Once the much better stuff is out all demand for the older style stuff ends abruptly ......

Intel's ongoing production shortfalls will continue to move AMD's older units (especially when the older units are properly priced) but it needs some time to work it all through as the whole PC market is somewhat soft at the moment.   TSMC will also get even better at 7nm full production during the meantime, increasing AMD's production yields and lowering the net price per wafer accordingly.

AMD can also continue to work on their 5nm stuff, getting it ready as well.   7nm is likely to be a short lived node anyway since the same equipment can run 5nm parts ...... this is both known and proven real by both TSMC and Samsung who already own the direct burn EUV lines that can do this trick.

Note please that Samsung and IBM just made a contract to run IBM's 5nm stuff at Samsung, this 5nm agreement is about the same cooperative agreement as they inked 2 years ago about 7nm lithography.



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



https://www.anandtech.com/show/13829/amd-ryzen-3rd-generation-zen-2-pcie-4-ei...

Further "confusions" stemming from Lisa Su's presentation at CES

When Lisa shows the silicon set and talks about APUs coming out now and again in mid year she is showing and talking APUs (combined graphics and processing units).    These are the new AMD laptop chipsets that are rolling out right now as we speak ......

When she is talking Mattisse, that is a desktop set up has no graphics unit in the big chip per se .......  it is a desktop processor that always has 1 each 8 core 7nm CPU chiplet built at TSMC.   Mattisse can then have a larger lithography 12-14nm I/O chiplet that is built at Global due to production constraints (or lower pricing, got to remember that).    The Mattisse CPU core chiplets are always the same 7nm TSMC as in APU, but the I/O chiplet on Mattisse is much much simpler and can be done at older lesser lithography levels very easily.

Remember, Mattisse and APU both swing two core filled die sets at this time, one is an I/O (and always has the on die rapid access buffer memory in it).   In the APU case the big chiplet includes Vega graphics cores (up to 10 of them).  In the Mattisse case the big chiplet does not include any graphics, just I/O and rapid access buffer memory.



Confusing?   Oh my yes -- now take these scenarios and add it to your confusion.  A failed sorted APU unit may be sorted and sold as a Mattisse graphic-less variant with the failed graphics seared off.  

On the 7nm Cpu chiplet side, sorted small CPU chiplets may have 1-4 non-functional compute cores seared off as well.

In full complexity of sorting, an I/O chip may start out intended as a 7nm TSMC based APU with up to 10 Vega graphics cores and an AI block inside the larger I/O chiplet on an AM4 socket silicon with 8 cores of 7nm TSMC CPU cores inside the smaller chiplet, and these CPU cores can be seared off as well to potentially lose all the the graphics and also lose one or more CPU cores in sorting.

So you got variants possible that have had graphics but lost them and had up to 8 compute CPU cores but lost some of those.    Or the variant can be built on purpose to never have had them in the first place if the demand is there and there aren't enough sorted units to meet the need.




WHAT IS LEFT HAS NOT BEEN SHOWN AND EXISTS ONLY OUT THERE IN THE FUTURE
(for sometimes for when Lisa Su decides it is needed).



This is the 12-16 core cpu (dual CPU chiplet) variant that is NOT NEEDED right now to keep up with Intel's best of the best i7-9970 8 core chipset --- Intel tops out at 8 cores in consumer at the moment and AMD will be content to outdo 8 core Intel using 8 better 7nm cores from TSMC.

This 12-16 core variant will not exist in Consumer until Intel actually goes there inside normal consumer computing.

NOTE:  This variant technically already exists in Server space, where Intel already has 16 cores per die (with 2 separate sockets per motherboard) and AMD has up to 32 cores per die right now using the chiplet technology (but it is on the larger single Epic server die set and socket scenario).

Intel is the one who likes to plug their old server die sets and sockets into Consumer space, they have done it several times already.

So far AMD likes to re-use the AM4 socket die set over and over, using smaller and smaller TSMC lithography in roughly the same sized chiplets to make it work again and again and again -- filling the space up with more and more cores as the lithography shrinks happen.

5nm chiplets are certainly out there in the future on the AM4 socket die, this is a no brainer coming from AMD.    Core counts will likely start at 8 and go to 16 and mebbe go to 32 potentially when 3nm comes around,  all staying in the AM4 socket size.

The AMD Epic server socket and die also will roll out with 5nm and 3nm chiplets when those waves begin, but your guess is as good as mine as to what the core counts will eventually be.  Many many, obviously --- 64 cores being the most likely next expansion.

Intel re-dies and re-busses their old motherboards at a bi-yearly pace, while AMD shows repeatedly they can use the same board and the same socket with just a BIOS refresh.

Some late night musing on the tech company year end roll ups given out in the last few weeks.

The overall tenor of the financial reports is very muted and very careful.   Tech overall looks to have a sub flat year for the first half of 2019, with financial warning flags being flown by TSMC and APPLE and everyone in between.   Shipments and earnings will be down, sharply.

Look to see people throughout the infrastructure to be buying on price/performance, not just pure performance.   Things look right grim for the overly pricey Intel stuff.

TSMC says they will ship only half the volume in first half of 2019 compared to what they did last year.  It may be better stuff, but there will be less of it shipped.

Also look to see everyone diligently search for COST ADVANTAGES over their competitors --- rolling products back to older cheaper lithography levels may become a repeated theme in 2019 as COST becomes KING.

Look to see smaller companies playing "skip this current node, jump to the next after this one when it gets cheap enough" sorts of thinking again, especially over in the Orient.

Choosing to stay where you are until it PAYS you to move again will become the new normal.   TRYING TO SURVIVE will be the name of the game for most small players, and many of them simply won't make it.

TSMC and Samsung have already bought the direct burn EUV production lines to go to 7nm, it is the same equipment that does 5nm memory right now and these lines are kept busy proving themselves out in 5nm memory production at both TSMC and Samsung.   This "over produced lots of 5nm memory will further depress the general pricing for memory products but those new lines simply have to run flat out producing something that can be sold just to make the payments on the new equipment.

As cheaper, better memory becomes more readily available, look to see builders use more forms of fast memory as it is a good way to build up some extra speed at "the same cost as what you were buying before".

Machine vendors are building alternative AMD mother boards for many existing laptop products so they can buy less expensive processors and components.  AMD style APU products will be preferred for this sort of thing as they are lower cost by their very nature.

I am watching new Chromebooks rolling over to AMD laptop APU processors in bulk as they are less expensive to make (they also have better performance, but that isn't why they are taking over, COST is the real driver here).

Remember, you can get nearly twice as many 5nm chiplets out of a wafer compared to 10nm chiplets.  So watch AMD optimize on this trick and roll over to 5nm as the total production of everything else drops sharply to free up the newest TSMC EUV equipment ......

7nm might be a very short lived node because the EUV production lines can make more money making a lot more 5nm chiplets on the same number of wafers than running the same wafers of 7nm chiplets.    

COST is KING, remember.

https://www.notebookcheck.net/When-cheaper-is-faster-Core-i5-HP-EliteBook-x36...

Buyer Beware of Intel latest Gen i7 and i9 chipsets -- they REALLY DO throttle down to sub i5 performance levels when installed in a laptop product ......

Nonetheless, we discovered some odd performance results between the two HP SKUs we tested. More specifically, the Core i7-8650U SKU would have slower CPU performance than the Core i5-8250U SKU. CineBench R15 Multi-Thread reveals that the Core i5 SKU can be 10 percent faster than the Core i7 SKU (493 points vs. 447 points).

A potential culprit could be throttling especially when dealing with small and super-thin convertibles such as our HP. To test this, we ran CineBench R15 Multi-Thread in a loop on each system with our results graphed below. Throttling occurs on each SKU and so the processors are fastest during the first benchmark loop before steadily declining thereafter. What's interesting, however, is that the Core i5 SKU manages to consistently outperforms the Core i7 SKU even after taking into account the throttling.

The takeaway message here is that the performance delta between Core i5 and Core i7 Kaby Lake-R can be very minuscule and variable depending on the system and implementation. In this case, users would be better off purchasing the Core i5 EliteBook x360 1030 G3 instead of the Core i7 option while saving a couple hundred dollars in the process.

This isn't the only case where we discovered faster performance from a cheaper Core i5 machine. The ThinkPad X1 Tablet exhibits similar results between its i5 and i7 SKUs as well.


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


First two AMD APU equipped laptops from HP are selling for an average $200 cheaper than the current Intel "kinda equivalent".

(and please note the AMD APU laptop units have much better built in graphics than Intel's best)

https://www.tomshardware.com/news/intel-earnings-4q-2018-desktop-shortage,384...

Intel Misses Revenue Targets and Issues Weak Outlook

Intel released its full-year and fourth quarter 2018 financial results today. Overall, the company posted yet another full-year revenue record of $70.8 billion, but its fourth-quarter performance fell short of analyst projections. The company also presented lower-than-expected guidance for 2019.

Intel's CEO search was top-of-mind as rampant speculation led many to believe that company would appoint its next leader during the earnings call. Unfortunately, Intel's interim-CEO Bob Swan dashed those hopes by saying the company is still searching for the right candidate, saying "I am convinced the board will close on a new CEO in the near future. In the meantime, we will not be distracted by the void."

Intel's notebook volumes dropped 10%, and desktop volume also fell 7% compared to the prior quarter, though the company has increased notebook volume by 4% overall compared to 2017. Desktops didn't fare as well, with a 6% decline in volume YoY, largely offset by an 11% increase in the average selling price (ASP).

Intel's growth in the data center segments cooled off, which the company attributed to large orders being processed earlier in the year finally being deployed. Industry analysts reported large orders earlier in the year as Chinese data centers built up stockpiles amid the looming threat of a trade war, so it's logical to assume that much of the new deployments come from that existing stock.

Slumping NAND prices hurt the Nonvolatile Storage Group's (NSG) profitability, so its 25% revenue growth on the year merely brought the unit to the break-even point. Intel CEO Bob Swan acknowledged that Micron had formally moved to purchase Intel's share in the IM Flash Technologies joint venture. Intel and Micron jointly produce NAND and 3D XPoint (Optane) memory through this venture, but Swan noted Intel has its own production facilities and long-term supply agreements with Micron that extend beyond the termination of the partnership.

Intel's projections for $16 billion in revenue for Q1 2019 is essentially flat year over year, and it projects a 1% gain in revenue to $71.58 billion in 2019.

Overall the company's record results and 2019 outlook are viewed as somewhat tepid in the investor community. Intel cited increased competition in several of its segments, along with the other factors listed above, as headwinds going into 2019, but didn't specifically mention AMD as the source.

Yeah, and you just dumped off 10% of your stock price within a 36 hour period, too .....      

https://www.extremetech.com/computing/284528-intels-10nm-cannon-lake-core-i3-...

https://www.anandtech.com/show/13405/intel-10nm-cannon-lake-and-core-i3-8121u...

https://www.zdnet.com/article/intel-details-first-10nm-cannon-lake-chip-comin.../





Yep, the "wonderful new Intel 10nm" is second to the bottom -- almost dead last --- only winning one slot above a dated and faded AMD "Internet of Things" Chromebook chipset.  

Yes, sadly Intel 14nm still clearly outperforms Intel's 10nm on both throughput and power consumption, while Ryzen 3 kicks all of Intel's various sorts and kinds of old lithography types in the arse quite soundly.  

AMD and TSMC are getting better at over twice the rate that Intel is "claiming" that they are going to do, and are actually getting better like 5 times faster than Intel actually ever achieves ......

What can be said for the Future of Intel 10nm processors as tested in a Lenovo product that also comes in other Intel 14nm chipset forms for easy and accurate comparisons?   A very fair "Apples to Apples comparison" is very easily do-able in this case and it has been done by all three reporting bodies.

Intel 10nm SUX ROCKS big time .......  Do not buy it !!!

Timing says that all three of the investigative web sites listed above all placed early orders to get a i3-8121u Intel 10nm chipped Lenovo built Ideapad so they could be the very first to review "the future of Intel as a chipmaker".

The laptops all showed up at the same time and the reviews have been written up and pushed out to the public with 24 hours of each other.

Do you remember me saying:

Do not buy a 10nm Intel chipset !!!!     yep, buy a 7nm Ryzen 3 chipset instead !!!!

Intel REACTS to the last few days of adversity ....

Intel counters the fumbling image they have projected of late by going and HIRING acting CEO Swan as Intel's permanent CEO ----- this does not give him any chops as a visionary whatsoever, but it legitimizes his rule over the company's future.

Intel uncorks a promised 28 core 14nm Xenon server processor to counter AMD's leaked 12 and 16 core processors in the consumer side and to try to EQUAL the 28 and 32 core Epyic server processors that AMD showed off at the last big server show.

Same issues remain for Intel --- AMD is showing real "building it now" stuff and Intel is pop talking vague future plans that are predicated on new plant buildings that have not broken ground yet and completely fictional future orders for ASML EUV direct burn lithography lines that are not even PO written yet (and Intel is not even showing up on the waiting list that ASML keeps for their new built production lines).


                   


REMEMBER  --- "10nm" Intel just got reality check reviewed, and it  SUX ROCKS  when benchmarked by THREE DIFFERENT test houses.

 
Intel was not alone in stumbling over 10nm.


Everybody stumbled at 10nm to some degree -- Intel just got pig headed for 6 long years about going past 10nm and moving on to something better.   Intel had no pressing competitor at the time and no pressing reason to spend any money on getting better.

Now that reluctance is slowly evaporating and Intel is starting talking plans that lead to 7nm and possibly lower.

This is good, but the timing sucks --- everyone else IS ALREADY AT 7nm and has plans to go to 5nm on CPU production starting late this year.

Intel is still stuck at 14nm, has failed yet again to get their best 6th generation 10nm to work out as good as their current 14nm is doing right now and only has some 2 year out "build a new building for it in 2020" style plans for new ASLM lithography lines that they need to use to move forward.

Lisa Su the visionary CEO at AMD shows real production chipsets at major shows that just EAT INTEL UP ALIVE, and smiles because she can afford to wait for her relatively tiny warehouse stocks of older chips to move on out through the sales channels before scheduling in the new much better chips as replacement inventory.

This entire year is an off year for electronics, a technically void year where no one is buying very much of anything supposedly.

Microsoft is giving Intel the STINKING BROWN SHITE STUDDED CROWN for this particular situation, as Intel's failures to advance their technology and their badly failed efforts in "predictive processing" have polluted the whole of computing with Meltdown and Spectre security risks that  SLOWS DOWN COMPUTING AS A WHOLE BY 30% simply to partially mediate these two issues.

Microsoft cannot show clean hands in this, as MS's gross programing overhead and very slow processing is another major cause of slow processing as well.

Clear Linux (by the Wintel boys) is an excellent example of MS and Intel proving to themselves that they could shite can their whole historically weighted x86 stack and still be able make up a much faster, better OS out of free small and light FOSS Linux bits and pieces.  

Now, after having done it themselves as a trial, MS is a believer ......

Seeing what MS and Intel have done with Clear Linux, Linus Torvalds himself is rallying all the FOSS troops for a wide scale Linux clean up again -- they do this about every 8 years or so when somebody shows the main Linux boys that they have gotten slow and fat and complacent around the edges again.   Last time they did this they were agreeing to mainline the faster FOSS Android code which would make the Android version into the Linux OS general standard as it was faster, neater, smaller and cleaner than the old desktop Linux code that the Distros were using at the time.

Having Mickey and Chipzilla show you your sins is simply too much to take, though, so it is time to clean Linux up again ......  

But once again it shows that OS lightness and speediness is currently as important as expensive hardware to get the whole show to move along quickly.

https://www.youtube.com/watch?v=KLz8gC235i8    it is a YouTube, click on it and watch it

I mentioned this when it first was announced, that Amazon was planning on dumping Intel Server Chipsets in favor of their own home grown ARM server chipsets.

What was not clear at that time was that this Amazon ARM based chipset was going to be FASTER than Intel's best Xenon chipset.

And it does it with ~50% less power consumed.   These are the things that count the most in Server World, relative speed and power consumption.

So, suddenly overnight we all realize that Amazon has suddenly become the world's largest manufacturer and user of large very powerful ARM chipsets.


                                                                             


So, Google runs on their custom water cooled TPUs and Amazon runs on their custom ARM server chipsets and Microsoft runs on their .........  HUGE very expensive bone stock Xenon double socket rig ups from Intel.    

China has copied the Amazon chipset (Chinese industrial espionage strikes again) and is using those in China right now along with the Chinese produced AMD Epic server stuff they licensed legally from AMD like 2 years ago.

Where is Intel in all of this ???   Still stuck at 14nm .......   and still making same ol' same 'ol very expensive stock Xenon double socket rig ups from the Intel Inside lovers out there.


   


And Intel, who is it exactly that is going broke buying all of your expensive Xenon chipsets again?    Is Mickey the only one left other than yourselves ???


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


ARE THE BIG BOYS (AND THE MID-SIZED BOYS) EACH PAYING CLOSE ATTENTION TO THE VARIOUS FOSS RISC-V "FREE" NO LICENSE FEE NEEDED CHIPSETS THAT ARE BEING MADE UP BY ALL THE OTHER VARIOUS FOSS PLAYERS?   Yep, they sure are.  

Note please that the current crop of RISK-V chips are starting out made for small, somewhat mundane functions that are often used by hard drives, routers and data switches and other "motherboard housekeeping" types of simple smaller stuff ---- small but very prevalent in large numbers in a great many different products so a whole lot of money is getting saved by using these RISC-V chipsets instead of buying Qualcomm or ARM or Broadcom branded chipsets to do these same tasks.


https://riscv.org/members-at-a-glance/   ........  sure looks like they are paying close attention, doesn't it?


https://www.youtube.com/watch?v=L8jqGOgCy5M      ........ it is a YouTube, click on it


In today's environment where COST IS KING, these RISC-V processors as done so far can be very thrifty to utilize especially over time as you only pay to develop it once and they can cost peanuts per wafer to re-produce at TSMC on programmable EUV direct burn cells.   TSMC can just call up the direct burn EUV 7nm lithography program on their ASML direct burn cells and burn as many or as few wafers as you need at the moment ......   Quick Changeover Agile Production at its best.

https://hackaday.com/2019/02/13/western-digital-releases-their-risc-v-cores-t...

Western Digital is PROUD to have released to FOSS all of their RISC-V chipsets that they designed and proved out in their own hard drive production.   The inference is that everybody else should be doing the same required "release it to FOSS" thing with their RISC-V chipset designs, eventually building up a FOSS library of proven RISC-V chipsets that can be re-purposed or slightly modified and used for other uses.

Last year, Western Digital made the amazing claim that they will transition their consumption of silicon over to RISC-V, putting one Billion RISC-V cores per year into the marketplace. This is huge news, akin to Apple saying they’re not going to bother with ARM anymore. Sure, these cores won’t necessarily be user-facing but at least we’re getting something.

This could be a big thing in a few years, when somebody posts a "proven by production" laptop grade RISC-V FOSS CPU design and proves it out in production ......




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



Read the line of text immediately above this one, grin a little bit and read on ..... read on.   It's just likely a small time bubble, a quick little time warp with "a few years" instantly passing in just a few hours.




https://www.seeedstudio.com/Sipeed-MAIX-I-module-WiFi-version-1st-RISC-V-64-A...

$8.90 cents for a motherboard the size of a large postage stamp bearing EIGHT tiny RISC-V FOSS chiplets and a dozen accelerator blocks -- something has changed in computing land, something really basic has changed and "the freebie mix and match" is now moving along now REALLY REALLY FAST.

....... right on up your leg.       zowie !!!

The speed of these things popping up is UNREAL ---- the question being will it decimate Qualcomm and ARM in general inside a year or so because is is Free, FOSS and faster than spit in creating new products .....

Sipeed MAix: AI at the edge

AI is pervasive today, from consumer to enterprise applications. With the explosive growth of connected devices, combined with a demand for privacy/confidentiality, low latency and bandwidth constraints, AI models trained in the cloud increasingly need to be run at the edge.

MAIX is Sipeed’s purpose-built module designed to run AI at the edge, we called it AIoT. It delivers high performance in a small physical and power footprint, enabling the deployment of high-accuracy AI at the edge, and the competitive price make it possible embed to any IoT devices.

As you see, Sipeed MAIX is quite like Google edge TPU, but it act as master controller, not an accelerator like edge TPU, so it is more low cost and low power than AP+edge TPU solution.

MAix's Advantage and Usage Scenarios:

MAIX is not only hardware, but also provide an end-to-end, hardware + software infrastructure for facilitating the deployment of customers' AI-based solutions.
Thanks to its performance, small footprint, low power, and low cost, MAIX enables the broad deployment of high-quality AI at the edge.

MAIX isn't just a hardware solution, it combines custom hardware, open software, and state-of-the-art AI algorithms to provide high-quality, easy to deploy AI solutions for the edge.

MAIX can be used for a growing number of industrial use-cases such as predictive maintenance, anomaly detection, machine vision, robotics, voice recognition, and many more.

It can be used in manufacturing, on-premise, healthcare, retail, smart spaces, transportation, etc.

In hardware, MAIX has the powerful KPU K210 inside, it offers many excited features:
1st competitive RISC-V chip, also 1st competitive AI chip, newly release in Sep. 2018
28nm process, dual-core RISC-V 64bit IMAFDC, on-chip huge 8MB high-speed SRAM (not for XMR), 400MHz frequency (able to 800MHz)  

KPU (Neural Network Processor) inside, 64 KPU which is 576bit width, support convolution kernels, any form of activation function. It offers 0.25TOPS@0.3W,400MHz, when overclock to 800MHz, it offers 0.5TOPS. It means you can do object recognition 60fps@VGA

APU (Audio Processor) inside, support 8mics, up to 192KHz sample rate, hardcore FFT unit inside, easy to make a Mic Array (MAIX offer it too)

Flexible FPIOA (Field Programmable IO Array), you can map 255 functions to all 48 GPIOs on the chip

DVP camera and MCU LCD interface, you can connect an DVP camera, run your algorithm, and display on LCD

Many other accelerators and peripherals: AES Accelerator, SHA256 Accelerator, FFT Accelerator (not APU's one), OTP, UART, WDT, IIC, SPI, I2S, TIMER, RTC, PWM, etc.
MAix's Module

Inherit the advantage of K210's small footprint, Sipeed MAIX-I module, or called M1, integrate K210, 3-channel DC-DC power, 8MB/16MB/128MB Flash (M1w module add wifi chip esp8285 on it) into Square Inch Module. All usable IO breaks out as 1.27mm(50mil) pins, and pin's voltage is selectable from 3.3V and 1.8V.

CHANGE, she comes .......
 
There is nothing stopping someone from posting a CPU chiplet design, a GPU chiplet design and an I/O block design then the tiny stamp sized motherboards can act kinda like an AMD RYZEN, jest building up to the power level that you need ........


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


https://liliputing.com/2019/02/odroid-n2-single-board-computer-coming-soon-fo...

Lookie, an ARM based 6 core $63 Computer shipped complete with Ubuntu OS    ODROID-N2 single board computer coming soon for $63 and up (Amlogic S922X processor, up to 4GB RAM)