Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
I thought Samsung skipped 20SOC to go to 14nm?
Only for its external foundry customers but it will use 20nm internally.
The A15/A7 frequencies look rather low.
Welcome to planar 20nm .
By following I mean the incorporation of similar instructions in other ISAs and subsequent implementations. On slide 19 of the Sparc pdf I linked they give the order of improvement over T5 in query decompression as 10. Now T5 had a weak core so probably 3-4 over their last M6 gen core. Generally 32-core 20nm M7 has a chip performance improvement over their 12-core 28nm M6 of 2.9-3.5 (slide 12) for 2.67 more cores. The individual cores are 8-threaded and look in the Silvermont class on first viewing and so the M7 should roughly have more or less a similar throughput performance as the top-bins of Power 8 or Xeon although about half in single-thread.
http://www.setphaserstostun.org/hc26/HC26-12-day2-epub/HC26.12-8-Big-Iron-Servers-epub/HC26.12.820-Next_Gen_SPARC_Phillips-Oracle-FinalPub.pdf
It all depends on how performant Oracle's hardwiring of query execution will be in applications and benchmarks. If it makes a real noticeable difference then expect others to follow in 4+ years time.
Hotchips Server/HPC pdfs
http://www.setphaserstostun.org/hc26/HC26-12-day2-epub/HC26.12-8-Big-Iron-Servers-epub/HC26.12.820-Next_Gen_SPARC_Phillips-Oracle-FinalPub.pdf
http://www.setphaserstostun.org/hc26/HC26-12-day2-epub/HC26.12-7-Dense-Servers-epub/HC26.12.721-SPARC-cache-Sivaramakrishan-Oracle_final_2.pdf
http://www.setphaserstostun.org/hc26/HC26-11-day1-epub/HC26.11-1-High-Performance-epub/HC26.11.120-SPARC64-Xlfx-Yoshida-Fujitsu-rev1.2.pdf
http://www.setphaserstostun.org/hc26/HC26-12-day2-epub/HC26.12-8-Big-Iron-Servers-epub/HC26.12.810-POWER8-Mericas-IBM.pdf
http://www.setphaserstostun.org/hc26/HC26-12-day2-epub/HC26.12-7-Dense-Servers-epub/HC26.12.730-%20OpenPower-Gschwind-IBM.pdf
http://www.setphaserstostun.org/hc26/HC26-12-day2-epub/HC26.12-8-Big-Iron-Servers-epub/HC26.13.832-IvyBridge-Esmer-Intel-IVB%20Server%20Hotchips2014.pdf
http://www.setphaserstostun.org/hc26/HC26-12-day2-epub/HC26.12-7-Dense-Servers-epub/HC26.12.741-C2000-Atom-Burres-Intel-Avoton%20arch%20-%20hotchips.pdf
http://www.setphaserstostun.org/hc26/HC26-11-day1-epub/HC26.11-4-ARM-Servers-epub/HC26.11.410-Opteron-Seattle-White-AMD-HotChipsAMDSeattle_FINAL.pdf
http://www.setphaserstostun.org/hc26/HC26-11-day1-epub/HC26.11-4-ARM-Servers-epub/HC26.11.430-X-Gene-Singh-AppMicro-HotChips-2014-v5.pdf
http://www.setphaserstostun.org/hc26/HC26-11-day1-epub/HC26.11-1-High-Performance-epub/HC26.11.110-SX-ACE-MOMOSE-NEC-v004.pdf
all from http://www.setphaserstostun.org/hc26/HC26.index.htm as the originals from http://www.hotchips.org/hc26-full-program/ now require passwords.
AMD Hotchips pdfs e.g. Kaveri/Opteron Seattle
http://www.setphaserstostun.org/hc26/HC26-11-day1-epub/HC26.11-2-Mobile-Processors-epub/HC26.11.220-Bouvier-Kaveri-AMD-Final.pdf
http://www.setphaserstostun.org/hc26/HC26-11-day1-epub/HC26.11-4-ARM-Servers-epub/HC26.11.410-Opteron-Seattle-White-AMD-HotChipsAMDSeattle_FINAL.pdf
description http://www.hotchips.org/hc26-full-program/
http://www.anandtech.com/show/8382/samsung-announces-exynos-5430-first-20nm-samsung-soc
http://www.anandtech.com/show/8379/samsung-launches-the-galaxy-alpha
More interestingly the international version of the device should sport LTE-A category 6 with help of an Intel XMM7260 modem. This would be the first device announced with Intel's new LTE modem and mark a break from Qualcomm's dominance in the sector.
This was Barrett's take at the time and he was right in the end only it was the P6 Pentium Pro's descendants that became the ubiquitous architecture not P7's Itanium.
http://www.embeddedlinks.com/chipdir/oth/p7.txt
One of the architects of Willamette told me they debated back
and forth whether to make the original P4 uarch 64 bit or not.
This was at a time before 1) AMD disclosed the fact that it was
working on x86-64 and 2) Merced ran into development problems
(all major OEMs were still publicly pledging to support IPF). If it came down from the top that Itanium would be the only 64 bit architecture supported by Intel then there would have been
nothing for the P4 team to debate.
Would it have been activated though if PA-WW -> EPIC -> Itanium had proved a big a success as hoped for in 64-bit and overall performance ? Prescott's 64-bitness was only revealed/activated once 64-bit Opteron started getting some traction. The account of Bob Cowell showed there were definitely two opposing camps in the design space, x86 and Itanium, that acted like ruthless internal competitors. Maybe Craig Barrett's plan was to let them both do their own thing and see who came through for him in execution as the one link I previously posted from him around that time showed him to be cautious when questioned about future architecture path albeit leaning in Itanium's direction. Definitely Itanium would have been the preferred winner as it was totally proprietary and would have cut out all the x86 licensees (e.g. AMD/VIA/IBM) from the leading edge loop ... forever.
I think theoretically in 1994 when the original HP-Intel PA-WW joint agreement was signed it was envisaged (and mentioned in PRs) that it would be so superior that it would eventually replace all x86 and Risc (both HP's and its competitors) and so there would be no need for a 64-bit x86 as the architecture could naturally obsolete itself at 32-bit. However once the first EPIC/Itanium silicon started coming back that notion was quietly dropped and x86 evolved as you had noted.
Catching up compared to the wimpy A8 cores they once had, sure, but still they are all under the AMD x86 Bulldozer level of performance never mind any Core with any serious clockspeed. Considering how flat out they are all going in terms of issue width and design exoticness on brand new architectures on a brand new 64-bit ISA the total end result is poor, remember in the 2000s AMD arguably slightly surpassed Intel's best when K7 and K8 came out, there is nothing on that level now from the ARMy and Intel can contain the threat with sku design/price manipulation at its leisure. Frankly a Power 8 desktop would be of more threat to Intel now than the entire efforts of the feeble ARMy. Yes, there will be parts of the ARMy above Atom but that is no big deal in the long run if Atom has profitable volume in its performance price range.
What is also interesting on slide 14 is the SpecINT2000 comparison between 1.3 GHz A7 and 1.4 Ghz Haswell Celeron. The Haswell Celeron has 30-40% better performance yet they have identical GeekBench3 numbers which shows just how unreliable GeekBench is remembering also this is third party Nvidia's numbers not Intel's.
http://www.hotchips.org/wp-content/uploads/hc_archives/hc26/HC26-11-day1-epub/HC26.11-2-Mobile-Processors-epub/HC26.11.234-Denver-Darrell.Boggs-NVIDIA-rev4.pdf
from http://www.hotchips.org/hc26-full-program/
Right, that solves the mystery of its performance profile, it's an in-order cpu like Itanium/Transmeta. On slide 14 the Bay Trail it is comparing against is the 1.6 Ghz N2910 Celeron which is 2/3 the clock speed of a high end Bay Trail so you have to add 50% to the white bars to get an apples to apples comparison and Denver looks not much better than A57 to me. Performance wise it looks an evolution to me for ARM not a revolution. Cherry Trail and A57 will be its competitors not Core-M e.g. they are claiming 50% greater SpecINT2K for a 10% higher clocked Denver than the A15 in 32-bit K1. This is not going to bother any Core clocked 1.5 GHz or over and remember Core goes up to 4.4 GHz now. At 2.5 GHz and 7-issue Denver also has very little growth potential in either direction. Sorry, it might be a step up for ARM but it's a big yawn by x86 standards.
Intel defined logic shrink as (gate * metal) pitch shrink and by that definition the 14nm shrink was a good one, 51%. Primarily it will allow Intel to catch up and possibly lead in gpu as well as increasing its cpu performance and performance/power lead.
If you think beating a purposely slugged clock limited Haswell Celeron is the height of achievement for a new 7-issue chip then you go for it but really at that issue width it should have higher performance/clock than Haswell not much less so something is not quite right with that design.
How many Haswell SKUs can fit in a fanless tablet like the Nexus 9?
Any of the -Y ones.
Denver also completely destroys Silvermont.
In your dreams. How do you think this Denver chip that has impressed you so much will do against Broadwell Core-M or Cherry Trail ?
That has a 1.4 GHz clockspeed, so for a 2.5 GHz Denver that is no big deal, in fact it is poor considering its issue width.
http://ark.intel.com/products/75608/Intel-Celeron-Processor-2955U-2M-Cache-1_40-GHz
I see many reasons to expect at least one of the Big Three will fall off of the balance beam in the next 5 years.
It won't be Intel in that time frame because Intel already has a 2 year advantage over its opposition (14FF) and pretty confident enough to give detailed stats about its next generation (10FF) while the competition only has 20nm planar confirmed as a genuine production worthy process at this moment. So the immediate question marks in the next 5 years are on TSMC, G-F and Samsung with INTC investors worries only about design execution.
What happens when/if the foundries catch up to a genuine 10FF, i.e. not just using 14nm BEOL will decide the 5-10 year future after then. There are encouraging signs that EUV might just be ready for 7FF so Intel could serenely march on to continue having a lead in that time period without missing a 2-year step after 10FF.
Stock now above all moving averages again (up to a few years out) which is a very benign situation .
Awesome detailed historical perspective in your post showing how the balance between winners and losers is so precariously balanced in every era and change of technology. I don't think many people/investors realize how close silicon lithography is to reaching the end of its long road and how a radical new technology will soon be required to increase the number of switching devices on a computing device. Forward projected earnings for any semi company that go beyond 6-12 years is fraught with this uncertainty although very few realize it. Where Intel will fit in the future is anybody's guess but its impressive successful historical R&D should place it in the vanguard of any future switching technology but that is not guaranteed just likely.
Windows systems set to be merged by Microsoft [Intel x86 a clear winner going forward]
http://www.bbc.co.uk/news/technology-28440288
Universal apps
Mr Nadella did, however, discuss the benefits of merging the Windows Phone OS with the other versions of the system, noting that it would swell the number of programs available to run on the firm's handsets as a result of having "one ecosystem".
"One of beauties of universal Windows app is it aggregates for the first time for us all of our Windows volume," Mr Nadella said.
"An app that runs with a mouse and keyboard on the desktop can be in the store, and you can have the same app run in the touch-first [mobile devices].
"[It] gives developers the entire volume of Windows, which is 300 plus million units as opposed to just our 4% share of mobile in the US or 10% in some countries."
The move puts Microsoft at odds with Apple and Google, which are both pursuing separate strategies for laptop/desktop computers and mobile devices - Apple with Mac OS X and iOS, Google with Chrome and Android.
But it does bring Microsoft closer to another OS developer, Canonical, which has promoted the idea of its Ubuntu system powering both phones and desktops. Canonical previously highlighted that one benefit of this strategy was that a handset could double up as a low-power desktop PC if it was plugged into a monitor and connected to a mouse.
It also paves the way for Microsoft to introduce its voice-controlled personal assistant, Cortana, to PCs. Mr Nadella mentioned the app several times during the call.
There are no one-way bets in global finance, but Japan's stock market comes close. The authorities are about to funnel large sums into Japanese stocks openly and deliberately under the next phase of Abenomics, both by regulatory fiat and by purchasing the Nikkei index directly with printed money.
Prime minister Shinzo Abe is unshackling the world's biggest stash of savings, the $1.3 trillion Government Pension Investment Fund (GPIF). Officials say the ceiling on equity holdings will rise from 12pc to around 20pc as soon as August, opening the way for a $100bn buying blitz.
Fund managers are suddenly in a race to get there first. Japan Post Bank - where Mrs Watanabe dutifully places the family money, confiscated from her Salaryman each month before he can spend it - is itching to rotate more of its $2 trillion holdings into equities before inflation pummels the bond market. So is Japan Post Insurance, no minnow either at $850bn.
Mr Abe's move comes sooner than expected and amounts to a market shock, though nobody should be shocked anymore as he keeps doubling down on the world's most radical economic experiment.
The Nikkei index stalled in December after rising almost 100pc since September 2012, even though the Bank of Japan (BoJ) is still showering the economy with money, buying $75bn of bonds each month. The BoJ's balance sheet will reach 70pc of GDP by March 2015, three times the US Federal Reserve's.
The bank is well on its way to defeating deflation under Haruhiko Kuroda, the first governor to launch QE with enough force to break the vicious circle. It is not yet a done deal. Three board members remain sceptical, and one (Takahide Kiuchi) thinks the undertaking is doomed to failure.
Yet core inflation has turned positive for the first time since 1997. The consumer price index rose to 1.4pc in April, adjusted for taxes. Real interest rates have plunged through the floor, transforming Japan's debt dynamics and the role of money in the economy.
Real borrowing costs on 10-year government bonds are no longer among the highest in the industrial world. They have fallen to -0.8pc. By this mathematical magic, public debt is levelling off near 245pc of GDP, after rising from 216pc in 2010. Japan has, for now, averted a compound interest trap.
http://www.telegraph.co.uk/finance/comment/ambroseevans_pritchard/10893607/Japan-to-keep-printing-money-for-years-to-come-so-learn-to-enjoy-it.html
The latter.
You don't think there will be a big demand for Core M ?
Pea Brain Ninny does not seem to realize Core M is a 14nm Broadwell variant.
That's because they are going to use Knights Landing in a standalone socketed form in some of the nodes for bandwidth reasons (and Haswell Xeons in others) so seriously how do you think a tiny individual gpu core would compare with the Silvermont based core in Knights Landing under the same standalone conditions ?
That's one alternative if you can pull it off but the Intel branded phone seems to have born and died with Medfield/Clovertrail+. Another alternative is gaining OEMs trust with sharing of the gross margins and Intel is doing that with its current low pricing of Atom. There has to be a happy medium somewhere as mobile products are incremental to Intel rather than crucial so pricing flexibility is available.
re: Avram Miller interview; great historical read back to the days when the PC/network computer was still being defined rather than just being refined. Embedded in all of it was this snippet about mobile OEMs which I have suggested before is the kind of inertia Intel has to overcome in getting them to use their chips ...
Elliot: Interesting observation. I’ve dealt with a lot of companies in the mobile space, and I’ve had a number of them talk to me about the Intel, who makes money business on this thing? And almost every single one of them says, “Intel’s making 40, 50, 60 percent gross margin on all this stuff and we get put aside for single-digit type of stuff,” so Microsoft and Intel make all the money in the PC market, in the laptop market, whatever, and the Dells don’t. The Lenovo don’t. They live on a very, very thin margin because all they’re really doing is packaging. And a lot of PR activity and their distribution channels, they’re all very important things, but the margin that they make is very small. And I’ve had a lot of mobile guys say, “That’s why we don’t do business with Intel because if we did business with, if we chose Intel’s processor architecture for this thing then they control us, they make all the margin and we’re stuck back here making four percent, seven percent, nine percent margins.”
Good info on CMOS scaling and especially liked the detail in one of the links in that article which is worth printing ...
http://semiengineering.com/will-7nm-and-5nm-really-happen/
The options
In the near term, the leading-edge chip roadmap looks clear. Chips based on today’s finFETs and planar FDSOI technologies will scale to 10nm. Then, the gate starts losing control over the channel at 7nm, prompting the need for a new transistor architecture.
One of the leading contenders for 7nm has been the high-mobility finFET, which is a finFET with III-V materials in the channels. The III-V finFET would supposedly consist of Ge for PFET and indium-gallium-arsenide (InGaAs) for NFET.
“Germanium is making good progress,” said An Steegen, senior vice president of process technology at Imec. “III-V is tricky. It still needs more work.”
In fact, III-V technology is challenging and could get pushed out to 5nm. “Ge and III-V channels are still hot contenders at 7nm,” added Aaron Thean, director of the logic program at Imec. “However, the narrow bandgap of these materials are presenting problems for low-leakage transistors. The outlook for these (III-V) materials is moving from 7nm to likely 5nm. This does not preclude the use of these materials in the source/drain in the nearer term.”
So with the possible delay of the III-V finFET, what’s next for 7nm? Imec is weighing several transistor options, namely gate-all-around, quantum well finFETs, and SOI finFETs. Considered the ultimate CMOS device in terms of electrostatics, gate-all-around is a device in which a gate is placed on all four sides of the channel. “At a certain point in your process (for gate-all-around), you are going to undercut that fin. Then you come in with a dielectric in the gate and you basically fill in underneath the channel, which is now a nanowire,” Imec’s Steegan said.
“There is, of course, SOI,” she said. “You can also have an effective quantum well. (In this structure), you build in an effective energy area to basically shut down the leakage path.”
For the channel materials at 7nm, Imec has narrowed down the options to two choices—an 80% composition of Ge for PFET; or a 25% to 50% mix of Ge for PFET and 0 to 25% of Ge for NFET with strain relaxed buffers. “The perfect candidate, of course, is germanium,” she said. “The silicon devices are operating at 0.8 and 0.75 volts. But the germanium devices are operating at 0.5 volts. So you have exactly what you want in performance as well as the electrostatic behavior. But, of course, you have lower Vdd, so you save power.”
Following 7nm, the industry is looking at several transistor options for 5nm—gate-all-around; quantum well; SOI finFETs; III-V finFETs; and vertical nanowires. “We are looking at all of the aspects for vertical nanowires,” she said. “We are exploring how we grow the channels. And do we use channel last or first integration schemes?”
Sooo if 64-bit Android performs better because of the replacement of Dalvik by ART this means that 64-bit enabled Silvermont will start make A9, A12, A15, A17 etc look anemic leaving the licensees with only the power hungry A57 and relatively wimpy A53 to compete.
http://www.pcworld.com/article/2391020/google-paves-way-for-64bit-android-l-devices-by-year-end.html
Intel has been testing 64-bit chips with Android in its labs and has seen performance benefits in graphics, data encryption and decryption, decompression and applications with large data sets. Mobile devices will be able to have more than 4GB of memory, and “fewer trips to the memory” will be required to process applications, said Doug Fisher, vice president and general manager of Intel’s Software and Services Group.
Intel is working with Google to made kernel-level changes for Android L, and will make an x86 distribution available at a later date.
Improvements in Android L also will help improve mobile device battery life, Insight 64’s Brookwood said. In addition, the notification and application launching mechanisms have been revamped in the OS.
Android L supports ARMv8, which is ARM’s 64-bit microprocessor architecture. Meanwhile, industry consortium Linaro is working on a 64-bit version of the mobile Chrome browser.
Google has said that Android applications will run two times faster on ART, the new Android 64-bit run-time environment for application execution.
Android developers won’t have to make many code changes to move to ART from the Dalvik runtime, Brockwood said.
Ultimately, Google’s move will spur a larger number of applications written with 64-bit and larger memory addressing in mind, Brookwood said.
This is from a study released this month and shows the leading edge of this technology which of course we had no doubt ...
http://link.springer.com/article/10.1007/s10973-014-3817-0?no-access=true#Fig8
Journal of Thermal Analysis and Calorimetry
June 2014, Volume 116, Issue 3, pp 1435-1444
Study on thermal decomposition of polymers by simultaneous measurement of TG–DTA and miniature ion trap mass spectrometry equipped with skimmer-type interface
Abstract
Simultaneous thermogravimetry–differential thermal analysis and miniature ion trap mass spectrometry (TG–DTA–ITMS) instrument equipped with a skimmer-type interface has been successfully developed. The system allows precise real-time monitoring analysis of activated organic compounds such as pyrolysates because gaseous transformation of the evolved gases from the TG–DTA is suppressed by the skimmer-type interface. Also, excessive fragmentation during ionization of molecules can be avoided with the soft ionization method by photoionization. In addition, the miniaturized ITMS is equipped with unique tandem MS capability. These features permit a better understanding of the complicated thermal behavior and the precise pyrolysates of materials. The pyrolysis of various standard reagent polymers such as polymethyl methacrylate, polystyrene, and polyphenylene sulfide has been examined by the TG–DTA–ITMS in inert atmosphere. The synergy effect of the skimmer-type interface and the ITMS was evaluated comparatively with conventional quadrupole mass spectrometry results. It was confirmed that the real-time monitoring ITMS/MS worked satisfactorily. Here, we demonstrate a valuable application of the TG–DTA–ITMS in the detailed analysis of commercial polymers.
I would be extremely surprised if Apple does not use Intel's 10nm to build its post 14FF/16FF products. The foundries are so far behind after this half-node I just can't see uber-rich Apple twiddling its thumbs for years waiting for them to catch up just so it does not have to use Intel as a foundry. Intel is a valued supplier to Apple and will continue to be so in the future as new opportunities arrive.
The point Charlie is missing is that it is worth much to Intel if it means OEMs don't look at the competition for their needs.
I go for not enough innovation but Knights Landing will seriously change that. Socketed Xeon Phis with massive cache/memory/bandwidth and 60 Silvermont cores obviating the need for Xeons as well as providing serious integer compute power, this is the stuff of vector computers. Tripling of TFlops at less power. This is a ridiculous great one gen jump and will fire up the HPC market no end. Gpus will be further squeezed.
competition in the server chip market from the likes of Applied Micro Circuits (AMCC) and Cavium (CAVM), which “could render 15% long-term [data center group] DCG growth targets too aggressive
These guys are delusional over a phantom threat. Power 8 is a more serious threat than these already obsolete ARM designs.