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.
Geekbench2 comparison between A7 and Bay Trail
http://browser.primatelabs.com/geekbench2/compare/2307544?baseline=2395855
and between A6 where it is only showing a ~50% integer increase over previous generation and less in floating point.
http://browser.primatelabs.com/geekbench2/compare/2307544?baseline=2208675
A7 comes up short in real integer benchmarks, even Merrifield should beat it. The crypto stuff has corrupted GB3.
Apple are past masters at finding benchmarks to prove their point even if the point they are making is not that great. The most power-efficient way of getting the maximum threaded performance out of the wide Cyclone is say 4-way SMT like the Alpha EV8 was going to have. They will probably make the caches bigger too. Add 250-350 MHz clockspeed and bingo there's your A8 .
A8 on 20nm should be a beast.
I'm just expecting a higher clocked A7 with maybe multi-threading. I doubt they will have the thermal budget to increase ipc further.
Don't assume perfect scaling with clockspeed. Are there any Geekbench2 scores for A7 so we can remove these cryptology scores from the integer total although the x86 FP total on that version is unusable. The only results I don't mistrust on Geekbench are Geekbench2 Integer and Geekbench3 Floating Point. You picked a dubious benchmark to make your point originally. The SPEC org really should release SPECCPU2000 as freeware for the general good !
Read the small print
http://browser.primatelabs.com/geekbench3/406668
Geekbench 3.1.4 Tryout for Windows x86 ***(32-bit)***
You also already know A7 has a big jump in Geekbench3 from 32 to 64-bit because you actually already put out a SA article showing it !
There is also an Avoton 32 to 64-bit comparison which shows a 15% increase in integer and 25% in floating point and a 17% increase in the total score.
http://browser.primatelabs.com/geekbench3/442333
Geekbench 3.1.4 Tryout for Linux x86 (32-bit)
http://browser.primatelabs.com/geekbench3/218430
Geekbench 3.1.2 for Linux x86 (64-bit)
http://browser.primatelabs.com/geekbench3/compare/442333?baseline=218430
Intel also put this slide out
Do you still insist that it 'doesn't make a damn bit of difference.' ?!
If Intel has an 80% utilization of their fabs today, they'll probably have less than 40% at 10nm
Doubtful as it will be monitored to fit long before then. They also will just add more cache to the dies as well as add more cores so the die sizes won't come down a lot except for the very price conscious ultra-mobility market.
You really think Tegra 4 is a superior SoC for a high end mobile phone?
Probably the best of a bad lot. Still going to throttle a lot, it is an A15 after all, but peak performance is probably better than S800's in both cpu and gpu.
Where were you at $12
Predicting a potential breakout at $16.5 if it ever went past it. It's all on Yahoo.
Oh, so stock ARM IP that any Tom, Dick, and Harry can license built on 28nm process is competitive, eh?
Samsung have got the A15 clockspeed up to 2.1 GHz and the A7 clockspeed up to 1.5 GHz on the latest model. The design may be stock but they seem to have the best implementation by far now.
Obviously, Intel didn't "nail it" with Merrifield/Moorefield and the sooner you accept that, the sooner we can move on.
From a cpu/gpu POV they look pretty good to me for the power consumption class they both will reside in. A 2.1 GHz dual-core Merrifield can embarrass quad-core chips and Alpha clones .
Pretending that the world doesn't want Intel chips because Intel is big (QCOM is a larger co. now, BTW - gg, Intel management) is simply absurd.
Not the world but the world's mobile OEMs. Ask yourself why even now with the highly competent Moorefield and highly capable 7260 *only* PC OEMs are going to take the combined design up and no it's not the camera or any other fancy feature. It's no accident and is something Intel bashers have been mentioning for some time and I finally agree with them having seen the empirical evidence. It's not that Intel is big but it is known to take an inordinate lot of the margins of ecosystems it dominates.
Show me a mobile SoC in the market today that is superior for high end devices to the Snapdragon 800.
Tegra 4 and Moorefield, K1 too when they tip up this year. The latest Exynos Octa does not look bad from a cpu POV and of course Bay Trail-M. Krait is an inconsistent and hot performer which has now been overclocked beyond the point of maximum efficiency and that's a trend that will continue with their even hotter 2.9 GHz 64-bit 20nm version.
In 2014, Intel admitted that Bay Trail is poorly designed for its intended market.
Not its intended market but its extended market.
I'd keep the "SnaplessDragon" comments to a minimum until Intel is actually winning designs worth a damn and RECOGNIZING REVENUE from each design.
I'm pretty sure it's recognizing revenue in Windows devices and you know yourself how snapless Krait is in real usage.
Those designs were won on getting LTE before everybody else not on the SnaplessDragon that resides in the SoC.
And Intel's "we're better than Snapdragon 800" nonsense really irritates me ... So, Intel basically needs to stop trying to "convince" us with MobileXPRT
I thought the results were funny and informative and undoubtedly show how over-engineered for their form factors the top phone chips currently are. It was amazing how bad S800 did on all fronts and A7 could not show its geeky superiority either.
And Intel still didn't win much of that market if at all if my sources are good. I warned you before MWC that Intel probably didn't have more than a couple of design wins to show (and nothing big), but I was off-base as the number came out to a nice round goose egg.
Only you was expecting a lot. The designs will come just as the CLT+ and Medfield designs keep coming. I am pretty sure an anti-Intel bias also has a lot to do with it, the mobile OEMs have seen what happened to the PC OEMs and don't want the same result for them. Intel will always be swimming hard against the incumbent tide which you will finally accept when Broxton comes out and is still ignored by the majority.
better camera
How many phones currently ship with more than 13MP capability ?
Merrifield gets 27,579 vs Tegra 4i's 26,131 with half the cores which suggests it's going to beat it across the board and especially user experience. Just learn to embrace reality, Merrifield's rivals will now be quad-A7/A9/A12/A17 etc and that's the biggest smartphone market segment out there now.
There are two main points I wish to convey about that article. First, don't cut corners/use different testing parameters i.e. 32/64-bit just to emphasize the main point you are making because now you have gullible readers thinking the A7 is twice as fast as Merrifield in general mobile cpu usage and I bet you just won't find another test to remotely back that up.
Secondly be more aware of how relevant popular benchmarks actually are to the processor model you are testing. To my mind the micro-benchmarks in Geekbench are best suited to test all the background housekeeping tasks a server does on all its data so backups, dumps, security encryption, mass data analysis etc and so would be a better fit for say judging Avoton against X-Gene or Hierofalcon in how fit they are for doing batch server work. Horses for courses.
A7 in 32 bit mode is still faster in GB3
You should have used that as a starting base though. Comparing 32-bit to 64-bit is sloppy especially when you did not bother to tell anyone. If you insist on using Geekbench3 as a 'pure cpu' measurement then strip out the memory, stream, encryption benchmarks and work out the new integer scores and use the floating point as is. Don't just rely on the combined aggregate total and declare that to be 'cpu speed' because that is just not an accurate statement.
Ideally what you should be endeavoring to do is using your contacts beg, steal or borrow a spec2000 suite so you can do a respected cross-architecture benchmark comparison and not one that just happens to be free.
That said, I paid the $20 and have been spending time looking through the MobileXPRT source
If you can amend the source to change the loop iterations performed that would give you an insight into how big a part throttling plays. You already have a Krait/Medfield/A7 to do some comparisons.
But for now all I am hearing is that geekbench sucks and MobileXPRT is the bomb diggity.
and all we hear is bias the other way making so many assumptions like this
'Unfortunately, this chart seems to be pretty off-kilter with reality now that some more processor-oriented benchmarks are in (rather than web-based benchmarks that are highly dependent on Javascript/HTML5 engines - in short, not "pure" CPU measurements). '
So geekbench is the arbiter of reality now, a suite that thinks encryption is integer and denorms are frequent in floating point ? I still contend web browsing speed is of more relevance to mobile users than the micro-benchmarks in Geekbench and game-playing cpu speed can be measured by the Ice Storm Physics score and Silvermont does just fine there.
One day there will be a geekbench released that does not have problems associated with it and answers more questions than it raises but not yet, mixing encryption and integer in the same grouping is the latest flaw. Ashraf compared a 64-bit A7 GB3 score with a Merrifield 32-bit score and did not bother mentioning it in his article even though A7 has a big jump from its 32-bit to 64-bit score and 64-bit Avoton also improves over Bay Trail. Also rather than finding out why Krait/A15/A7 run relatively poorly (length of run most likely) in WebXPRT he dismissed it as biased automatically just because the results did not tally with benchmarks that run in a fraction of the time. He also thinks a benchmark that basically consists of compressing/decompressing/encryption is more relevant to a mobile user than something that tests Web speed. That article was truly something else and not in a good way.
WebXPRT runs for about 20 minutes so I suppose it would slug the mobile power hogs.
I actually would have expected Merrifield to clock much higher in turbo mode. If really density was the reason then it was a bad one.
There is also power savings to be had in not clocking so high. To me Merrifield looks just the right balance of performance and endurance and 2.3 GHz quad-core Moorefield will not quite have its battery life which looks about 20% better than the ARM competition. Asus, Dell and especially Lenovo with its ex-MM team need to put nice slim light chassis around it that usually ARM quad-A7s go into and let it loose on the buying public. Discerning buyers, at least in the West, will know the quality they are buying.
Because gate pitch and metal pitch are not the same thing. Basically metal pitch determines how big your die is going to be eventually due to all the wires your transistors must have and gate pitch determines how many transistors there will be on it. Intel's gate pitch lead at 22nm vs foundry 28nm is greater than its metal pitch lead (posted previously by me) and users have quibbled about the poor gate pitch scaling at foundry 20nm already.
Probably why it is so bad at WebXPRT if it is forced to do so many iterations. Just goes to show both CLT+ and Merrifield are fit for purpose while Krait is a benchmarketing processor. Perish the thought you would actually want to play an intensive game on it for more than a minute . Actually we already know the answer to that as it drops the cpu to around 1 GHz, like A15, when the gpu wants to do any real work.
A7 has a 4MB block of SRAM as an L3$ that Merrifield lacks
That would account for some of the cpu performance advantage A7 sometimes displays.
I'm just in shock that Intel tried to tell us that Z3480 was faster than the Snapdragon 800, when it's pretty easy to verify that in most, non-Intel written/non-JS benchmarks that this isn't the case.
Maybe because it actually is on real applications and benchmarks that have application code in them. The real question you should be asking if you wanted to do some real analysis is why is Krait such a big fail in WebXPRT and other ARM processors like Apple A7 are not. Yet you just took the view that antutu/geekbench are all that counts because everybody uses them and benchmarks that don't agree with them are automatically wrong. Seriously is that what you are saying ?
I bet the Krait in your RazrM is also faster according to geekbench/antutu then the Medfield in your RazrI yet you did not subjectively feel this in real usage either. Instead of bemoaning how Intel does not perform well in not very relevant benchmarks perhaps you should be pointing how irrelevant those benchmarks are and what a collective lazy ass all the review sites are for putting up with them. Also why is Intel at fault for producing a processor that's good at javascript ?
something along the lines of: ...
You live on another planet to the rest of us. That 'apology' is negative PR highlighting perceived weaknesses and promoting competitors. Any PR dept worth its salt would just laugh at it. You should try working for someone and seeing how office/company politics really work and how important saving face is to your work/company standing and the extreme attempts people go to ensure it. Also why are you blaming Intel for showing its products in their best light ?!!
Here's how companies work, they try and execute but if they fail they either give out no excuse and hope you did not notice or a very limited one designed for sufficient damage limitation if there is a public outcry. You don't give out a shopping list of perceived failures so your competitors can have a future ball at your expense. Intel already released a general apology about mobility last investor day, that's good enough.
There also is no public outcry over Intel's mobile plans not executing as no analysts expected it to be a success in the first place. The public outcry if there is one is the failure to stem the general PC decline or at least increase profits in that decline. You are projecting your personal disappointment about mobility and trying to make it out to be a major crisis when it is not. The bigger crisis in reality is where are all the Haswell-Ys and Haswell Iris Pros ? They mean more to increasing Intel's bottom line now than Merrifield, Moorefield or any mobile Atom chip.
take a look at the Geekbench and AnTuTu results
Benchmarks I have NEVER taken seriously unlike you. There is a lot to be said for someone actually producing relevant mobile benchmarks rather than using 20+ year old micro-benchmarks.
Why doesn't Intel just license ARM stock cores and clock them high on its process?
because they probably would still burn too much power.
So what do OEMs selecting components base their decisions on?
I doubt it's any benchmark showing that does it just paper specs and features.
Seriously stop acting like a noob dazzled by these silly benchmarks. You know Merrifield is better than that in real usage.
and the part that's original (Silvermont/CPU) is not good.
Over reacting as per usual on two mickey mouse synthetic benchmarks. I have seen no real technical proof of what you was telling me, in fact quite the opposite given Intel's benchmarks doing real application work for some length of time, and what you just wrote up above just leaves me shaking my head at the language you use even if you don't really mean it literally.
The only problem I had with my Razr M was the slow-as-molasses MSM8960.
The mighty OoO 3-issue Krait I believe .
SoFIA LTE (the one on 28nm) will feature a modem that "drafts off of the 7260"
Even though it is advertised as 3G ? Makes sense though as it is already 28nm.
I'd imagine that the 14nm SoFIA will feature a cost optimized 7360 modem.
Did Intel also confirm that as a dual-core Silvermont processor ?
In terms of graphics, I'm very pleasantly surprised regarding 6400/6430@533mhz.
For comparison Anand says the A7's 6430 is around 450 MHz
http://www.anandtech.com/show/7460/apple-ipad-air-review/4
Dell also is an important respected American brand. I suspect they will have little trouble selling Intel x86 LTE phones in North America if they were competitively priced and well made.
Look on the bright side, think of all the margin-sapping contra-revenue it is saving .
Intel intends SoFIA for these ultra-cheap devices.
A discrete Merrifield and LTE chip will still be in a valid sku pricepoint between the original 28nm 3G SoFIA and Moorefield. However the 14nm SoFIA looks a true shrink of 22nm Merrifield (or will it even be a dual-core Broxton ?) with all the performance/size advantages that ensures. Do you know which of the 7160 or 7260 they will pair that with in that later SoFIA 14nm chip ? Common sense suggests the more versatile/powerful 7260 but do you know/can you find out ?
I define "junk" as "this thing wasn't competitive at launch in terms of performance, integration, or time to market".
There is nothing wrong in real-world performance with any phone that has a 2GHz Medfield/Clovertrail+ in it, all the user reviews I have read have invariably ranged from good to excellent. CLT+ was also fine against the original 1.5 GHz quad-core Kraits, it was only when they took the clock to 1.9-2.3 GHz that it fell behind a bit but it is still not totally outclassed thanks to H-T.
You are rewriting history to unfairly blame these fine chips just to avoid facing the inescapable undesirable conclusion that some OEMs will never use Intel chips regardless of how good they are because they prefer the light touch ARM ecosystem.
Intel chips will have to be consistently better and the customers will have to be educated as to the chip's advantages for meaningful and sustainable mobile marketshare to be attained. This is a mammoth task and more mammoth a task than the ARMy trying to get into PCs and servers because Intel is only one company facing a combined market cap of companies assembled against it in the trillions of dollars who don't particularly care for the Wintel business model to be repeated in mobility. Intel will basically have to enable new mobile OEMs to displace the incumbents, again a mammoth task.
You may have to brace yourself now to the notion that even if Broxton is as good as Intel claim it may still get ignored by the vast majority of mobile OEMs. This is going to be trench warfare and it will take many years of close quarter fighting to defeat the ARMy. The unglamorous sockets that Medfield/CLT+ are now in maybe as good as it gets for Intel for quite a while until their OEMs win the customer hearts with ultimately superior products.
Your stock price models based purely on technical specs are too simplistic for the multifaceted design-orientated and IP licensing-favoring mobility markets.
Merrifield is cheap enough to make to not need subsidizing on auxiliary chips which cost about $2 all told.
Why is every phone at the Verizon store, sans the iPhone, powered by a Snapdragon SoC?
That's more a modem issue than an application processor issue. You yourself tested Medfield against a dual-core Krait in the Razr and considered it superior yet now it's junk. Inconsistency from one day/week/month/year to another depending on your mood.
p.s. Bluetooth/Wi-Fi/NFC chips are not expensive.
Keep in mind that the G6400 in Merrifield is an area-optimized version of the G6430 found in the A7.
which is also performing better in Intel's 22nm than the performance variant is in Samsung's 28nm
Merrifield apparently couldn't win against lowly Snapdragon 400 parts for a number of key low end designs launching today, nor could it win hero devices.
Maybe it was not ready or those OEMs will choose to ignore Intel chips regardless of their capabilities for self-control reasons.
Without comms/connectivity integration, it will get eaten alive by the MediaTek based A12/A17 based SoCs.
Even if Merrifield was paired with a 7260 ? This is where having a non-integrated chip helps, any OEM can mix and match it with any discrete modem.
Merrifield is not a piece of junk like Medfield and CLT+
There you go again with your unprofessional inaccurate smack talk. In sustained performance/power/price they were and still are good phone chips.
Well if they are only selling ones and twos at the moment and they have a big sq ft area to do more clearly that's not going to be profitable. However add another launch or two or a big 1stDetect contract and all these 1stDetect production costs get swallowed up as noise. That is what happened in previous profitable quarters.
Using the known dimensions of Bay Trail as discovered by you (9.723 * 10.477mm)
http://seekingalpha.com/article/1707482-intel-answering-the-margin-question
I estimated a Silvermont pair block to be 3.58 * 2.21mm = 7.9 sq mm.
From this I estimated the Merrifield die size as 7.9mm height * 9.3mm width = 73 sq mm.
I did however adjust the displayed width up by ~20% to account for the slanted perspective using the Silvermont block width as my guide so there is a margin of error there.
Regardless it is around 70% the die size of the 28nm A7 which has a very similar but lower clocked gpu. Now this not a big deal for Apple who will move serenely onto the expensive 20nm but for those fabless companies staying on the most cost-effective foundry 28nm process it indicates that Intel is going to beat them in mobility on both gpu performance and die size at 22nm by some margin using the same licensed PowerVR IP.
This is the clearest example of what Intel's process advantage means in the real world and when fabless companies are trying to match Merrifield's eventual lowest selling price for similar gpu performance before its retirement they will find it hard to make any money.
Merrifield is a very efficient design. It may not be glamorous but it will cause problems going forward for the ARMy when it descends the food chain like Medfield has which has a similar die size.
You also have to take into account 1stDetect costs in that gross margin figure as well as any capital expenditure required. The fact remains that when the number of launches go up the gross margin shoots up consistently.
Probably the ex-CFO dumping his 300K+ shares after losing his case.
Samsung's S5 flagship smartphone offers 6GB less storage than it claims
http://www.dailymail.co.uk/sciencetech/article-2567574/The-Galaxy-S5-isnt-perfect-MailOnline-discovers-Samsungs-flagship-smartphone-offers-6GB-storage-claims.html
I wonder if it an attempt to limit the potential contra-revenue damage by making BYT exclusively 64-bit Android, at least the ones that are not dual-OS. Maybe Intel realized after their Investor Day that the GM hit did not go down too well with investors and adjusted their approach and will then perhaps quickly replace Bay Trail with Moorefield in Android tablet sockets going forward. Just speculating .
so I guess that means all the "Contra Revenue" is being applied to CT+ and BT?
Just Bay Trail really as I understand it because CT+ was designed as an Android part and Bay Trail is being pressed in early to do the job Merrifield/Moorefield should have been doing in Android. The failure of the Merrifield design target objectives IMO is not as a phone chip but that it was also expected to be a main flagship tablet android chip when it was released, that's my belief.
As to additional losses/profits I don't think any of us have enough granularity insight to speculate accurately as to what pricing Intel applies to each OEM tablet/phone deal. Feel free to speculate on the negative side if that pleases you though.