The Intel Atom Thread

[Nothingness]

Homeles states that their are more improvements to 64-bit than just more address space. But as far as I know, that isn't true. 64-bit address space is just that: the ability to use more RAM (+ about 10% performance boost).


Sure, but what has that to do with "64-bit"?

Well if you double the number of registers, that's more than only enlarging the address space, no? Here is a summary: http://en.wikipedia.org/wiki/X86-64

 

[cbn]

According to the die picture Merrifield is ~72mm² big.

So it should be pretty inexpensive then.

With the following info (and rumor) in mind:

http://www.dailytech.com/Microsoft+Confirms+Plans+to+Repair+Windows+81/article34389.htm

Microsoft also announced some upcoming Windows OEM changes, including lower licensing prices and relaxed hardware spec requirements. Likely realizing it might lose low-end users with the termination of Windows XP sales in April, Microsoft has announced it will now only require 1 GB of DRAM and 16 GB of hard drive or NAND storage to install Windows 8.1 on 64-bit chips.

http://www.dailytech.com/Report+Mic...s+by+70+for+Budget+PC+Makers/article34381.htm

According to Bloomberg, Microsoft will slash Windows 8.1 licensing fees by 70 percent for makers of low-cost (priced less than $250) computers and tablets. This means that licenses will now be priced at $15 instead of $50.

We could end up seeing some pretty cheap x86 Windows devices using the Merrifield chip.

 

[TuxDave]

Homeles states that their are more improvements to 64-bit than just more address space. But as far as I know, that isn't true. 64-bit address space is just that: the ability to use more RAM (+ about 10% performance boost).

Sure, but what has that to do with "64-bit"?

Here's something to freak you out.

64-bits doesn't even really refer to the address space. You only get 48-bits of physical address. What you should say is that "64-bits" in this context physically refers to the width of the general purpose registers (which can be used by integer instructions and not just address space related instructions). So saying 64-bits is just address space is wrong any way you look at it.

However, that narrow focus on the physical presence of "64-bits registers" isn't the same as "64-bit extension of the x86 ISA" because that's x86-64 which comes with a gigantic package of stuff including an increase in the NUMBER of registers.

I'm not really a big fan of nerd fights over semantics. You know what everyone else is talking about.

 

[Sweepr]

Super Micro backs Intel Atoms - New range of embedded servers

Intel will be over the moon that another key hardware maker has adopted its Atom chips at the heart of its embedded servers.

[...] A1 series motherboards
Supermicro's new A1 series motherboards feature 4-8 cores, 14-20W SoC, 64GB ECC memory, and quad GbE LAN ports. The A1SAM/SAi motherboards are optimised for low-power server/storage, web-hosting applications, while the A1SRM/SRi motherboards with integrated bulk Crypto (10Gb/s) Intel QuickAssist technology are targeted for network security appliances.

More here: www.techradar.com/news/computing/servers/super-micro-backs-intel-atoms-1229432

XS35V4 Shuttle: An affordable Bay Trail fanless nettop

http://translate.google.com.br/tran...bay-trail-fanless-abordable-en-approche-15665

 

[witeken]

Will you also update this thread with Merrifield products? That would be very nice.

 

[Sweepr]

Will you also update this thread with Merrifield products? That would be very nice.

Yes. I'll be posting news/articles related to different products based on the Silvermont core (including Merrifield/Moorefield).

Intel to get its SoC to Motorola phones

We got one thing confirmed by Intel at the Mobile World Conference in Barcelona. Since Intel has announced that its portfolio of clients includes Lenovo, Asus, Foxconn and a few others, we asked if the Lenovo deals means getting Intel into future Motorola phones.

Lenovo acquired Motorola from Google and Intel has implied that the Lenovo deal will cover the future development. They didn’t say Motorola will launch phones with Intel SoCs inside, but they said that we can assume that future Motorola phones that come under Lenovo will be influenced by this multi-year and multi-device deal.

www.fudzilla.com/home/item/34066-intel-to-get-its-soc-to-motorola-phones

 

[witeken]

Awesome, especially since there's a war going on at Seeking Alpha about Intel's mobile competence.

 

[bullzz]

@witeken - thats looks really bad for intel. a CPU company not getting it right on CPU? looks like even a contra revenue might not save merri/moore field

 

[witeken]

What do you mean? Silvermont is a great CPU.

 

[bullzz]

for tablets, yes. for phones going by the geekbench scores, it is not looking good

 

[mikk]

For phones as well, the problem is that Intel decided to put only two Silvermont cores into Merrifield SoC which is a design flaw for the segment Merrifield is aimed for and now they are in a hurry to bring Moorefield.

 

[witeken]

I think Silvermont was actually designed mainly with smartphones in mind. Silvermont has great single threaded performance at a very low power consumption. Two Silvermont cores should easily be on par with thermally bound quadcore Krait 300+ cores. The clock speed bump in the new S801/805 will only make it throttle even faster. Also, as far as I know, Merrifield is meant for midrange phones, like a SGS5 Mini or so.

 

[Khato]

Alpha

Must say that I'm rather disappointed with that article in particular. Citing the over-all geekbench score as proof of performance is a great way to 'prove' a point to the uneducated masses, just like Intel's use of WebXPRT to 'prove' the opposite point. Is A7 faster? Sure, especially when running the encryption tests with accelerated instructions on the 64 bit A7 code path but not on Silvermont, or when running floating point code. The floating point advantage is genuine, no denying that... it just isn't terribly applicable to the target market. Whereas the encryption portion, if it actually matters, doesn't make much sense - it almost looks like Geekbench isn't making use of AES-NI on Silvermont. Either that or Intel severely reduced performance of that block on Silvermont.

Oh yeah, there's also the matter of markedly different power consumption, but that's clearly an unimportant detail for the target market. (Which is actually somewhat true because these devices don't spend much time at full load power draw.)

 

[witeken]

Yes, I agree. Some of his articles such as A7 vs Silvermont were actually very good, but as some people have pointed out in the comments the other link, it seems that he is writing more articles to make more money, while delivering less quality.

Personally, I think some people (including him) have too high expectations. Silvermont is without doubt a great product, but you can't expect to make it Intel's market share rise from a 5% (or 0.2% in smartphone market) to 50% within a few months. It's a process that will take a few years. I think that by 2017, Intel will be leading the tablet and smartphone market the same way as it does with desktops now and also making nice profits (in 2017 Intel will have a full year of 10nm vs 20nm FinFET).

 

[Roland00Address]

I think intel would be happy to have a 15% marketshare in phones in the next 12 months. I don't forsee this happening. Now in the next 24 months that is definitely possible, for they will finally have a fully integrated chip.

 

[witeken]

The deal with Lenovo, Asus and Dell is nice, but I wonder if companies with bigger market share are also planning to make phones with Intel inside in 2014, like Samsung, LG, Huawei, HTC or Sony. For example, the HTC 816 would have been a great phone with Merrifield, but now it has 4 slow A7s.

 

[jfpoole]

Whereas the encryption portion, if it actually matters, doesn't make much sense - it almost looks like Geekbench isn't making use of AES-NI on Silvermont. Either that or Intel severely reduced performance of that block on Silvermont.

Geekbench takes advantage of AES-NI on Silvermont.

 

[Khato]

Geekbench takes advantage of AES-NI on Silvermont.

Thanks, which I guess implies that Silvermont is hitting some other constraint when it comes to the AES test in particular. After all, the other encryption tests look to be roughly where they should be if you compare the sub-test results to one of Intel's big core results. It's only AES where instead of the expected ratio you get something about 3x worse.

 

[jpiniero]

I think intel would be happy to have a 15% marketshare in phones in the next 12 months. I don't forsee this happening. Now in the next 24 months that is definitely possible, for they will finally have a fully integrated chip.

15% is going to be incredibly tough. It'd have to come entirely from Android, and probably more likely Samsung adopting it.

 

[Homeles]

15% might have been doable if Merrifield had launched in a decent time frame, but it's way, way past that point.

 

[SiliconWars]

Smartphones hit 1 billion units in 2013, of that Intel barely sold 2 million. They are rumoured to be pulling out of phones next year.

 

[Homeles]

Smartphones hit 1 billion units in 2013, of that Intel barely sold 2 million. They are rumoured to be pulling out of phones next year.

Rumored by some people that are incredibly high on drugs, perhaps.

 

[Arachnotronic]

Geekbench takes advantage of AES-NI on Silvermont.

Which actually means that comparing SLM to any 32-bit ARM chip isn't quite fair ;-)

 

[Arachnotronic]

Yes, I agree. Some of his articles such as A7 vs Silvermont were actually very good, but as some people have pointed out in the comments the other link, it seems that he is writing more articles to make more money, while delivering less quality.

I see. When it's a positive Intel article, it's quite good, and when it says, "hey, wait a minute...why does Apple's A7 show a markedly different result in Geekbench than in Intel-sponsored WebXPRT?" it's "poor quality"?

Did it ever occur to you that the author went with 64-bit for A7 because in 32-bit mode, the AES-NI instructions aren't available for the A7 but are available for Silvermont?

There are 64-bit results for BYT, and it doesn't make much of a difference. Compiling GB3 with ICC rather than GCC/MSVC++, however, makes a real difference.

 

[Khato]

I see. When it's a positive Intel article, it's quite good, and when it says, "hey, wait a minute...why does Apple's A7 show a markedly different result in Geekbench than in Intel-sponsored WebXPRT?" it's "poor quality"?

Did it ever occur to you that the author went with 64-bit for A7 because in 32-bit mode, the AES-NI instructions aren't available for the A7 but are available for Silvermont?

There are 64-bit results for BYT, and it doesn't make much of a difference. Compiling GB3 with ICC rather than GCC/MSVC++, however, makes a real difference.

No, there's simply a difference between an article that performs in-depth analysis to give readers all the facts - http://seekingalpha.com/article/1848061-intel-vindicated-very-competitive-with-apples-a7 - and one that ignores any details which don't support the desired message - http://seekingalpha.com/article/2057243-apple-outguns-intel - see the difference?

I can understand the desire to paint Intel in a bad light considering that they aren't moving anywhere near so fast as investors would like into the phone and tablet markets... But that doesn't change the fact that the Silvermont CPU architecture is competitive with Cyclone on integer workloads while providing all the encryption and floating point performance that's needed in the intended markets. As well, Silvermont provides that performance at a fraction of the power consumption.

Anyway, here's to hoping that we don't have to wait too long for Anandtech to receive a merrifield device for testing... though sadly I wouldn't be surprised if it took months instead of weeks.