The Intel Atom Thread

[witeken]

The first Silvermont phone is rumored to be the 5" Lenovo Vibe X2-TO, with a Merrifield SoC at 2.2GHz. Unfortunately, this won't be the main version of the phone; the regular X2 has a MediaTek SoC.

 

[North01]

The first Silvermont phone is rumored to be the 5" Lenovo Vibe X2-TO, with a Merrifield SoC at 2.2GHz. Unfortunately, this won't be the main version of the phone; the regular X2 has a MediaTek SoC.

The phone portion of the ASUS Transformer Book V, that was announced at Computex 2014, will use Intel's Moorefield SoC.

The ASUS FonePad 7 (FE375CG) with Moorefield is launching in China later today.

 

[Arachnotronic]

The first Silvermont phone is rumored to be the 5" Lenovo Vibe X2-TO, with a Merrifield SoC at 2.2GHz. Unfortunately, this won't be the main version of the phone; the regular X2 has a MediaTek SoC.

Why Merrifield and not Moorefield? 2 SLM cores is no match for 4x A17s...

 

[witeken]

Why Merrifield and not Moorefield? 2 SLM cores is no match for 4x A17s...

Silvermont probably has a higher IPC frequency, and lower power consumption. 4 cores is mostly important in benchmarks, so let's wait for reviews.

 

[Arachnotronic]

Silvermont probably has a higher IPC frequency, and lower power consumption. 4 cores is mostly important in benchmarks, so let's wait for reviews.

4x A17 @ 2GHz should theoretically be faster than 2x Silvermont @ 2.13GHz. But yes, you're right, let's wait on the reviews.

 

[frozentundra123456]

The phone portion of the ASUS Transformer Book V, that was announced at Computex 2014, will use Intel's Moorefield SoC.

The ASUS FonePad 7 (FE375CG) with Moorefield is launching in China later today.

So is that phone a totally separate device, or does it detach from the tablet/laptop somehow?

I like the Transformer T-100, but depending on cost, this five in one just seems to be trying a bit too hard to be all things to all people.

 

[witeken]

4x A17 @ 2GHz should theoretically be faster than 2x Silvermont @ 2.13GHz. But yes, you're right, let's wait on the reviews.

From memory, Silvermont has about the same IPC as A15 (Source: AT SLM), and A17 should be slower because it's a re branded A12. I haven't looked at benchmarks in quite some time, so maybe I am wrong.

But besides performance, there's also price, graphics, etc. Clover Trail+ wasn't the fastest SoC, but it did manage to get quite a few design wins. And if a company does't like Merrifield, Intel also has Moorefield.

 

[witeken]

So is that phone a totally separate device, or does it detach from the tablet/laptop somehow?

You place the phone in the tablet, you place the tablet in the laptop.

http://www.cnet.com/products/asus-transformer-book-v/

Moorefield SoC for phone and tablet, and there's probably Core M in the laptop (and 2 more GB of RAM).

 

[Nothingness]

From memory, Silvermont has about the same IPC as A15 (Source: AT SLM), and A17 should be slower because it's a re branded A12. I haven't looked at benchmarks in quite some time, so maybe I am wrong.

Cortex-A17 should have an IPC close to Cortex-A15, so I wouldn't be surprised if it came very close or even beat Silvermont. OTOH the GPU in the Merrifield version of the phone is much better and we all know how much the GPU matters.

BTW I'm not sure any public Cortex-A12 benchmark is available.

 

[Shivansps]

Today i spotted another Chinese BT tablet
http://www.pandawill.com/voyo-winpa...40d-80-ips-screen-2gb-32gb-silver-p90401.html

2GB, 1.46gb usable?!

I still have my eyes on the $99 Toshiba Exite go with 3735g and android, if it only could boot windows somehow... any news on how the BT boot system works? for example, Allwinners SoC are locked to look for primary boot on external sd, that makes them to be able to boot linux no matter what the OEM do.

 

[witeken]

Cortex-A17 should have an IPC close to Cortex-A15, so I wouldn't be surprised if it came very close or even beat Silvermont. OTOH the GPU in the Merrifield version of the phone is much better and we all know how much the GPU matters.

BTW I'm not sure any public Cortex-A12 benchmark is available.

I have to read the A12 coverage of AT, but I'd be surprised if the A12 comes quite close. Although Silvermont is dual-issue, it has some CISC tricks to make it wider than it appears to be.

 

[Nothingness]

I have to read the A12 coverage of AT, but I'd be surprised if the A12 comes quite close. Although Silvermont is dual-issue, it has some CISC tricks to make it wider than it appears to be.

Issue width is not the whole story 🙂 Anand wrote this:

The Cortex A17 name was deliberately chosen as ARM expects to be able to deliver similar performance to the Cortex A15 (in mobile apps/benchmarks, likely not in absolute performance), but in a much smaller area and at a lower power. I can't help but wonder if this is what the Cortex A15 should have been from the very beginning, at least for mobile applications.

ABI Research also claims that A17 has 0.9 performance of A15 while consuming about half as much.

I hope we'll soon see some real benchmarks rather than such claims, but if this is correct then CPU-wise the Merrifield phone and the A17 phone should be close. But again it looks like the Merrifield GPU is beefier (and IMHO 2 cores is a better fit for phones).

 

[witeken]

Interesting, but it doesn't seem like A12/17 will be used much, since Qualcomm will go straight to A57. A17 probably wasn't a big upgrade compared to Krait anyway. Since Qualcomm has 100% market share in flagship phones, A12 probably won't harm Intel too much.

 

[Arachnotronic]

Issue width is not the whole story 🙂 Anand wrote this:
ABI Research also claims that A17 has 0.9 performance of A15 while consuming about half as much.

I hope we'll soon see some real benchmarks rather than such claims, but if this is correct then CPU-wise the Merrifield phone and the A17 phone should be close. But again it looks like the Merrifield GPU is beefier (and IMHO 2 cores is a better fit for phones).

It's pretty impressive what ARM was able to do with the A17.

 

[Nothingness]

It's pretty impressive what ARM was able to do with the A17.

Indeed. Too bad it comes one year too late...

 

[NTMBK]

The A17 sounds like a great core for normal smartphones (i.e. not overpriced phablet flagships). A big.LITTLE with 2 A17s and 2 A7s sounds about right.

 

[Arachnotronic]

Indeed. Too bad it comes one year too late...

Yeah, but it gives ARM a good base to build off of in the future.

Quite honestly, I wonder how long companies like Qualcomm will continue doing custom cores when ARM's continue to get better and better?

 

[witeken]

It's pretty impressive what ARM was able to do with the A17.

Why?

And do you have benchmarks with performance, power consumption and efficiency?

And the fact that it's so late to market makes it a lot less impressive. If AMD or Apple delivers Sandy Bridge performance when Skylake it out, no one will find that impressive either.

Quite honestly, I wonder how long companies like Qualcomm will continue doing custom cores when ARM's continue to get better and better?

Probably as long as ARM can't deliver compelling, efficient cores (A15, anyone) in time. I doubt A57 will be such a core, but its TTM is better than Qualcomm's own core, so that helps.

 

[Arachnotronic]

Why?

And do you have benchmarks with performance, power consumption and efficiency?

And the fact that it's so late to market makes it a lot less impressive. If AMD or Apple delivers Sandy Bridge performance when Skylake it out, no one will find that impressive either.


Probably as long as ARM can't deliver compelling, efficient cores (A15, anyone) in time. I doubt A57 will be such a core, but its TTM is better than Qualcomm's own core, so that helps.

What impresses me about ARM more than anything else is how quickly they can spin new cores/revisions with meaningful changes/improvements.

 

[North01]

You place the phone in the tablet, you place the tablet in the laptop.

http://www.cnet.com/products/asus-transformer-book-v/

Moorefield SoC for phone and tablet, and there's probably Core M in the laptop (and 2 more GB of RAM).

The keyboard is a dumb dock, the tablet portion is using Core M to run Windows. If you ignore the phone, this is just a standard Windows 2-in-1 using Core M.

The Moorefield SoC inside the phone is only running Android when docked into the tablet portion.

 

[krumme]

The A17 sounds like a great core for normal smartphones (i.e. not overpriced phablet flagships). A big.LITTLE with 2 A17s and 2 A7s sounds about right.

Yes. But cores is good for marketing and a a7 is 1.5mm2 so i doubt we will see less than 4 of those. We have power gating on all cores today so the there is only a insignificant cost side to it for some good bragging rights.

 

[VengenceIsMine]

Interesting, but it doesn't seem like A12/17 will be used much, since Qualcomm will go straight to A57. A17 probably wasn't a big upgrade compared to Krait anyway. Since Qualcomm has 100% market share in flagship phones, A12 probably won't harm Intel too much.

A12/17 will make strong headway in the midrange and below tablet & mini android pc segment, there are a boatload of products in the pipeline based on Rockchip 3288 that will replace the Allwiner A31, tegra 3 & rockchips 3188 that make up a ton of the $99-$149 market. A lot of bang for the buck in the white box tablet market.

 

[Nothingness]

Here is a Geekbench comparison of Cortex-A12 @1.8GHz vs Z3745 @1.86GHz. I guess it's Cortex-A12 because the part number is 3085 while it should be 3087 for Cortex-A17.

It looks like my previous claim of C-A1[27] IPC being close to Silvermont IPC was correct.

 

[Arachnotronic]

Here is a Geekbench comparison of Cortex-A12 @1.8GHz vs Z3745 @1.86GHz. I guess it's Cortex-A12 because the part number is 3085 while it should be 3087 for Cortex-A17.

It looks like my previous claim of C-A1[27] IPC being close to Silvermont IPC was correct.

Good find. Though I take Geekbench with a grain of salt, this does give a good general picture.

ARM did a solid job with A17.

 

[Khato]

What impresses me about ARM more than anything else is how quickly they can spin new cores/revisions with meaningful changes/improvements.

Though just how quickly is this actually? If you consider their 'performance' offerings of A9, A15, A57 they've been on a 2 year cadence in terms of actual shipping SoCs (2010, 2012, then probably this year or early next year for A57.) The 'value' segment seems to be following a similar path with A5, A7, A53 (I'm not certain if A17 should be put there or in a new 'mid range' segment.) Regardless, this is actually a slower cadence of core releases than what you see on the x86 side, which is part of the reason why they can make more drastic changes... and they need to make such changes because of how much ground they can make up by doing so.