Samsung Tapes Out 1 Gigahertz+ ARM Cortex-A15 SoC

[Guest]

Samsung and Synopsis claim to have been first with a tape-out of an ARM processor based on the Cortex-A15 core.

Samsung Tapes Out 1 Gigahertz+ ARM Cortex-A15 SoC : Read more

 

[saturnus]

I thought there was a Cortex A15 in the Krait SoC? Guess not.

Should be interesting. They promise 50% better performance than A9 and 20% lower power consumption even without the BIG.little heterogenous setup with A7 ultralow power idle aux cores.

Also interesting to see that Tom's so far has completely missed the release of the Intel Medfield which didn't rock any boats as promised but was only able to par 2 year old top models like SGS2 and Iphone 4 (not 4S), and that's with a 33% overclock advantage on both CPU and GPU and much larger reference form factor, some would say clunky.

 

[ohim]

Wonder when phones will start again to have 3-4 days between rechargers .. (not talking about 1 week +) for now most of the smart phones are totally useless if you plan to be away from a power outlet more than 24 hours.

 

[neuromancer2701]

TI announced OMAP5 in Feburary so this is basically the same thing for Samsung. What does "Tapes Out" mean in this context?

 

[scook9]

[citation][nom]saturnus[/nom]I thought there was a Cortex A15 in the Krait SoC? Guess not.Should be interesting. They promise 50% better performance than A9 and 20% lower power consumption even without the BIG.little heterogenous setup with A7 ultralow power idle aux cores.Also interesting to see that Tom's so far has completely missed the release of the Intel Medfield which didn't rock any boats as promised but was only able to par 2 year old top models like SGS2 and Iphone 4 (not 4S), and that's with a 33% overclock advantage on both CPU and GPU and much larger reference form factor, some would say clunky.[/citation]
Krait is not based off the arm core, they only license the instruction set, not the cores. Just like AMD cpus are still x86 but not the same core design as Intel.

And the SGS2 is almost 1 year old....and Intel did that well with a single core - dual core with improved architecture on the way.

 

[saturnus]

[citation][nom]scook9[/nom]Krait is not based off the arm core, they only license the instruction set, not the cores. Just like AMD cpus are still x86 but not the same core design as Intel.And the SGS2 is almost 1 year old....and Intel did that well with a single core - dual core with improved architecture on the way.[/citation]

True. February 2011. Seems like 2 years ago. My bad.

I disagree on the Medfield performance though. It's hyperthreading enabled single core. That has always been Intels attempt to avoid having to go multicore. However, it does not consistently out-perform a dual Cortex 9 core even in single threaded workflows although being clocked 33% higher. CPU stress tests on other sites also show very poor battery performance when the CPU is heavily taxed but that might be an early issue that will be addressed.

I won't hold my breath for Intel ever making a competetive SoC for the smartphone space, not in this or next generation. A viable tablet SoC perhaps for those few requering legacy x86 support for their Windows 8 tablet.

 

[Guest]

What is this???
This was done long time ago .. dual core Exynos 5250 @ 2 GHz
http://www.samsung.com/global/business/semiconductor/minisite/Exynos/products5dual.html

 

[Guest]

@neuromancer

it's a manufacturing term, it should be Tape-out without the s, it basically means the Photo mask has been finalised and ready for mass production

 

[830hobbes]

[citation][nom]ohim[/nom]Wonder when phones will start again to have 3-4 days between rechargers .. (not talking about 1 week +) for now most of the smart phones are totally useless if you plan to be away from a power outlet more than 24 hours.[/citation]
That's not a technological challenge but an engineering design decision. It's easy to make a phone that lasts for 4 days, just add a big enough battery compared to the hardware. The constraint is that the engineers are told to make the smallest phone possible with a 6-hour or so battery life. If companies were convinced people would buy a thicker phone with a longer battery, they'd make it (RAZR MAXX?). It's a simple matter of market demand determining engineering constraints, as it should be. If people don't like it, they should let cell phone manufacturers know they'd buy thicker phones. If companies aren't listening and it really is a profitable market segment, a competitor should emerge that will support it.

 

[shasheng]

Etudions bien et faisons des progres chaque jour.

 

[ojas]

@saturnus

What???!!! What review of the Medfield SoC have you read? Go read the xolo x900 review on AnandTech.

This is a die shrink of the Core 2 architecture! That's a 4 year old architecture whipping A9 dual cores in most CPU benchmarks. It was standing toe to toe with phones running ICS while it was running gingerbread.

Battery life was mid-range but not bad by any means, plus the X900 has a smallish battery anyway.
The graphics chip is a non-intel component. Plus the X900 is based on intel's reference design.
If they want to, all they need to do to is add a faster graphics chip and throw in a larger battery. It's a thick unit anyway.

Add the 22nm Silvermont Atom and they've done it.

I don't know about you ppl, but I'm downright excited about x86 SoCs

 

[aicom]

I don't expect Intel to be the ultimate epic solution to everything SoC related, but I do believe they'll be a viable competitor to ARM, if only due to their manufacturing advantage.

 

[Guest]

Medfield actually was the champ when you compare performance with battery life. It has the best performance per watt of anything out there right now. Intel most certainly is competitive. It won't be the fastest until they bring it down to 22nm and then go out of order in the instruction set, but they are VERY competitive. Intel can also do 64-bit which ARM cannot at the moment which kills ARM on the server side. No one wants a 32-bit server. That is a waste of space.

 

[Cazalan]

The 22nm Atoms will surpass ARM, and the 14nm Atoms will be unbeatable.

ARM has a lot of work to do to maintain their lead if even possible.

 

[noob2222]

[citation][nom]ojas[/nom]@saturnusWhat???!!! What review of the Medfield SoC have you read? Go read the xolo x900 review on AnandTech.This is a die shrink of the Core 2 architecture!

.I don't know about you ppl, but I'm downright excited about x86 SoCs[/citation]

Sure, the 1.6 ghz Atom looks great compared to a 1.2 ghz A9, but throw the 1.5 ghz into the mix ... performance vanishes. The one s is a dual core.

 

[blazorthon]

[citation][nom]saturnus[/nom]I thought there was a Cortex A15 in the Krait SoC? Guess not.Should be interesting. They promise 50% better performance than A9 and 20% lower power consumption even without the BIG.little heterogenous setup with A7 ultralow power idle aux cores.Also interesting to see that Tom's so far has completely missed the release of the Intel Medfield which didn't rock any boats as promised but was only able to par 2 year old top models like SGS2 and Iphone 4 (not 4S), and that's with a 33% overclock advantage on both CPU and GPU and much larger reference form factor, some would say clunky.[/citation]

You people all do realize that Medfield is single core and you're comparing it to dual core A9s in order to say that it isn't great, right? If Intel made a dual core version, it would beat the A9s with ease.

Not only is Medfield only a single core, but it's hardly even more than just a beta version for Medfield; a proof of concept. The 22nm and 14nm Atoms should come out with hue improvements within the next year (22nm) to two years (14nm). Even if the architecture wasn't better than ARM's architectures, it would have the process node advantage.

[citation][nom]noob2222[/nom]Sure, the 1.6 ghz Atom looks great compared to a 1.2 ghz A9, but throw the 1.5 ghz into the mix ... performance vanishes. The one s is a dual core.[/citation]

Furthermore, the current Medfields were never supposed to be the top of the high end, just to get into the high end. That the One S is about twice as fast as the Medfield being tested shows us that the Medfield has near the performance per core as the One S has (only ~10% lower).

 

[tuch92]

All of these phones are capable of much higher battery life.

I got a Galaxy S 4G in November and with moderate use (games occasionally, texting all day, reading xkcd, googling) it got about 8 hours of battery life, maybe a bit more. Since then I have upgraded roms with a low voltage kernel, and after letting it charge completely I got 40 hours of battery life. Ri-diculous.

So if you don't like your battery life head over to xda developers and find a low voltage kernel for your phone. It's a lifesaver.

Note: I live on a campus where Wifi is always available, so wifi is turned on all the time, and 4G about half the time, GPS only when I'm driving.

 

[blazorthon]

[citation][nom]tuch92[/nom]All of these phones are capable of much higher battery life. I got a Galaxy S 4G in November and with moderate use (games occasionally, texting all day, reading xkcd, googling) it got about 8 hours of battery life, maybe a bit more. Since then I have upgraded roms with a low voltage kernel, and after letting it charge completely I got 40 hours of battery life. Ri-diculous. So if you don't like your battery life head over to xda developers and find a low voltage kernel for your phone. It's a lifesaver. Note: I live on a campus where Wifi is always available, so wifi is turned on all the time, and 4G about half the time, GPS only when I'm driving.[/citation]

Does it underclock/under-volt any parts of the phone to help get such high battery time? If so, then it might take a performance hit.

 

[ojas]

[citation][nom]noob2222[/nom]Sure, the 1.6 ghz Atom looks great compared to a 1.2 ghz A9, but throw the 1.5 ghz into the mix ... performance vanishes. The one s is a dual core.[/citation]
Lol. Think about it. That's a 4-year old arch and a single core proc running gingerbread. You're comparing it to the top-of-line ARM stuff running on ICS.

Even still, there's a benchmark that looks like this:

Just saying, there're both sides to everything. Wait for dual core medfield SoCs and the Ivy Bridge based stuff. That's what i'm waiting for, it'll be like classic AMD vs Intel just with ARM instead of AMD

 

[blazorthon]

[citation][nom]ojas[/nom]Lol. Think about it. That's a 4-year old arch and a single core proc running gingerbread. You're comparing it to the top-of-line ARM stuff running on ICS.Even still, there's a benchmark that looks like this:Just saying, there're both sides to everything. Wait for dual core medfield SoCs and the Ivy Bridge based stuff. That's what i'm waiting for, it'll be like classic AMD vs Intel just with ARM instead of AMD[/citation]

At that time, we might have A15 and similar competing against Medfield, so it might be a level playing field. Besides, AMD seems to be catching up.

 

[noob2222]

[citation][nom]ojas[/nom]Lol. Think about it. That's a 4-year old arch and a single core proc running gingerbread. You're comparing it to the top-of-line ARM stuff running on ICS.Even still, there's a benchmark that looks like this:Just saying, there're both sides to everything. Wait for dual core medfield SoCs and the Ivy Bridge based stuff. That's what i'm waiting for, it'll be like classic AMD vs Intel just with ARM instead of AMD[/citation]
One benchmark Intel was able to win, one.... Not sure how thats bragging rights or such a great thing. As far as claiming pure single core, your forgetting it still runs HT, so its single +. As for making a dual core, they need to make sure 22nm can actually use less power as its single core runs middle of the pack on battery life. Dual core would drain the battery in a matter of minutes if its anything like Ivy bridges power saving of 4%. Intel has a long way to go, and ARM isn't just sitting there waiting to be overtaken.

 

[blazorthon]

[citation][nom]noob2222[/nom]One benchmark Intel was able to win, one.... Not sure how thats bragging rights or such a great thing. As far as claiming pure single core, your forgetting it still runs HT, so its single +. As for making a dual core, they need to make sure 22nm can actually use less power as its single core runs middle of the pack on battery life. Dual core would drain the battery in a matter of minutes if its anything like Ivy bridges power saving of 4%. Intel has a long way to go, and ARM isn't just sitting there waiting to be overtaken.[/citation]

Medfield, once again, is a 32nm die shrink and modification of Core 2, just updating the architecture would help far more than the die shrink would. You also seem to forget that the current Medfield phones have weak batteries and using a proper batter with the dual cores would mean similar, if not better, battery time than the x900. Intel doesn't have a long way to go to compete with ARM, but do they have a long way that they can go.

That it took Hyper-Threading to get the single core of the Medfield to more or less match the performance that a single one of those Krait 1.5GHz cores from the HTC One X doesn't matter because that is how it performs. It still beats out the mid-ranged smart phone and tablet processors and most of the high end smart phone and tablet processors.

Intel just needs to use a much more optimized architecture in addition to scaling down to 22nm and later on to 14nm.

This isn't particularly related, but Ivy averaged 3.7% faster while using almost exactly 3% less power (i7-3770K versus i7-2700K), so it's more like almost almost 7% more energy efficient than Sandy Bridge (varies a little, depending on the workload), at least in the high end.

 

[ojas]

[citation][nom]noob2222[/nom]One benchmark Intel was able to win, one.... Not sure how thats bragging rights or such a great thing. As far as claiming pure single core, your forgetting it still runs HT, so its single +. As for making a dual core, they need to make sure 22nm can actually use less power as its single core runs middle of the pack on battery life. Dual core would drain the battery in a matter of minutes if its anything like Ivy bridges power saving of 4%. Intel has a long way to go, and ARM isn't just sitting there waiting to be overtaken.[/citation]
I know, one benchmark isn't bragging rights, was just saying that this is possible too. Even on all the other CPU benchmarks, the Xolo was in the top 5 or 3. THAT however, is bragging-rights worthy.

As far as HT goes, it really didn't make too much of a difference in one of the benchmarks in which it was explicitly tested (read the article, feeling too lazy to link ) so it's pretty much single core, single threaded performance.
Anyway, even in the desktop world, it's well known that actual physical cores perform far better than logical HT cores.

Battery life: Again, they used a weak-ish battery as this phone's based on the reference design. Normalized battery life, which takes into account the battery and all wasn't award winning but it was decent. No "it'll last 5 min on a charge!". Plus it was more than a lot of ARM-based chips that showed poorer performance, so really i can't see a reason for the negativity.
The reviewer at AnandTech clearly states that the battery life fears regarding x86 are a "myth".

Dual core Atoms are for tablets only, at least for now.

The increase in efficiency realized by IB might not be much compared to SB, but compared to the Core architecture of the Core 2 series, IPC alone should lead to a huge jump in battery life, plus lower TDP of the 22nm Atoms will be a bonus too.

As an example of IPC improvements:

Read full article here.

As an example of resulting efficiency improvements:

and combining the net score for both single and multi-threaded benchmarks:

Read the full thing: http://www.tomshardware.com/reviews/sandy-bridge-efficienct-32-nm,2831-8.html