Samsung's Exynos 5 Octa: Checking out the chip inside the Galaxy S 4

This new chip has eight CPU cores, but that's only part of the story.

Read the whole story

 

[Torrijos]

The one true subject that should be treated is : Are the promised benefits in energy consumption present?

Anyway this is all for naught since this isn't the phone the american consumers will get (still no idea when, or at what price?).

 

[veilrap]

Can someone explain why there are different processors delivered to different countries? At first glance it makes little to no sense.

 

[Donkey Hotay]

Can someone explain why there are different processors delivered to different countries? At first glance it makes little to no sense.

In an acronym, LTE.

Radio functions are increasingly often included into System on Chips now, to further streamline power requirements. The U.S.' use and spectrum for LTE is unique and therefore requires separate hardware; in this case, Snapdragon's 600 SoC was chosen along with Qualcomm's LTE radio.

This is roughly what Samsung did with the Galaxy S III as well.

The last three paragraphs of this article should preface every article in the U.S. about the Galaxy S IV: instead we will be inundated by useless hardware comparisons and hipster angst regarding the power conservation versus the performance of the Exynos big.LITTLE implementation.

 

[LosD]

Can someone explain why there are different processors delivered to different countries? At first glance it makes little to no sense.

In an acronym, LTE.

Weren't the International version supposed to support LTE from the get go this time around?

In that case, it really makes no sense (unless the International version doesn't support the american frequencies, of course).

 

[kray28]

Can someone explain why there are different processors delivered to different countries? At first glance it makes little to no sense.

In an acronym, LTE.

Radio functions are increasingly often included into System on Chips now, to further streamline power requirements. The U.S.' use and spectrum for LTE is unique and therefore requires separate hardware; in this case, Snapdragon's 600 SoC was chosen along with Qualcomm's LTE radio.

This is roughly what Samsung did with the Galaxy S III as well.

The last three paragraphs of this article should preface every article in the U.S. about the Galaxy S IV: instead we will be inundated by useless hardware comparisons and hipster angst regarding the power conservation versus the performance of the Exynos big.LITTLE implementation.

So what happens if you take the American version of the S3/S4 abroad? Are you limited to 3G or slower?

 

[MaskofSanity]

Andrew Cunningham wrote:
Big.LITTLE pairs two distinct CPU cores, one larger and faster (in this case, a Cortex-A15 running at 1.2GHz) and one that is smaller and more power-efficient (a Cortex-A7 running at 1.6GHz).

The speeds are backwards.

 

[Errum]

Ignoring for a moment the question of what smartphone software or usage pattern really benefits from 4 cores of any kind, can the big.LITTLE architecture possibly be an efficient way to handle the range of tasks? After all, my car doesn't have a V8 under the hood and and the added expense and complication of a lawnmower engine in the trunk. Given that normal cores can be throttled or shutdown as required, I'm skeptical.

Oh yeah, and enquiring typographical minds want to know why it isn't "BIG.little"?

 

[THT]

Andrew Cunningham wrote:
Big.LITTLE pairs two distinct CPU cores, one larger and faster (in this case, a Cortex-A15 running at 1.2GHz) and one that is smaller and more power-efficient (a Cortex-A7 running at 1.6GHz).

The speeds are backwards.

Look at the benchmarks in the article. In particular, the Exynos Octa versus the 1.9 GHz Tegra 4. They both have Cortex-A15 cores, somewhat similar memory architectures. The vast majority of the difference between the two will be driven by clock rate.

 

[BullBearMS]

Can someone explain why there are different processors delivered to different countries? At first glance it makes little to no sense.

In an acronym, LTE.

Radio functions are increasingly often included into System on Chips now, to further streamline power requirements. The U.S.' use and spectrum for LTE is unique and therefore requires separate hardware; in this case, Snapdragon's 600 SoC was chosen along with Qualcomm's LTE radio.

Qualcomm produces separate LTE chips which could easily be combined with Samsung's Exynos 5 Octa.

Apple uses them in the new iPhone alongside their custom SOC.

That's the Qualcomm LTE modem chip highlighted in blue.

 

[TheGame]

The Exynos 4 Quad that was in the International version of the S3 also came to the US in the Galaxy Note 2, so it stands to reason that the Note 3 will come with the Octa in tow worldwide...which leads me to the question: Why is the Octa not worldwide for the S4? The Exynos 4 Quad in the Note 2 seems to demonstrate that it can be done. Just seems like an odd step back by Samsung given that they've worked out this problem before.

 

[jalexoid]

Andrew Cunningham wrote:
Big.LITTLE pairs two distinct CPU cores, one larger and faster (in this case, a Cortex-A15 running at 1.2GHz) and one that is smaller and more power-efficient (a Cortex-A7 running at 1.6GHz).

The speeds are backwards.

Nope. Everyone else is reporting the same. The A15s run at a lower frequency than A7s

 

[jalexoid]

Can someone explain why there are different processors delivered to different countries? At first glance it makes little to no sense.

In an acronym, LTE.

Radio functions are increasingly often included into System on Chips now, to further streamline power requirements. The U.S.' use and spectrum for LTE is unique and therefore requires separate hardware; in this case, Snapdragon's 600 SoC was chosen along with Qualcomm's LTE radio.

Qualcomm produces separate LTE chips which could easily be combined with Samsung's Exynos 5 Octa.

Apple uses them in the new iPhone alongside their custom SOC.

That's the Qualcomm LTE modem chip highlighted in blue.

Except that iPhone5 still has different SKUs that have different LTE chips.

 

[BullBearMS]

Andrew Cunningham wrote:
Big.LITTLE pairs two distinct CPU cores, one larger and faster (in this case, a Cortex-A15 running at 1.2GHz) and one that is smaller and more power-efficient (a Cortex-A7 running at 1.6GHz).

The speeds are backwards.

Nope. Everyone else is reporting the same. The A15s run at a lower frequency than A7s

The A15 was really designed more as a tablet CPU since tablets have huge batteries and can dissipate the heat A15's put out over a larger surface area, but they can be shoehorned into phones if you keep the clock frequency low.

 

[BullBearMS]

Except that iPhone5 still has different SKUs that have different LTE chips.

Having different LTE modems in different countries is worse than just giving the US the crap CPU?

 

[lowlymarine]

Except that iPhone5 still has different SKUs that have different LTE chips.

Having different LTE modems in different countries is worse than just giving the US the crap CPU?

A 1.9 GHz quad-core Krait is 'crap'? Don't be ridiculous.

 

[Wintersdark]

Except that iPhone5 still has different SKUs that have different LTE chips.

Having different LTE modems in different countries is worse than just giving the US the crap CPU?

I assume they get a good rate on their Qualcomm LTE chipsets by bundling them with Qualcomm SOC's as well.

If they get the Snapdragon's at a much reduced cost compared to the Exynos, it's not at all surprising.

Bastards. I'm still grouchy that my Note has a crappy NA processor as well. The international version was much better.

 

[dagamer34]

Can someone explain why there are different processors delivered to different countries? At first glance it makes little to no sense.

In an acronym, LTE.

Radio functions are increasingly often included into System on Chips now, to further streamline power requirements. The U.S.' use and spectrum for LTE is unique and therefore requires separate hardware; in this case, Snapdragon's 600 SoC was chosen along with Qualcomm's LTE radio.

Qualcomm produces separate LTE chips which could easily be combined with Samsung's Exynos 5 Octa.

Apple uses them in the new iPhone alongside their custom SOC.

That's the Qualcomm LTE modem chip highlighted in blue.

Except that iPhone5 still has different SKUs that have different LTE chips.

It has different frequency bands, but the same LTE chip (MDM9615).

 

[Putrid Polecat]

Ignoring for a moment the question of what smartphone software or usage pattern really benefits from 4 cores of any kind, can the big.LITTLE architecture possibly be an efficient way to handle the range of tasks? After all, my car doesn't have a V8 under the hood and and the added expense and complication of a lawnmower engine in the trunk. Given that normal cores can be throttled or shutdown as required, I'm skeptical.

Oh yeah, and enquiring typographical minds want to know why it isn't "BIG.little"?

The issue isn't whether the car has a lawnmower engine for light driving, the issue is that this chip has 4 lawnmower engines for light driving.

Also, I would like a future article to address precisely how the cores are utilized. For example, can two A15s be active with two A7s? Can all cores be shut down except for one A7? How does the chip now to switch from an A7 to an A15, and based on what formula does it choose a mix of cores. There are a lot of permutations available.

 

[dagamer34]

Andrew Cunningham wrote:
Big.LITTLE pairs two distinct CPU cores, one larger and faster (in this case, a Cortex-A15 running at 1.2GHz) and one that is smaller and more power-efficient (a Cortex-A7 running at 1.6GHz).

The speeds are backwards.

Nope. Everyone else is reporting the same. The A15s run at a lower frequency than A7s

When first announced, I distinctly remember the A15s running at 1.8Ghz and the A7s running at 1.2Ghz. My guess is that they ran into power issues as the smaller Exynos 5250 can draw way more power than a smartphone can provide at peak (almost 7W by AnandTechs testing) and that was with only 2 A15 cores at 1.7Ghz.

 

[armin.hammer]

Ignoring for a moment the question of what smartphone software or usage pattern really benefits from 4 cores of any kind, can the big.LITTLE architecture possibly be an efficient way to handle the range of tasks? After all, my car doesn't have a V8 under the hood and and the added expense and complication of a lawnmower engine in the trunk. Given that normal cores can be throttled or shutdown as required, I'm skeptical.

Oh yeah, and enquiring typographical minds want to know why it isn't "BIG.little"?

You may not have a separate lawnmower engine in your vehicle, but many people find that having an additional electric motor + batteries makes things more fuel efficient. Hybrids probably make a better analog for multiple batteries rather than multiple core clusters...but I hate car analogies in the first place

Having the additional A7 cluster doesn't add much expense, and can supposedly yield up to a 70% increase in battery life (Obviously, CPU isn't the only drain on the handsets battery though and we can't expect an overall increase that big.) DVFS can still be used and scaled for either cluster.

This solution just adds additional flexibility.

 

[idealego]

If you're going to call the Exynos 5 Octa an 8-core then you should call the Tegra 4 (and 3) a 5-core. However, I think it's more accurate to call the Exynos 5 Octa and the Tegra 4 both quad cores since neither can run more than 4 cores at a time.

 

[Riemann Zeta]

That music in the Samsung video doesn't belong anywhere outside of a 1980s elevator.

 

[armin.hammer]

If you're going to call the Exynos 5 Octa an 8-core then you should call the Tegra 4 (and 3) a 5-core. However, I think it's more accurate to call the Exynos 5 Octa and the Tegra 4 both quad cores since neither can run more than 4 cores at a time.

big.LITTLE MP mode allows all eight to be used at once. That will not be the mode implemented on the S4, but the Octa itself isn't inherently limited.
ARM whitepaper on the architecture.

 

[name99]

Ignoring for a moment the question of what smartphone software or usage pattern really benefits from 4 cores of any kind, can the big.LITTLE architecture possibly be an efficient way to handle the range of tasks? After all, my car doesn't have a V8 under the hood and and the added expense and complication of a lawnmower engine in the trunk. Given that normal cores can be throttled or shutdown as required, I'm skeptical.

Oh yeah, and enquiring typographical minds want to know why it isn't "BIG.little"?

The issue isn't whether the car has a lawnmower engine for light driving, the issue is that this chip has 4 lawnmower engines for light driving.

Also, I would like a future article to address precisely how the cores are utilized. For example, can two A15s be active with two A7s? Can all cores be shut down except for one A7? How does the chip now to switch from an A7 to an A15, and based on what formula does it choose a mix of cores. There are a lot of permutations available.

EXACTLY. big.LITTLE seems like a bizarre way to solve the problem, insanely expensive in silicon.
To put it differently, the problem it seems to be solving is an assumption that OSs are frozen in how they will treat the chips and cannot be changed. IF this were true, then the only way to make a lower power system would be to have each OS visible core consist of the big and the LITTLE part, with OS-transparent switching between them.
But this seems like a totally flawed assumption. I imagine (given Tegra) that Android can perfectly well handle switching between 2 or 4 active cores and just 1. iOS can obviously be rewritten to support this if/when Apple wants this functionality. And if Windows and Blackberry are unwilling to update their OSs, honestly who cares?

All in all it seems like a really strange direction for ARM, one which I don't understand. It's really hard to imagine a realistic scenario where having 4 active low-power cores is a better match to the compute/energy parameters of a problem than a single low-power core.
Is it purely to get around patents?

 

[Skiddywinks]

As far as I am aware, both CPUs have individually gated cores, so the performance range is from one A7 up to four A15s.

 

[WaltC]

About cell phones in general, what is the point of talking up the number cores the respective mobile cpu has? It isn't remotely akin to discussing the number of cores in an x86 cpu, for instance, and always leads inevitably to the same old ill conceived x86 vs. ARM daydreams--that state of the art ARM is on a performance par with soa x86. It's not even close.

And in terms of what a cell phone is supposed to look like, just how many variations can you imagine for a smartphone? I mean, the face of the thing has got to be all touchscreen, doesn't it? Form follows function perfectly in this case. And as long as Samsung keeps putting the word "Samsung" on the face of its cell phones, it certainly isn't possible to assume that anyone will confuse Samsung and Apple products (Never mind the myriad of differences that manifest on screen after turning the phones on.) Smartphone cell phone exterior design cannot by its nature be radically different among cell phones, just as automobile tires and steering wheels (and many other things) cannot be radically different among competing automobile manufacturers--form has often got to follow function by necessity.

 

[BullBearMS]

Ignoring for a moment the question of what smartphone software or usage pattern really benefits from 4 cores of any kind, can the big.LITTLE architecture possibly be an efficient way to handle the range of tasks? After all, my car doesn't have a V8 under the hood and and the added expense and complication of a lawnmower engine in the trunk. Given that normal cores can be throttled or shutdown as required, I'm skeptical.

Oh yeah, and enquiring typographical minds want to know why it isn't "BIG.little"?

The issue isn't whether the car has a lawnmower engine for light driving, the issue is that this chip has 4 lawnmower engines for light driving.

Also, I would like a future article to address precisely how the cores are utilized. For example, can two A15s be active with two A7s? Can all cores be shut down except for one A7? How does the chip now to switch from an A7 to an A15, and based on what formula does it choose a mix of cores. There are a lot of permutations available.

EXACTLY. big.LITTLE seems like a bizarre way to solve the problem, insanely expensive in silicon.
To put it differently, the problem it seems to be solving is an assumption that OSs are frozen in how they will treat the chips and cannot be changed. IF this were true, then the only way to make a lower power system would be to have each OS visible core consist of the big and the LITTLE part, with OS-transparent switching between them.
But this seems like a totally flawed assumption. I imagine (given Tegra) that Android can perfectly well handle switching between 2 or 4 active cores and just 1. iOS can obviously be rewritten to support this if/when Apple wants this functionality. And if Windows and Blackberry are unwilling to update their OSs, honestly who cares?

All in all it seems like a really strange direction for ARM, one which I don't understand. It's really hard to imagine a realistic scenario where having 4 active low-power cores is a better match to the compute/energy parameters of a problem than a single low-power core.
Is it purely to get around patents?

Depending on the OS, the ability to use many cores simultaneously itself is in doubt.

Blackberry 10 and iOS both run native code and are both based on an OS that has always supported spreading threads for native code across multiple CPU cores. (Symmetric Multiprocessing) Blackberry 10 is based on QNX and iOS is based on Unix.

Things get trickier when you move to Android and Windows Phone. Both of those OSes run apps on top of virtual machines and those virtual machines were designed to be multithreaded, but only on a single CPU core.

Both OSes give coders an escape hatch for games where you can ignore the virtual machine and write directly to the hardware, but that's not what the vast majority of apps for those platforms do.

Google added some support for SMP in Android 4, but as Intel discovered while porting Android to x86, it's got a long way to go:

"If you are in a non-power constrained case, I think multiple cores make a lot of sense because you can run the cores full out, you can actually heavily load them and/or if the operating system has a good thread scheduler.

A lot of stuff we are dealing with, thread scheduling and thread affinity, isn't there yet and on top of that, largely when the operating system goes to do a single task, a lot of other stuff stops. So as we move to multiple cores, we're actually putting a lot of investment into software to fix the scheduler and fix the threading so if we do multi-core products it actually takes advantage of it."

http://www.theinquirer.net/inquirer/new ... processors

You may or may not recall that dual core Android devices were marketed and sold to the public long before Android had any support for the second core. Now we're seeing 8 cores marketed on an OS whose support for multiple cores is still dodgy.

You'll see benchmarks that look good, but those are written to ignore the virtual machine like the games do. Then you see performance numbers for regular apps written to run on the virtual machine and article authors scratching their heads about why that app didn't see the expected performance increase.

 

[marcusj0015]

When you need even MORE power, does it use all 8 cores?

Honestly, it's an interesting idea, but I still feel like quad "mini" cores, is overkill.

 

[marcusj0015]

Any news on ARM64 btw? That could be a game changer.

 

[teh31337one]

Did ars do anything to confirm the devices were running the Exynos chip? From what I've heard, the devices on the show floor were had Qualcomm's snapdragon in them...

 

[tipoo]

I'm glad this story was run, everyone calling it an octacore was irritating, since only 4 cores will ever work at once and 4 are low power energy saving cores.

The loss of the GPU is probably the bigger deal since the CPU isn't faster than what we can get here, the Adreno 320 is good but it still doesn't catch up to PowerVR, and this one would be clocked even higher than the iPhone has it.

 

[Deleted member 192806]

Since batteries seem to be the universal limiter, what's the current state of battery technology for cellphones?

 

[tipoo]

Samsungs video spells performances wrong, "performences", lol

 

[tipoo]

I wonder how much the 4 low power cores really help out, over the 1 low power core in the Tegra 3 and 4. I guess they mean you can do more low power tasks at once, but I wonder if 4 are really necessary or if it's mostly for the marketing that comes with "Octa". For many tasks you'd likely want to get it done as fast as possible so it can get back to idle and shut down those A15 cores, I wonder how many use cases will make good use of the four low power ones.

 

[ChronoReverse]

I think people are going to be surprised by how well a Krait 600 running at 1.9GHz will perform in comparison.

What I'm interested in is how well the A7 cores save power. These cores should be more efficient than even the low power core in Tegra 3. Tegra 3's low power core is just an A9 built on a different process and clocked down (one of the reason why the S4Pro versions of the same phone tend to have better battery life and equal performance). The A7 is designed specifically to be even more efficient than A9.

I'm also thinking that even Anand has made a mistake about the clock frequencies. I'm pretty sure it'll be 1.6GHz A15 and 1.2GHz A7 but we'll see.

 

[KevinN206]

....
Depending on the OS, the ability to use many cores simultaneously itself is in doubt.

Blackberry 10 and iOS both run native code and are both based on an OS that has always supported spreading threads for native code across multiple CPU cores. (Symmetric Multiprocessing) Blackberry 10 is based on QNX and iOS is based on Unix.

Things get trickier when you move to Android and Windows Phone. Both of those OSes run apps on top of virtual machines and those virtual machines were designed to be multithreaded, but only on a single CPU core.

Both OSes give coders an escape hatch for games where you can ignore the virtual machine and write directly to the hardware, but that's not what the vast majority of apps for those platforms do.

Google added some support for SMP in Android 4, but as Intel discovered while porting Android to x86, it's got a long way to go:

....
http://www.theinquirer.net/inquirer/new ... processors

You may or may not recall that dual core Android devices were marketed and sold to the public long before Android had any support for the second core. Now we're seeing 8 cores marketed on an OS whose support for multiple cores is still dodgy.

You'll see benchmarks that look good, but those are written to ignore the virtual machine like the games do. Then you see performance numbers for regular apps written to run on the virtual machine and article authors scratching their heads about why that app didn't see the expected performance increase.

Windows Phone 8 shares the same NT kernel as Windows 8 and shares a subset of the new WinRT API. The NT Kernel can use up to 64 cores, if needed. Please excuse my ignorance, but what virtual machine is Windows Phone 8 using? My understanding is that everything is ultimately compiled to native code regardless if you use C++, C#, or JavaScript to write your apps.

Here's an interesting excerpt from Peter Bright's article: http://meincmagazine.com/features/2012/10 ... w-again/5/

Microsoft had two competing technologies that were suitable as the basis of a modern API—COM and .NET—but each had drawbacks. .NET has rich metadata, safe programming languages, and fits neatly with many conventions of modern programming languages (such as their use of interfaces and object-oriented inheritance). On the other hand, .NET uses a complex runtime with a virtual machine, rather than native code, which potentially exacts a performance penalty, and is somewhat awkward to use and integrate with existing native C and C++ programs.

COM is weaker in many regards—less descriptive metadata, no built-in notion of inheritance, unsafe programming languages, and in most ways far more awkward to use than .NET. But COM does have an important advantage: it has no virtual machine, being native code from the ground up. COM is also the technology used by many of the big, old Windows programs, including the all-important Office.

Within Microsoft, there are also certain political considerations at play. Internal opinion about .NET is divided. Many teams use the technology to good effect and regard it as important. The Windows division ("WinDiv"), however, has a different view. The many developmental difficulties that occurred during the (essentially abandoned) development of Windows Longhorn were attributed, at least in part, to the use of .NET code. The team also believes that native C++ development is what most developers want. This has tended to lead to an avoidance of the use of .NET even when it's an appropriate or desirable technology.

When the Windows team created WinRT, indications are that these non-technical concerns weighed at least as heavily as any technical reasons. As a result of this distaste for .NET, combined with COM's native nature and extensive use in major pre-existing Windows applications, the decision was made: WinRT is built on COM.

But it's COM with a twist.

 

[BullBearMS]

Please excuse my ignorance, but what virtual machine is Windows Phone 8 using?

You're quoting an article that answers your question. .NET uses a virtual machine, as does Android.

However, you're off when you claim that Windows Phone supports WinRT just like Windows 8 does.

http://www.infoworld.com/t/microsoft-wi ... e-8-196479

 

[Hinton]

Anyway this is all for naught since this isn't the phone the american consumers will get (still no idea when, or at what price?).

It's not completely for naught, some other countries have people that aren't completely worthless. They're the exception of course, but saying that it's /all/ for naught is an exaggeration.

 

[Putrid Polecat]

Anyway this is all for naught since this isn't the phone the american consumers will get (still no idea when, or at what price?).

It's not completely for naught, some other countries have people that aren't completely worthless. They're the exception of course, but saying that it's /all/ for naught is an exaggeration.

It's interesting even if no phone uses the Exynos.