The handheld electronics market is spoiled for choice when it comes to microprocessors. Qualcomm, Nvidia, Samsung, and quite a few others, are all vying for that precious space on your phone’s circuit board, which determines how fast and energy efficient your handheld device will be. But with so many manufacturers, each offering several different chips, it can be a pain to even begin to understand the subtle differences, and sadly even most shop assistants probably won’t be able to help you out either.
I’m going to keep this guide as jargon free as possible, but to make sure we’re all on the same page here’s a quick overview of what a system on a chip (SoC) is.
A SoC integrates all of the important components needed by an electronic system into a single chip. Rather than having separate parts for a CPU (central processor) and GPU (graphics processor) on a circuit board like you might see in a typical PC, a SoC combines these circuits into a single chip to make mass manufacturing easier and more cost efficient. So the SoC in your smartphone houses a CPU, GPU, RAM memory, audio converters, and often a radio modem in a single chip.
Right, let’s take a look through some of the companies involved in the mobile business, and their latest SoC offerings, starting with the older generation.
We’ll start with one of the big dogs, Samsung’s Exynos 4 line of processors. The Exynos 4 system on a chip has been used in the hugely popular Galaxy S2, S3, and Note range of smartphones, and can probably be considered the current benchmark which all others are compared against.
The Galaxy S2, S3, and Note 2 all use various versions of the Exynos 4 chip.
The most recent version of this chip is called the Exynos 4 Quad (4412) and is the model used in Samsung’s newest top of the line smartphones like the S3 and Note 2. It features four ARM Cortex A9 CPUs, which currently sit as the mid-performance processor offered by ARM, the Cortex A7 being the lowest power chip and the A15 being the most powerful. With four cores clocked at somewhere between 1.2 and 1.5GHz, this chip will easily cope with multitasking requirements, video content, and gaming.
Graphics wise, the Exynos 4 packs in a four-core ARM Mali 400, so the same graphics chip is used in the Galaxy S2 and S3. However, the newer version is clocked at 440 MHz, compared with 266 MHz before. There’s a decent performance improvement there, which will translate into higher frame rates in games.
The newest versions of the Exynos 4 (the Dual 4212 and Quad 4412) are faster and more power efficient, thanks to a smaller SoC size, than the first iteration.
The OMAP 4 line of SoCs stick an older dual-core configuration in this generation rather than opting for more cores. The SoC uses the same mid power Cortex A9 processors seen in Samsung’s Exynos 4, but provides less peak performance than the Samsung chip when it comes to multitasking or running multi-core optimized applications.
Graphics wise, the chips uses a PowerVR SGX540 clocked somewhere between 307 and 384MHz. This isn’t a really fast chip compared with some of today’s high end devices, but it’s a powerful enough mid range piece of hardware. In the larger tablets the GPU is changed for a SGX544 which can be clocked up to 384MHz, thus providing a bit of extra power needed for higher resolution displays.
This SoC has been used in a wide range of top quality smartphone and tablet products, including the Samsung Galaxy Nexus, Galaxy Tab 2 7.0, and Kindle Fire HD, but doesn’t really cut it compared with the Exynos 4.
Nvidia had a different take on mobile processors at the start of this generation compared with its competitors. It was the first company to really consider minimum power consumption, and was keen to push its GPGPU line of graphics processors. The Tegra 3 chip has appeared in quite a few tablets, HTC’s One X+ smartphone, and will be powering the OUYA Android gaming console.
An example of additional graphical effects available for Tegra 3 devices.
This SoC utilities four Cortex A9 processors, just like the Exynos 4, clocked up to 1.6GHz for peak performance. However it also has another lower power “companion core” which is limited to only 500MHz. This vastly improves on minimum battery drain when performing basic tasks compared with Exynos or OMAP processors, but still keeps the maximum achievable performance high.
Nvidia is known for its GPU prowess when it comes to the PC market, but the Tegra 3 GPU doesn’t offer any real performance benefits over high-end chips from other manufactures. However, Nvidia has teamed up with some developers to build Tegra specific versions of their games with additional graphical features, so if you really enjoy gaming, a Tegra 3 device isn’t a bad choice.
Just like the Exynos 4, the Tegra 3 is starting to show its age. The 40nm SoC size means that it will run warmer than some competitors, and isn’t as energy efficient as it could be, despite the inclusion of a companion core. The Tegra 3 is a step ahead of the competition when it comes to balancing peak performance and energy efficiently, something which the next-generation of competing processors are keen to catch up on.
Qualcomm hasn’t made this easy, as it has four different models in the S4 line-up, and there are several revisions of each of these model types. The S4 range of processors are definitely ahead of the curve, as they were only released in 2012, so consider them more of a generation 1.5.
Snapdragon S4 processors power a lot of recent high-end handsets.
Rather than wade through each SoC we’ll just look at the two most popular versions, the dual-core S4 Plus and the quad-core S4 Pro. The S4 Plus SoC was used in the Motorola Droid RAZR, Nokia Lumia 920 and HTC Evo 4G, while the latest S4 Pro has worked its way into many new high-end product, such as the LG Optimus G, Nexus 4 and Xperia Z.
Both of these versions use Qualcomm’s newest Krait CPUs, which Qualcomm has based loosely around ARM’s Cortex architecture. The Krait processor is clocked up to a maximum 1.7GHz and offers performance somewhere around that of the Cortex A15, so that makes the S4 Plus and Pro some of the best performing processors currently available. Speaking of performance improvements, Krait processors work asymmetrically which allows the speed of each core to be individually controlled, thereby saving on power consumption in low power states.
In addition to the difference in the number of cores, the S4 Pro features an improved Adreno 320 GPU, compared with the Adreno 225 in the S4 Plus, which is one of the fastest GPUs available in many of the benchmarks I’ve seen. The 320 offers more performance than Nvidia’s Tegra 3 and doubles the performance of the older Adreno 225 chip.
Devices using the Snapdragon S4 Pro SoC are some of the fastest and most efficient currently available and should remain competitive even when Nvidia and Samsung finally get around to releasing its next generation Cortex A15 chips. Just check out the video below (minute 3:41) if you want to see how the S4 Pro stacks up against the best Exynos 4 chip.
Intel was late to the party in putting its mobile processors out on the market, only making an appearance last year. Instead of modifying ARM processors to fit its needs, Intel has adapted its line of Atom CPUs used in netbooks to fit the low power consumption required by smartphones and tablets.
The result was actually quite an impressive chip, with the Saltwell CPU outperforming ARM’s popular A9 CPU. Rather than opting for a quad core chip like many of its competitors, Intel has chosen to go with dual-core design with Hyper-threading enabled. It’s an interesting choice; by keeping the numbers of cores low it ensures that battery drain is reduced, whilst keeping support for multi-tasking when needed.
Graphics wise the SoC packs in the familiar PowerVR SGX series, but the chips launched last year contain a rather dated SGX545, so it lags behind a lot of the competition and isn’t a good a choice for gaming compared with the Tegra 3. Still, this SoC has proven popular in tablets, powering the Asus VivoTab, Acer Iconia and the HP ElitePad 900.
The most recent revision of this architecture, Clover Trail+, has been used in the new Levono K900 Android phone we saw earlier this year. Whilst it’s unclear at the moment exactly what performance improvements the + brings to the SoC, we can probably expect at least some battery and performance improvements.
There are a few other chips worth a mention which have been used in popular devices, but fit into categories types covered above.
ST-Ericsson’s NovaThor U8500 utilises a 1GHz dual-core Cortex A9 CPU and Mali 400 GPU, which is a mid-range chip used in devices like the Galaxy Ace 2, S3 Mini, Xperia U and Xperia P. This chip is very similar to the OMAP 4 range, but goes for the same GPU as the Exynos 4 processors instead.
MediaTek’s MT65xx series offers something a little more unique, using more energy efficient Cortex A7 processors in its quad-core chip. But again chooses a dual-core A9 configuration and a PowerVR 5 SGX series GPU in its other chips, just like Texas Instruments and ST-Ericsson.
A Cortex A9 dual-core setup has been the go to configuration for most handsets this generation, with the most recent high end smartphones and tablets opting for quad-core variants of the same processors. Qualcomm is the only real exception to this rule, branching out with its own CPU design. But are other manufacturers willing to follow suit later this year?
If you’re holding out for a newer device with some beefed up hardware then we can have a quick look at what will become available later in 2013 as well.
The Exynos 5 Octa is a misleading marketing ploy. Samsung’s newest SoC will feature two sets of quad-core processors, four ARM Cortex A15 CPUs and four low-power Cortex A7 processors, so it’s not a “true” eight-core. This new design aims to improve on Nvidia’s efficient Tegra design, by using up to four low power chips for the most common tasks, whilst allowing for higher peak performance than the older Exynos 4 processors.
The Exynos 5 range will also appear in dual, quad, and octo-core configurations, depending on how much peak power your device needs. But only the top end chips will make use of big.LITTLE to save on battery life.
The medium range quad-core chips will see Samsung stick with ARM for the GPU as well, using the new Mali-T604 processor for your 3D needs. It can offer a significant performance boost over the older generation and will be more energy efficient as it allows for scaling between one and four cores.
Eight-core versions of the Exynos 5 will opt for a clocked up PowerVR SGX544 tri-core GPU, which boast performance similar to the latest iPad.
The big.LITTLE architecture of the Exynos 5 Octa is certainly the most interesting concept in the next-generation of SoCs. Departing from standard the CPU configuration in favour of balancing power consumption will yield dividends for consumers, but more cores will come with a hefty price tag. Looking even further into the future, the ARM A53/A57 big.LITTLE combination could really blow the competition out of the water.
Nvidia is keeping the same CPU design as the Tegra 3 in its next generation, four main cores and one low-power core. Just like the Eyxnos 5, the new SoC will also be using ARM Cortex A15 processors, which should result in a 40% performance improvement over the previous generation, bringing Nvidia up to par with the current Snapdragon chips.
On top of CPU improvements, Nvidia is boosting the new Tegras graphics power six fold, bringing the GPU core count up to 72. More importantly, Tegra 4 will confirm with new graphics APIs, like OpenGL ES 3.0 and DirectX® 11, which will allow devices utilising this chip to make use of improve graphical features when gaming.
Project Shield will be powered by Tegra 4.
Overall Tegra 4 brings more performance than the previous generation, but Nvidia appears to be playing catch-up to Qualcomm in terms of performance, and won’t offer the same performance/battery flexibility as big.LITTLE.
I’m hesitant to put this in the next-generation category, as really the 4i is just a polished Tegra 3. It will still use the same Cortex A9 configuration, but the processors will be beefed up with more memory and an impressive 2.3GHz clock speed. It features a pumped up GPU, offering 60 cores for a smoother gaming experience than the Tegra 3.
The Tegra 4i also adds a built in LTE network modem, which should make it more appealing to smartphone manufacturers, but it won’t offer performance anywhere close to a Snapdragon chip. Overall this is likely to be a SoC for future mid range handsets, it’s not badly designed but it doesn’t offer anything that hasn’t already been available in the market for a while.
We’ve seen that the new HTC One will be using a Snapdragon 600 processor, but the 600 isn’t really a brand new SoC. In fact, it’s essentially a ramped up S4 Pro Krait CPU, and the GPU is exactly the same.
The LG Optimus G Pro also sports the new Snapdragon 600.
I won’t delve in to the specifics, but the tweaks to the existing Krait architecture and upping the clock to 1.9GHz nets somewhere around a 20-40% performance improvement over the S4 Pro.
The Snapdragon 800 is more of a step up from the current generation, it will feature four new Krait 400 cores running at up to a whopping 2.3GHz, and will pack a new Adreno 330 GPU which is rumored to offer a nearly 50% improvement over the already zippy 320. So it should be faster than both Nvidia’s Tegra 4 and Samsung’s Exynos 5. Just like Nvidia, Qualcomm is keen to make sure it’s up to date with the latest graphics APIs, ensuring that Snapdragon devices will be on the radar for gamers.
The Snapdragon 800 will definitely be one to watch in the future, but we don’t know which devices it will crop up in yet.
Texas Instruments and ST-Ericcson are also each offering a new line of chips this year. The OMAP 5 system on a chip will be available in Q2 2013, and will somewhat resemble ARM’s big.LITTLE architecture. Two powerful Cortex A15′s will be backed up with two lower power Cortex M4′s to keep the processor very energy efficient, whilst offering performance improvements over the previous generation.
It will also pack in a cut down GPU from the one seen in the Exynos 5, using just a dual-core PowerVR SGX544, rather than the three-core variety chosen by Samsung. This will put the OMAP 5 processors as good mid-range chips, but nothing is being done here which is going to scare the competition.
ST-Ericcson is probably the most boring of the bunch, simply powering up and optimizing its NovaThor range rather than introducing an interesting new architecture. An energy efficient quad-core Cortex A9 chip clocked up to a mighty 3GHz will be shown off at MWC, but it will only feature a single-core PowerVR SGX544, leaving it lagging severely behind the competition in terms of GPU performance.
The clock speed is certainly impressive, but when the competition is focusing on energy efficient designs and using newer processor architectures, you can’t help but feel that ST-Ericcson is struggling for innovation.
If you feel like your head is about to explode I totally understand, there are just so many chips with such incremental differences. Although we haven’t managed to cover absolutely everything, hopefully this article has shed some light on the SoC hierarchy, and might just help you know what you can expect when choosing a new device.
2013 is going to be an interesting year for the processor industry, as competition is fiercer than ever. The next generation of chips is going to be quite a big leap up from the current line-up, both in terms of peak performance and energy efficiency. Qualcomm looks to be in the best position at the moment, as it is the first to market with devices using its improved Krait CPUs and Adreno GPUs. But don’t count out big.LITTLE, which is the most interesting architecture in my opinion.
Battery life is not even mentionned once. When you feel no difference in UI on a dual core, don’t upgrade for these marketting 1000000 horsepower chips.
yeah! battery life is so important these days of powerful SoCs, whats worse is that more and more phones come ultra slim and non removable batteries. no one seems to be interested in battery life – it’s either the soc/ camera/ screen/ – while making them all power efficient is one way – investing in battery life would make a huge impact.
Well, that new Novathor chip that does 3GHz only uses 0.6V at 1GHz, which means it uses half the battery power of other 28nm chips, which use about 35% less power than last generation chips. Most of the time, the CPU in your phone is crawling along at a few hundred MHz, so unless you do something you don’t do at the moment, it you’ll get better battery life.
Where did you see information in TI OMAP5? TI has exited the phone/tablet SOC market and is now concentrating 100% on automotive electronics, enterprise communication, industrial micro controllers etc.
OMAP5 may very much end in , but it may end up in the Car Entertainment systems of high end cars such as Audi, BMW, Benz etc. but expecting that in a phone or tablet near you in 2013. Very little chance.
TI hasn’t abandoned the smartphone market from anything I’ve read, but I could be mistaken. It said that it was focusing on diversifying way from smartphones, but TI still list the OMAP 5 under mobile processors. (http://goo.gl/MiHPK)
A quote from the website, they consider the OMAP 5 suitable “for next-generation Smartphones, Tablets, and other mobile devices”
True though, it might never make it’s way into a smartphone.
TI’s message has been muddled. Or muddied.
They said they were exiting mobile, then seemed to back away from that a bit.
Meanwhile, no design wins for new phones means they’re leaving, voluntarily or not. Technical/competitive issues alone are sinking/have sunk them.
What OEM phone designer would choose OMAP at this point given that they seem sunk ?
You should’ve name the article “Android SoCs”, because it seems you forgot one major actor.
Marvell designs its own cores, too.
It’s interesting that you suggest that the Snapdragon 800 will be more power efficient than Tegra 4 and Exynos 5 Octa. Four of the cores in the Exynos chip are low power cores, and the fifth core in the Tegra is a low power core. On the other hand, the cores in the Snapdragon can be individually scaled for voltage and frequency, while those in the other chips are all scaled in their clusters, and can only be off or running at that speed. The use of FD SOI in the Novathor chip allows it to run at upto 1GHz on 0.6V. All of these techniques are aimed at power efficiency in normal use with loads of processing capacity when you need it.
Right. Marvel also has the architectural license. From Wikipedia :
“Companies can also obtain an ARM architectural license for designing their own CPU cores using the ARM instruction set. Distinct ARM architecture implementations by licensees include Apple’s A6, AppliedMicro’s X-Gene, Qualcomm’s Snapdragon and Krait, DEC’s StrongARM, Marvell (formerly Intel) XScale, and Nvidia’s planned Project Denver.”
Well I can’t cover every chip or we’d be here all week. But a couple of you have asked for it, I’ve added some Intel info.
You’ve made a few goods points there, I’ve revised a bit about the big.LITTLE to make it clearer.
As for the Snapdragon 800, based on the rumored speeds it will be clocked faster than the A15′s used in Tegra 4 and big.LITTLE. So as long as the architecture is roughly as fast, the 800 “should” be the fastest chip around. But we’ll have to wait for the benchmarks I guess.
Thanks for the added informations (still wonder why not talking about Apple Swift but I understand it’s a bit off topic here on AA).
Krait 400′s performances is not know at the moment but I guess they won’t reach pure A15 levels.
The fortcoming competition between Krait400, A15 and Swift V2 will be very interesting I think.
No mention of duel core Intel atom from lenovo k900.
Ask and you shall receive!
Great article, but there are few errors.
1. The ARM Cortex A7 is not the least powerful ARM Cortex core, the A5 is. The Cortex A7 sits very close to the A9 in terms of performance, while offering substantial power effiecny and using way less area than the Cortex 9.
2. The IT PowerVR that is roumerd to be used in the upcoming Samsung Exynos SOC that will power the next generation Galaxy S is not the same used in the latest ipad (4th gen) the 4th gen ipad uses the PowerVR 554 which offer twice the performance per core of the 544 (6.4 GFLOPS Vs. 12.8 GFLOPS).
The tri core PowerVR is roumerd to have higher clock speed, therefor it should sit anywhere between the iPhone 5 and 4th gen ipad.
“The Snapdragon 800 … will pack a new Adreno 330 GPU which is rumored to offer a nearly 50% improvement over the already zippy 320. So it should be faster than both Nvidia’s Tegra 4 and Samsung’s Exynos 5.”
actually that really depends for the Exynos 5. The Mali T604, if that is the version they use, will be blown out of the water. But the other two options are the T628 and T678. The T628 should come in around on par with the 320, but the T678 is a totally different beast, about twice as fast as that. Ive seen it in print that the “theoretical max performance for 8 cores” is 550 gflop/s.
PS> To play Crysis 3 at 1080p and max settings but no AA (or a really lightweight AA like FXAA) you need ~750 gflop/s .. we’re talking three, maybe even two years before your phone will do that. :)
loved this article, and like most really good articles it attracts really good comments too!
I think you meant HTC EVO 4G LTE for the Snapdragon S4 Plus. Not the HTC EVO 4G (the original), which used a Snapdragon S1 (very old at this point).
Good article otherwise. Cheers.
There’s needed a writing correction:
After Exynos 5 picture when it says the coming models of it, it was written “octo-core”