It’s been a busy month for the processor industry, as companies have been unveiling their latest technologies at various conferences and lining up manufacturers to take their products to market in new handsets and tablets. In the light of the recent Tegra 4 benchmarks, it’s a scary time for competitors to be talking about going head to head, but considering the release of the rumored Exynos 5 Octa-powered Galaxy S4 is just around the corner, I think it’s about high time we finally stacked these next generation chips up against each other.
Of course we’re yet to get our hands on any devices using these two chips, so everything discussed in this article is based on the best information currently available, rather than benchmarks I’ve conducted myself. But even so, we should be able to judge how these two SoCs compare relatively accurately.
Both Nivida’s latest high end chip and Samsung’s next generation Exynos 5 SoCs will be using the new Cortex A15 architecture for the lead CPU. This is partly responsible for Nvidia’s impressive benchmark results, as the Cortex A15 is the fastest processor available in the market at the moment, substantially beating out the widely used older Cortex A9s from devices like the Galaxy S3, Note 2, and a plethora of tablets. The Cortex A15 also takes the performance crown from Qualcomm’s S4 Pro, and should also beat out the beefed up Snapdragon 600 as well.
The evolution of ARM Cortex-A series processors, resulting in smaller, faster, multi-core chips.
The Tegra 4 sticks with the same 4+1 design as the Tegra 3, utilizing four main cores for general processing and a lower clocked companion core to run background processes and save on idle battery drain. The main processors will be clocked up to an impressive 1.9GHz, whilst the companion core will peak at 800MHz.
Samsung’s Exynos 5 series will be using four Cortex A15 cores, just like the Tegra 4, which will be clocked at 1.8GHz. This is ever so slightly slower than the Tegra 4′s peak frequency, but in real world applications users are not going to notice any difference in peak CPU performance.
The Exynos 5 deviates from standard processor designs by introducing the new big.LITTLE architecture. In this design, the four Cortex A15s are backed up by four separate Cortex A7 cores, but rather than turning this chip into an eight core monster, the A7s are designed to keep power consumption low whilst the device is performing basic tasks.
You’ll notice that the A7s are pretty low down on ARM’s performance sheet, as they aren’t designed for heavy duty processing, but four simple cores are actually more than you’ll need for basic Internet browsing, email syncing or playing music. But the real benefits from big.LITTLE shine through when it comes to prioritizing processing needs, as the chip can switch between cores in just 20,000 cycles. To put that seemingly large figure in perspective, it takes just 20 microseconds to switch over when the processor is running at only 1GHz.
You don’t have to compromise between leaving a process running on a low power core or draining your battery, processing can be transferred between A7 and A15 cores mid-cycle, ensuring that you always have maximum performance on stand by, whilst keeping your battery healthy. This demonstration from MWC shows off big.LITTLE in action.
Overall the Tegra 4 and Exynos 5 Octa are going to provide users with virtually identical peak performance, and are both head and shoulders above any other processors currently on the market. We’ll have to look a bit further at GPU performance and energy efficiency to decide which is the better chip.
It’s hard to directly compare the potential GPU performance without a head-to-head benchmark, but we can probably hazard a good guess based on the bits and pieces of information floating around.
Nivida’s new and improved 72-core GPU blitzed Apple’s A6 chip used in the Phone 5, and comes up trumps against PowerVR SGX543MP4 GPU in the benchmarks we saw last month, scoring an impressive 4148 on Geekbench and 57fps on the Egypt HD compared with the iPhone 5′s 1640 and 27fps, respectively. That’s a two fold increase in terms of frame rate, which is nothing to be shy about.
Next gen GPU performance is going to be key if devices like Project Shield hope to capture the interest of Android gamers.
But how does this stack up against the top of the line tablets? Well, the fourth generation iPad scored a respectable 52fps on the same Egypt HD benchmark, whilst the MatlT604 powered Nexus 10 only managed 33fps. So Tegra 4 is fast, very fast, and is the quickest GPU on the market by a fair margin. Granted this is only a single benchmark result, but it’s a decent enough indicator of whereabouts real world performance should lie.
Samsung’s Exynos 5 Octa chip is going to utilize a tri-core PowerVR SGX544MP3 clocked up to an impressive 533MHz, or significantly faster clock-wise than the same GPU which is used in Apple’s A6X, which is only clocked at 300MHz. Of course the absence of a fourth core, compared with the A6X chip, is going to lop off a decent amount of performance, but perhaps the increased clock speed can make up the gap.
There is good evidence to suggest that an overclocked three-core SGX544 could end up besting its quad-core brother. If we recall back to the previous SGX543 chips used in Apple products, the iPhone 5 featured a higher clocked three-core GPU than the iPad’s quad-core, and managed to best the chip by a couple of frames on the Egypt HD benchmark (according to Anandtech). So there’s a good chance that the Eyxnos 5 Octa could pull performance roughly around the iPad 4, which is very respectable.
The Exynos 5 Octa will definitely perform better than the MaliT604 seen in the dual core Exynos 5 variant, and will likely sit very close to the quad core SGX544MP4 used in the latest iPad. The difference in performance is likely to be slight and will vary based on the application. The Exynos 5 Octa will in all likelihood fall short of the mighty Tegra 4, but of course we’ll have to wait for the benchmarks to be exact.
What is likely to interest many handset developers is what other features, each of the SoCs add to their products, besides brute processing power.
Firstly and most disappointingly, neither of these two chips will come with a build in LTE modem. Considering Qualcomm has had built in LTE compatibility available for a while you have to wonder what these two tech giants are up to. Unless of course they are targeting these SoCs more at tablets than smartphones, in which case they are forgiven.
As LTE adoption continues to grow, missing out on producing an integrated LTE chip seems like a poor business decision.
Nvidia has upped it’s game since the Tegra 3, and not just in terms of hardware specs. Its new chip comes with support for higher resolution displays, up to a massive 3200 x 2000, and will happily export images to 4K resolution screens via an HDMI output. We’ve also heard lots of news recently about Tegra-specific graphics options in games, and support for new APIs like DirectX11 and OpenGL ES 3.0. Nvidia is clearly keen to live up to it’s reputation as a gaming chip this time around.
Tegra 4 also supports up to 4GB of DDR3L memory and a wide range of LTE and HSPA+ bands, so if manufacturers are prepared to add on an optional LTE chipset the Tegra 4 can finally be used on networks all around the world.
Information on the exact features of the Exynos 5 Octa are a little thinner on the ground, but from what I’ve found out, the chip shares many similar features as far as network support, memory and connectivity. The real difference comes in the form of graphics, the SGX544 GPU is older than the Tegra 4, and is only certified for older APIs like OpenGL ES 2.0 and DirectX 10.1. Unfortunately the white sheet doesn’t mention supported resolutions so I wouldn’t hold my break for 4K compatibility, but at least 1080p displays and codecs are confirmed.
We know that the Exynos 5 Dual is based on a 32nm process, but the Octa is supposedly going to use a 28nm die just like the Tegra 4. So heat production should be roughly similar.
Tegra 4 deserves to win this round as it seems to support a larger number of optional technologies, but this could change as the last pieces of Exynos 5 information drip through.
This is where I believe the two chips are really going to clash heads. Nvidia was the first to really get to grips with the need for more energy efficient SoC designs when it came to low performance tasks, but has big.LITTLE manage to take the idea one step further?
With ever more power hungry processors our poor batteries are being drained faster than ever. Balancing performance and power is even more important in the next generation of SoCs.
We’ll start off by looking at the Tegra architecture, which, as I mentioned, runs from five Cortex A15 cores. The main four cores run at a peak frequency of 1.9GHz, whilst the low power companion core runs at between 700 and 800MHz. Just like the Tegra 3, the fifth core won’t be visible to the operating system to assign applications too. Instead it operates solely in the background, performing tasks, like syncing emails, checking Twitter updates, or processing audio, whilst your device is in standby mode.
The Tegra design does go some way to help reduce active battery drain as well, by adjusting clock speeds and gating each of the cores to use them only when needed. But this is common practice amongst nearly every chip manufacturer, and doesn’t provide any perks over other manufacture’s designs.
Many users were disappointed with the Tegra 3′s battery life as the 4PLUS1 core architecture only provided decent battery saving benefits whilst the device is asleep, but didn’t save significantly on energy whist it’s awake. Sadly, Tegra 4 doesn’t do anything to address this complaint, and with more powerful Cortex A15 processors draining even more power than the Tegra 3′s A9s you can expect some significant battery drain whilst gaming, etc.
The Big.LITTLE design in the Exynos 5 on the other hand has totally switched things up since the Exynos 4. The SoC is split into two sets of quad-cores: four A15 for heavy duty processing, just like Tegra 4, and four lower power Cortex A7s for general processing. The diagram from the big.LITTLE white paper says 1000 words: the peak drain from an A7 doesn’t even come close to the minimum power drain from a single Cortex A15.
As we saw in the video at the start of the article, the Exynos 5 can switch between cores incredibly quickly, keeping battery drain to a minimum when performing moderately demanding tasks like quickly loading flash heavy webpages or starting up a new application. This is something that Tegra 4 simply cannot do, adjusting clock speeds and gating cores simply isn’t as efficient as being able to switch down to a lower power core.
The Exynos 5 Octa also benefits massively from having all of these processors visible to the operating system, allowing for core assignment on the fly. As Tegra doesn’t allow for applications to be assigned to the low power core, evening running something simple like an email client will require an A15 to be switched on and the battery begins to drain, whilst the Exynos 5 can do this easily on an A7.
The big.LITTLE architecture should win in this category by a long way. There simply isn’t another architecture like it available from any of its competitors.
It’s tough to say exactly who comes out on top in this match up. Performance-wise both chips are very similar, but I’d give the slight edge to Nvidia’s Tegra 4 thanks to its impressive GPU benchmarks. On the other hand the Exynos 5 Octa offers and incredibly power efficient design, which is becoming more and more important as high performance chips eat up ever more battery.
Nvidia also comes out slightly ahead when you look at additional chip features. But even though Nvidia supports outputs to 4K displays, I don’t see a major deal breaker here for the average consumer. Neither side comes with a built in LTE modem, so manufacturers won’t have a clear preference there either.
If you are to pressure me for a winner, I’d probably have to pick the Exynos 5 Octa simply because of big.LITTLE. Nvidia’s chip comes out slightly ahead when you jot everything down on paper, but prolonged battery life whilst maintaining very similar performance is surely the better deal. And I’m positive that the majority consumers will prefer big.LITTLE over 4K HDMI support, DirectX11, or a few extra frames per second whilst gaming. But your opinion will obviously depend what you are looking for.
So does this mean that I think the Tegra 4 will be doomed to live out its days in just a few devices like the Tegra 3? Well no, Tegra 4 is an excellent high performance chip, and is the preferable choice when it comes to picking a SoC for gaming. Better performance, the latest API support, and Tegra specific graphics options will definitely make devices like Project Shield appealing to Android gamers, and it would be a very welcome SoC in a high performance 10 inch tablet as well.
I personally expect the Exynos 5 Octa and Tegra 4 to be more popular in high-end tablets than smartphones.
However, I don’t see the Tegra 4 being popular amongst smartphone manufacturers due to the relatively high power consumption compared with competing chips and lack of built in LTE.
Similarly I think the Exynos 5 Octa is much more suited to tablets than handsets, although big.LITTLE does open the door for a smartphone, like the leaked EU Galaxy S4, to dabble in high performance quad-core goodness. Time will tell just how much big.LITTLE can prolong your battery life, but personally I’d rather see a dual-core big.LITTLE A15/A7 chip for smartphones.
Either way, I’m looking forward to seeing how devices using the two chips stack up in the real world. Feel free to let me know if you agree with my analysis, and if you’re particularly excited for either chip set.
Like this post? Share it!
Can you run just 1 A7 instead of all 4 A7 in the LITTLE? Personally I don’t think you can, but still doubtful.
And if you cannot run a single A7 alone, then your power consumption analysis from the graph is wrong because that is just one A7′s power consumption whereas you will be running 4xA7 cores.
Yes you can that’s what this slides says which Samsung presented at ISSC 2013
How can you compare tegra to exynos if you do not have the exynos idiots.
ha! Well I think this was a great article, he did a lot with little set-in-stone details. Great coverage Mr. Triggs!
Even if the 4 A7′s are better for battery, it’s not as powerful as an A15…so what about fastest to sleep time? Tegra 4′s companion A15 core, may be able to finish tasks faster, go to sleep faster and save better power, compared to A7.
and since the companion core is now much more powerful in T4 (compared to T3), single threaded operations could be done on that core almost most of the time…check out anandtech’s article on power savings for T4.
Sounds like rumors suggesting Samsung big.LITTLE is got some thermal issues…
But with the t4 the companion core is recognized by the os. One the screen is on, its in all out quad core battery draining mode
not necessarily…the main 4 cores can be power-gated, right? So, a single A15-main core can still run by itself…get tasks done faster and power down.
curious, does T3/4 companion core only run during sleep mode? I thought it still ran even during regular operations….
I prefer Exynos than Tegra 4.
My site: http://www.technewz.gr
According to Samsheep Exynos 5 is untouchable! Octa core my mass! It’s just dual quad core with only 1 quad core working at anytime!
“Octa core my mass!” just sounds so good!
That would be a concern but if you watch the video it appears at 0:09 that there are 3 A7s and 1 A15 running at the same time.
GPU part is totally messed up.
- iPad 3 : SGX543MP4
- iPhone 5 : SGX543MP3 (same as Exynos Octa)
- iPad 4 : SGX554MP4 (nex generation graphics chip, not 543/544)
One more GPU-thing: “[The Tegra 4 scores]… 57fps on the Egypt HD compared with the iPhone 5″ This is true, but it misses the prize. GLbench’s official benchmark list online has complete Egypt HD scores for the A6X too: in fps it is 52fps. The Tegra 4 therefore is not just better than the iPhone 5, but even better than the A6X-based (ie, the latest) ipad.
Read the article again dude.
Wow. I really enjoyed this article – the first real breakdown I’ve seen comparing the upcoming chips. It’s still all SPECulation but great thoughts nonetheless. Thanks!
This is all great and wonderful but everyone is missing the fact that nothing in Android has been developed to take advantage of even last years dual cores. The software needs to catch up. All I see this doing is letting software makers be lazy. Two ways of getting great performance. Put a V8 engine in a poorly tuned car, or put a v4 engine in a very well tuned car, get the analogy? The engine = proc the tuned car = software. We need to start writing apps that have multithreading support. Thats not to say we don’t need to keep making more advanced chips like the Exynos 5 or Tegra just that if we don’t write software that takes advange of it theres not really a point.
That being said Tegra has had optimized games out for a while, but really only a few are actually optimized the other just get the moniker slapped on them.
I have to agree, most Androids are already very powerfull yet with bad performance, battery life or not optimised this is all due to bad software, drivers, kernels (from manufacturers), and on the other side bad apps wich are powerhungry, not optimised to take advantage of the hardware.
Many vendors have added on stuff, we usually seen described as “custom UI” elements, which may take advantage of it though. Samsung for instance has a multitasking system in theirs right?
I agree with you mostly, Sony and recently Motorola, does a fair job with UX optimization, but the issue isn’t about the UX, which takes very little power to run. The issue is writing multithreaded apps. Very few non games can use multiple cpu threads.
And this could be fine. If the distributed load is done properly, but at this point we are talking about more power spent on knowing when to distribute the load to each processor.
As for Samsung its hit and miss with their UI elements.
Nature UX is pretty fast but if you play with a Gallaxy Note 10.1 you will discover how horribly Samsung has optimized or rather the total lack of optimization.
Hope this helps. Got nothing against those who like Nature UX, just not my thing.
I totally agree on this as well, i have both iphone5 and Galaxy S4 (i9505) and my dual core iphone feels smoother most of the time, sure some say its not “real” multitasking but i personally think its Google’s fault for having a to open policy for the application and not putting enough pressure on developers to make up to date and efficient apps. Also Google itself needs to make the framework more streamlined so that developers dont make up their own rules and make strange things in the apps. Sometimes it feels like a puzzle figuring out how some apps work.
But as you say the Android phones are far far more superior in hardware but very poorly in software. I love my S4 but i wouldn’t mind if it could be more snappy.
Renasas have done a dual core big little with integrated lte modem FYI
Renasas’s APE6 will be one of the world’s first SoCs using the PowerVR series 6 Rogue graphics engine. Can’t wait!
Why do you claim the Octa lacks LTE? Recent leaks shows LTE capabilities.
any Octa-core released in the USA will not have on-silicon LTE, unless Samsung strikes a deal with Qualcomm to avoid their conflicting patents (both have patents on the tech).
article aside, wouldn’t mind seeing the tegra 4 in the next nexus
Renesas Mobile Octa core shown at last week’s MWC is a much higher performance device w 4x2Ghz A15s, 4×1+Ghz A7s, SGX6MP4 GPU. Why no love?
Because 12W+ power consumption :-D
I always facepalm when I see a comparaison between the S600 vs Tegra 4 and Exynos 5 Octa, the latter two are not optimised for phones and can only be used on tablets (and phablets with 3000+mAh batteries), and the only you can put either on a phone is either to lower the frequency to around 1Ghz or even lower or slash the number of cores to two instead of 4, the S600 is already optimised and work perfectly with phones while being very efficient.
On another note I doubt on the CPU side the performance difference wouldn’t be that huge of course the A15 would be faster but again it depends on many other factors such as memory used, brandwidth, cache….., Scorpio vs A9 is a prove of that and now Krait vs A15 should be similar especially that now the krait architecture been out for a while and didn’t cease to improve
It would be interesting to see how these two would fare against AMD Temash Soc which too is coming in tablets
Xaxaxa , this autor is amateur ! -1
The GLBenchmarks is a STUPID.
GFLOPS Performance (offscreen) :
Tegra 4 (Wayne) is a 96 GFLOPS = 48 + 24 x 0.672 x 2
IT PowerVR SGX554 MP4 (A6X) is a 80 GFLOPS = 8 x 4 x 0.280 x 9
ARM Mali T604 MP4 is a 72 GFLOPS = 4 x 17 x 0.533 x 2
IT PowerVR SGX544 MP3 is a 57 GFLOPS
IT PowerVR SGX544 MP3 is a 57 GFLOPS
Xaxaxa , this autor is amateur ! -1.
Thus, from Exynos 5250 to 5410, the GPU of the Galaxy S4 is a disappointing downgrade compared to the Nexus 10.
sorry but i will not adopt into big.LITTLE untill its seen real world applications. the switching will be the biggest kill if its not done right. this was a hard lesson learned for tegra2. it will be very noticable in music, games, videos if its done wrong. and since this is first attempts yeah there will be problems and sadly no software patch or help from android modding community will get rid of it (only mitigating but yeah..). tegra2 could not turn on its dual core or off its dual core efficiantly enough and music skippding, and frame drops in videos and games were noticable. tegra3 fixed that issues cause nvidia learned of this mistake (maybe that companion core really did the trick who knows) but when this chip switches even if its supposedly 20ns im sorry but these 3 things continue to runso if its not a positive handoff it will be noticable.
who know it may not have that issue but then again it could. it happened once and to a very big name in the computer industry that has plenty of experiance and quality and excellance yet a major screw up was there. and this big.LITTLE is a completely new step in the industry so it could happen. i would wait untill others try out and waite for reviews. if all 4 shut off of a7 and then a15 turns on then yes there will be problems and that 20ns will be very noticable in music, games and videos. but if 1-3 of the a7 turn off then 1-4 a15 turns on then the last a7 turns off then i think all will be well that will be a smooth change. i dunno. the instruction set will make it or break it for this idea since if the instrction set for a dual core is wrong and all these issues are present and that was just turning on/off a core while the first was never touched and here we all talking about a complete switch of cpus im worried.
Snapdragon 800 with adreno 330 gpu beats out either soc.