This year’s 64bit Mediatek processors may have gained them a larger market share, but the next generation of SoC could see them dominate the processor market.
Mediatek’s stellar year has been aided by the release of their 64bit processors and the performance advantage they offer over similar spec Qualcomm chips. Compared to previous years when fans wished brands would opt for a Snapdragon SoC, this year fans are opting for MT6752 processors over SD615, and in the near future higher end Qualcomm chips could also be under threat.
Today, images from a Mediatek sideshow tell us details of the Mediatek MT6797, the worlds first 10 core, tri-cluster processor.
The idea behind 10 core is basically an evolution of big.LITTLE which we see in some chips today, but instead of just the large and small the tri-cluster set up allows for TINY.MEDIUM.HUGE. With each step comes a boost in performance, the lower the step the better that battery performance.
Gizchina News of the week
So far we have only seen details of one processor using the new architecture, the MT6797. This chipset will use a Tiny core running on 4 x A53 cores running at 1.4ghz to offer the best power effciency. Opening more tasking applications will bump the system up to its Medium setting where 4 x A53 cores running at 2.0Ghz will take over. For more extreme gaming and processor heavy tasks the Big stage of the chip will kick in running on 2 x A72 cores at 2.5Ghz.
According to the slides the MT6797, aka the Helios 20 will be capable of Antutu benchmarks of over 70,000 points!
Other details of the SoC are still unknown, but the chipset will likely retain LTE support, as well as support for higher resolution displays and next generation camera sensors.
What do you make of this processor evolution? Is this what we have been waiting for or is it all a little overkill?
[ Source ]
Mediatek has come such a long way! But I wonder about the different A53 cores. Can’t they just make them dynamically tuned? Like, they start with lower frequencies @ 1.4Ghz and when needed they boost to 2Ghz? The other two cores, the A72 are totally different designs so they can stay but I wonder if the different clocked A53’s are really needed. It would surely mean a hell of a less surface on the chip = smaller chip = less power consumption = less cost + longer life on the battery. Right? (probably not that easy ;))
Absolutely right.
In fact, I’d have wished them to put in two extra A72s instead, for being symmetric with the quad A53 workload, but that would bump the cost up as well.
Yeah but if this is going to be there hi-end flagship processor. Would it really cost so much more to dump the mid-range A53’s 2Ghz processors and make it:
4x A53 low power consumption 1.4Ghz
4x A72 hi-end 2.5Ghz
As the user Realjjj says below here; it probably sounds better for marketing. And besides that; there are probably a dozen other technical aspects we don’t know anything about 😉
I’m with you… but you’d lose the “marketing magic” 😛
I’m with you… but you’d lose the “marketing magic” 😛
It’s all about marketing.
The fact they used Antutu to promote it should make us think.
Those A53 over clocked at 2GHz are just for the show.
They are not meant to reach those frequencies, and definitely can’t sustain heavy load for long period without throttling. They are perfect to run benchmarks though.
mediatek soc made for antutu
there are/will be other solutions, I believe that kind of configuration will be used as well when they will announce a “smaller and less energy eater” SoC.
For a smartphone it would be desirable an hexacore instead:
A 2*A72 at 2.0Ghz+ 2*A53 at 1.7Ghz+ 2*A53 at 1.2Ghz
Lower price, balanced performance and greater battery life.
That chip belongs to the tablet market. Period.
It’s not the same thing. This is about optimization for power/performance. This is a static optimization and it’s done by modifying the HW itself. Deepening on what HW implementation you choose you can have an A53 core that cand go up to 2 GHz but use a bit more power or one that can go up to 1.4 GHz but use very little power.
There’s also dynamic scaling of cores but that’s not as effective as doing a HW modification to create a low power core.
Yes, I though it would be something like that. Seems like a waste of surface but it’s likely to be the most efficient this way.
Also we must keep in mind the marketing reasons ( 10 cores sounds really good and better than 4 or 8 ) and the fact that given well optimized app a SOC could actually use these cores together.
Imagine a game using 2 threads running on the poweful A72 cores… it could also run 2 secondary threads on the 2GHz A53 while the some system background threads could share the 1.4 GHz A53 cores.
We live in a world where parallel is the name of the game.
Absolutely, more cores on lower frequencies are better but only if the code is optimized for multi-core use. Now most apps use only two or four cores.
True but Android smartphones have many processes running at the same time so even if you have one game in background using say 4 cores there’s still 6 more cores to handle background tasks.
And if the hardware is there with many cores maybe the software will support these many-core SOCs eventually.
Yes, but I thought that would start with the first octa-cores. But I haven’t found any evidence of any real performance with so many cores. Nowhere I found a document stating the workings of how Android can put so many different processor though the processor. It’s all only about four cores.
With the launch of the MediaTek 6592 octa-core they told the media that Chrome would be able to use all 8 of the cores but I haven’t found any proof of that.
In short; as MediaTek starts to get bigger and bigger maybe developers will start to build apps with 8 (or 10) core support in it. Until that time it’s good to see they use the many cores for power-level balance.
Yes, but I thought that would start with the first octa-cores. But I haven’t found any evidence of any real performance with so many cores. Nowhere I found a document stating the workings of how Android can put so many different processor though the processor. It’s all only about four cores.
With the launch of the MediaTek 6592 octa-core they told the media that Chrome would be able to use all 8 of the cores but I haven’t found any proof of that.
In short; as MediaTek starts to get bigger and bigger maybe developers will start to build apps with 8 (or 10) core support in it. Until that time it’s good to see they use the many cores for power-level balance.
True but Android smartphones have many processes running at the same time so even if you have one game in background using say 4 cores there’s still 6 more cores to handle background tasks.
And if the hardware is there with many cores maybe the software will support these many-core SOCs eventually.
Absolutely, more cores on lower frequencies are better but only if the code is optimized for multi-core use. Now most apps use only two or four cores.
Dont think it is so easy to implement, most apps use 2-4 cores and many only one.
But to run A72 and A53 at 2ghz when loaded and 4 slower A53 cores when idle is a good thing.
What would make more sense would be a 4/4/2 configuration. (A72 / A53@2GHz / A53@1.4GHz).
A53s@1.4GHz when topped, will consume the same power as the middle ones at the same speed, so if you bottleneck the lower 1.4GHz then it’s already time to put the middle ones to work without power penalty, because the 2GHz can scale back to 1.4GHz on the same power envelope. What the middle ones can’t scale back is to the base clock frequency of the 1.4GHz when on the first multiplier.
That’s the only reason the 1.4GHz ones exist.
Of course 4xA72 that would be a lot more expensive (and ocuppy a lot more circuit real estate), so maybe it’s not possible/economicaly viable.
I always wondered about those setups: what happen’s when a single-core process is running? It obviously doesn’t max the quad A53 group (only one processor is used) so the process would not migrate to the powerful A72s and you phone would be left to seem slow while it isn’t…
It would. Consider that A53 @ 1GHz (stellar performance / power consumption frequency) is as fast as A7 @ 1.35GHz.
Probably it depends whether it bottlenecks that one processor. If it does, it should migrate.
But how would the os know? Android may well balance the load on all four cores, i.e. the OS would think that only 25% of the cluster is used (one core’s load balanced to four cores). Or am i missing something?
What would make more sense would be a 4/4/2 configuration. (A72 / A53@2GHz / A53@1.4GHz).
A53s@1.4GHz when topped, will consume the same power as the middle ones at the same speed, so if you bottleneck the lower 1.4GHz then it’s already time to put the middle ones to work without power penalty, because the 2GHz can scale back to 1.4GHz on the same power envelope. What the middle ones can’t scale back is to the base clock frequency of the 1.4GHz when on the first multiplier.
That’s the only reason the 1.4GHz ones exist.
Of course 4xA72 that would be a lot more expensive (and ocuppy a lot more circuit real estate), so maybe it’s not possible/economicaly viable.
Marketing? If the do the right thing right away, what will they sell later? Next tri-cluster generation will follow this idea and will be marketed as something new.
Marketing? If the do the right thing right away, what will they sell later? Next tri-cluster generation will follow this idea and will be marketed as something new.
you all need to understand – not every piece of silicone are the same, not every A53 core capable of 2.0ghz, so, you want mediatech to throw them all? just look at PC CPUs or even mobile SOC, when you overclock them, you probably won’t have the same results as other users.
Sweet 🙂
Thanks for sharing. That was a great article.
I gotta say that Mediatek is really going gaga. Not a bad arrangement I must say, but do we really need all this processing power in a smartphone ???
I see a future when our smartphones will run octa-core i7 grade CPUs and GeForce 9XX ultra grade GPUs too.
They should slow down a bit. Though I know that this is good for business and thromping Qualcomm.
Beware Qualcomm … HUGE BLOW coming …
The pace at which everyone is going is insane. Heard Sharp is working on 5.5inch 4k display with a mind blowing 806ppi. Do we really need that? Soon phones will pack 8k displays, 8gb RAM, 1TB memory and processing power make antutu show error in computation.
It will make Antutu throw a division by 0 error :))
point is , will those $5000 phones be WORTH ALL OUR TIME ?!
who on EARTH will do hardcore gaming on 4k Smartphone screens when the PCMASTERRACE and consoles already HAVE THAT OPTION aplenty…
Someone that wants to play Madden while on the train/plane/anywhere but home.
homeless 😛 ?
Someone that wants to play Madden while on the train/plane/anywhere but home.
“Do we really need that?”
YES if you VR, not if you don’t.
In fact 4K is barely enough for VR, 8K would be better. Have you tried Google’s cardboard or Gear VR those giant pixels are fugly. Then again most people don’t VR (or even don’t know of its usefulness) so yeah it would be an overkill for most…
even with all this monstre hardware, 90% of the users will continue using their smartphone for candy crush, whatsapp and all the other 1-core-needed apps from playstore. software is not growing on par with hardware and the mainstream one is even more basic (since it has to work with old hardware and entry level models), like having a ferrari and using it only inside a crowded urban centre.
That’s where Ferraris are used the most anyway.
Urban centres, the more crowded the better it is.
What’s the point on driving a Ferrari if nobody is looking at you?
“used most” doesn’t mean “used properly” 🙂
Agreed, but isn’t this happening with phone as well?
😀
exactly, that was my point (and yours as well) 🙂
you forgot the 16gb of ram 😉
Of course … I forgot about the RAM too. Hehe … lets keep watching them
Well they waste die area with all those A53 cores , at some 4.5mm2 for each quad cluster, they could have went with 2 or 4 A53 cores and use a bigger GPU. Guess 10 cores sounds better for marketing.
Depending on when it comes to market and how it’s priced , it will be nice or will suck. If it comes next year, it’s not exciting since Qualcomm will likely have better chips,.if it come this summer,then it’s fantastic.
This would beat the Galaxy S6 in single threaded CPU perf and maybe match the ipad Air 2. So pretty much would be the fastest phone SoC in CPU perf until SD820 arrives but it would have to arrive before it.
What i like the most is the 128-bit memory bus, should help plenty and nice to see that from MT since their midrange SoCs use single channel 32 bit and the top chip dual chan 32 bit so this doubles that.
One article says mass production is in July. If that is true, we can expect phones with that chipset late third quarter or fourth quarter. It’s built on 20nm process.
I also want to see those processors in my pc and notebook.
x86 processors are still much faster than these arm chipsets
Mediatek, we don’t need more power, give us more power-optimized SoCs and GPS performance on the same level as snapdragon.
Agree totally, GPS performance is my biggest problem.
I don’t think GPS is a problem for mediatek SoC since 6595/6732/52 or newer. Even in my experience, my jiayu s3 perform better gps locking than my iuni u3.
the GPS problems is an old story of mediatek.
While I do agree that it’s better then 6592/6582/6589, 6595 (Lenovo Vibe X2) still required 3 times more the time for a lock then my oneplus one in equal conditions.
While I do agree that it’s better then 6592/6582/6589, 6595 (Lenovo Vibe X2) still required 3 times more the time for a lock then my oneplus one in equal conditions.
I would not mind about 2 minutes lock, the problem is accuracy.
+1 we need more features from mediatek (like Qualcomm’s IZCat, 7.1 surround etc). We need more open source approach so 3rd party developers port-build ROMS for our phones. Processing power is nice but graphics power on mediatek’s SoCs is still a bottleneck in an era where portable devices used as portable multimedia players most of the time.
GPS is fixed. GLONASS and Beidu becomes a standard. The crappy build quality of the antennas and the buggy software are another subject to discuss.
+1 we need more features from mediatek (like Qualcomm’s IZCat, 7.1 surround etc). We need more open source approach so 3rd party developers port-build ROMS for our phones. Processing power is nice but graphics power on mediatek’s SoCs is still a bottleneck in an era where portable devices used as portable multimedia players most of the time.
GPS is fixed. GLONASS and Beidu becomes a standard. The crappy build quality of the antennas and the buggy software are another subject to discuss.
Good thing is they didn’t design it as true 10 core A53.
Another good thing is dual Cortex-A72. I would like a quad A72 cluster better instead to 2 quad cluster of A53, but at 2.5GHz those should have deliver up to 1900 single core score at Geekbench, based on Huawei’s numbers about A72.
I suppose Mediatek’s goal is reaching better Antutu score with cheaper chip again. CPU part’s die area of that chip will be less than a quad A72 + quad A53 chip like Snapdragon 620.
That 3 cluster arrangement could help power consumption a bit I guess.
GPU part and lithography is missing. I guess they could go for 20nm because it’s cheaper than 16nm FinFET.
GPU could be PowerVR GX6250 like MT8173. They implemented PowerVR G6200 first in MT8135 around mid 2014, then in MT6595 and lastly in MT6795.
According to a Chinese article, this chipset is manufactured on 20nm process.
Good thing is they didn’t design it as true 10 core A53.
Another good thing is dual Cortex-A72. I would like a quad A72 cluster better instead to 2 quad cluster of A53, but at 2.5GHz those should have deliver up to 1900 single core score at Geekbench, based on Huawei’s numbers about A72.
I suppose Mediatek’s goal is reaching better Antutu score with cheaper chip again. CPU part’s die area of that chip will be less than a quad A72 + quad A53 chip like Snapdragon 620.
That 3 cluster arrangement could help power consumption a bit I guess.
GPU part and lithography is missing. I guess they could go for 20nm because it’s cheaper than 16nm FinFET.
GPU could be PowerVR GX6250 like MT8173. They implemented PowerVR G6200 first in MT8135 around mid 2014, then in MT6595 and lastly in MT6795.
Whilst this makes more sense than what I imagined when I first heard about the 10 core idea, I’m still more interested in what they have planned in the GPU department as all this processing power is gonna be useless if they don’t implement a decent GPU to match.
a slide like that suggests that they have already done internal tests with antutu and have a 70k result .. it should be pretty fast .. but with 10 cores sharing the load, .. single thread performance (for those long sql transactions we’re all doing with our mobile phones) would be slow.
New high score for the dual A72 tablet SoC http://browser.primatelabs.com/geekbench3/compare/2378944?baseline=2375021
This phone SoC is slightly higher clocks and the wider memory bus would help perf further so , no matter what others do , the perf will be nice.
impressive but what we really need is better GPS and a very competitive GPU…
GPS working fine or not depends mostly on the Chinese phone design, not on the chipset. From what I have heard, GPS on MT6732 and MT6752 phones are working great.
Agree the MT6732 in the ulefone be pro is spot on I can’t fault it, but the MT6752a in the kingzone Z1 is very poor as is LTE reception compared to the be pro.
The GPU actually matters little.
First it’s just for gaming.
Second you got 2 categories of mobile gamers. One is older people that play Angry Birds type of stupid games and those don’t require much of a GPU. The other is kids and they might require a good GPU.
Third even in more demanding gaming you only need above 30FPS min frame rates and phones max out at 60FPS. So there is no ponit in a GPU that can push 120FPS if the screen can only go 60. And ofc FPS depends on screen resolution. Unfortunately very few test FPS in actual games and because of that people want the best perf instead of the perf they need.
So a big GPU is mostly all marketing because after all it has a limited purpose and is improperly benchmarked by pretty much everybody.
Chip makers waste a huge die area on the GPU as it is even if it’s a lot less useful than the CPU.
The Samsung Exynos 5433 is about 113mm2. The CPU cores and cache are about 19.7mm2 while the GPU is 30.9mm2.
Apple’s phone SoC allocates some 12.2mm2 to it’s 2 cores plus cache and some 19.1mm2 to the GPU.
We would all be better off if tech sites stop the nonsense and properly benchmark GPUs so chip makers can shift back to focusing on things that we actually need instead of GPUs that are too big.
the earlier antutu screenshot leak suggests that the GPU outdoes the HTC One M8’s adreno 330…very excited to see which one MediaTek put in this time..
No wonder why they used that Antutu slide.
Part of their success comes from that too.
Not very professional but hey… It’s all about money!
While their primary goal is performance it’s interesting to see that they are focusing on efficiency as well.
In not sure this is the right path due to those A53 cores clocked at 2GHz, i guess we have to wait before saying it.
If it’s going to be a 20nm process then there’s not much room for a big GPU.
This is plain stupid! Common people is it really that hard making intelligent way to use frequency table & hot plug? A72 @ 1GHz won’t use more power than A53 @ 1.5 GHz & performance will be the same. This is so stupid way to waste silicon that I don’t have words for it.
When can we expect devices with 6797 SoC ? What Snapdragon/Exynos/Kirin will have then?
If they also start using AMD’s GCN architecture in their hi end SOCs in the future, they will end up the top brand, leaving Qualcomm and the rest far behind.
Why not set the six Cortex A53s clock dynamically..or maybe use two Cortex A53 , four Cortex A53e (newer version, more energy efficient at higher clock speeds) and another quad Cortex A72?