Mediatek are currently promoting their next generation MT6797 Helio 20 SoC to manufacturers revealing more details of the chipset.
Mediatek are courting Chinese phone makers in an attempt to lure them over to using the next generation Helio 20 in flagship devices rather than the Qualcomm Snapdragon 810.
2015 has been a great year for Mediatek so far, with their 64bit chipsets performing well and finding their way in to Chinese and international brand phones, a trend they hope to continue with the MT6797 Helio 20.
We’ve already posted details of the Tri-Cluster, 10 core processor, but a new comparison table against the Snapdragon 810 gives us an easy reference for the full specification of the chipset.
As previously mentioned the MT6797 will run a small, medium and large core ranging from 4xA53 at 1.4Ghz to 2xA72 at 2.5Ghz for more intensive use. This new table also confirms that the MT6797 will be made on a 20nm process to create a smaller and more efficient chipset.
Gizchina News of the week
The MT6797 will also support 25 mega-pixel main cameras, super slow motion video recording and Native3D 2.0 for three-dimensional stills. 4K video recording and playback will both be possible, with 4K HDR video being a new feature for the chip.
Graphics are handled by an ARM Mali T880MP4 700mHz GPU, while screen support remains at WQXGA 2560 x 1600 resolution panels.
While the MT6797 Helio 20 does look like one formidable processor the Snapdragon 810 still has a few feature highlights. Cat 9 LTE being the most important but there is also support for higher resolution screens and LPDDR4 memory.
Helio 20 MT6797 processors should start shipping to phone manufacturers in Summer with phones reaching stores with the chipset by late Autumn/Early Winter.
[ Source ]
I think the Helio X20 will be a little faster than the Snapdragon 810, but not significantly. Its main advantages will be efficiency and camera performance. I guess that’s what Mediatek want to focus on – and I like that. Top notch performance really isn’t what we need, especially when last generation Qualcomm chipsets still can’t be maxed out with todays applications and games.
No, it ill be a hell of a lot faster on the CPU side.
In theory A72 at 2.5GHz would be some 40% more or less faster than the Galaxy S6 and that one is plenty faster than the 810.
The SD620 with A72 at just 1.8GHz should easily beat the SD810. and more or less match the Exynos 7420.
If ARM meets it’s targets with A72.it will be pretty great.
That applies to chipsets that only use A72 cores. You forget that Mediatek does not use all those cores for a performance benefit only, but mainly due to efficiency reasons. The A72 cores won’t be used all the time. They will come to use if a complicated task needs to be done very quickly to be able to revert back to slower cores as fast as possible. By doing that you can increase energy efficiency significantly. The big.LITTLE architecture and also this new Mediatek tripple-core architecture never have been invented for a performance benefit. Energy efficiency always is priority no.1 with those. If you want to get maximum performance you would go for x cores of the same kind. That’s what Mediatek did with the MT6752 for example and it is the reason why it is so much more powerful than the Snapdragon 615.
That’s one weird way of looking at it and it’s wrong.
The small cores are used on tasks that don’t saturate them. So they don’t hit max load and they can deal with the task. There is no real loss in performance vs using the bigger core.When the small core gets a task that would saturate it so it would take longer, the load goes to the bigger ones.
Any loss in perf is mostly because of the time lost switching not because you use the small cores but ideally that would be negligible.In real life that does get a bit tricky and it’s not always handled well enough.
The goal for bigLITTLE is to save power without impacting performance not to sacrifice performance to save power.You actually can clock the big cores higher for short bursts because the small cores allow for a bit more TDP room.
BigLITTLE serves the same goal as dynamic voltage and frequency scaling where the core clocks and voltages are dialed down or up to adjust to the current need. BigLITTLE extends that and adds further power gains at a small die area cost by using cores that are best fit for the task instead of clocks that are best fit for the task.
Not to mention that SD810 also has little cores.
Now it is true that this chip has only 2 big cores but in mobile you don’t really see scenarios where 4 big cores would be pushed to high load at the same time and usually muti-core perf is TDP limited so you can’t really push 4 big cores at 100% for more than a few seconds anyway.
You simply didn’t get what I said. I never claimed there would a performance loss.
lol ok let me repeat what you said, just in case you decide to edit
“That applies to chipsets that only use A72 cores. You forget that
Mediatek does not use all those cores for a performance benefit only,
but mainly due to efficiency reasons. The A72 cores won’t be used all
the time. They will come to use if a complicated task needs to be done
very quickly to be able to revert back to slower cores as fast as
possible. By doing that you can increase energy efficiency
significantly. The big.LITTLE architecture and also this new Mediatek
tripple-core architecture never have been invented for a performance
benefit. Energy efficiency always is priority no.1 with those. If you
want to get maximum performance you would go for x cores of the same
kind. That’s what Mediatek did with the MT6752 for example and it is the
reason why it is so much more powerful than the Snapdragon 615.”
Now maybe you can clarify what you said.
But i did what i could to explain it in easy to understand terms so i’m done.
Oh well, is that so. Me too then. No need to edit anything. Please show me where I wrote something about a performance downgrade, because I can not find that part. Anyway, your explanation above is right, I perfectly know how those architectures are working. But again, you simply didn’t understand what I wrote, so your explanation doesn’t proof my words wrong at all and neither of our claims is wrong.
You claimed that the X20 will be “a little faster than the Snapdragon 810”, i replied that it will be a lot faster only for you to claim that “That applies to chipsets that only use A72 cores.”.
Let’s all pretend we’re computer engineers with a doctorate in microchip design and know what we’re talking about so we can look cool on the internet.
LOOOOOOOOOOOOOOOOOOOOOOOOOOOOL
I cant help laughing bro. What you have here is a mix of geeks, nerds and also computer engineers so yes,some people actually know what they are talking about. Not everyone here are just phone fans.
Some are more advanced than that 🙂
At least They are no hurting anyone.
Look at those who give medical advices. And there are lot of people taking those advices seriously, sometimes more than doctors.
Jesus, you really have no other problems, don’t you? 😀
So what is the definition of “faster” and a “a little faster”? Depends on the point of view obviously. So lets get things straight:
– The Snapdragon 810 is a big.LITTLE chipset with 4x A53 and 4x A57
– The MT6797 is a tripple cluster chipset with 4x A53, 4x A57 and 2x A72
– The basis of the MT6797 pretty much equals the Snapdragon 810 (except frequencies) but adds two A72 cores on top
– The A53 cores are the most power efficient cores but also the slowest. A57 is faster but needs more power. A72 is supposed to be a lot more faster but in reality probably not all that much more than A57 cores.
– The A53 cores of the MT6797 feature a lower clock speed than the ones on the Snapdragon 810. The A57 cores are similar plus the additional A72 cores result in a much higher “raw” processing power.
– Core Pilot 3.0 will manage the whole thing, making sure every core is used in the most efficient way possible which includes consideration of temperature, clock, voltage and core load while assigning tasks.
– The overall performance gain of this is about 15% going after Benchmark results (71k vs 61k in Antutu) which obviously don’t say much but are the only thing we can refer to right know.
So yes, the performance gain is there (I never denied that) but it isn’t anything groundbreaking. And I for myself never expected it to be groundbreaking, since (as I said before) big.LITTLE architectures and also Mediatek’s tri-cluster architecture (which is based on the idea behind big.LITTLE) never have been built for the sole purpose of getting the biggest performance boost possible. The main target is efficiency. Efficiency = ratio between power consumption and performance.
What I wanted to say with my claim “that applies to chipsets that only use A72 cores”: If you want to get the biggest performance gain possible, you will never go for such an architecture as a chip manufacturer since efficiency always means a certain degree of performance cut-down. The Allwinner A80T is a very good example for this. This chipset performs awful but becomes quite a power house as soon as you disable the big.LITTLE management and let the chipset do everything with its big cores and only use the little cores if the big cores are maxed out. Of course that also results in a cut-down on battery life of about 68% (personal experience on the Teclast P98 Air which I tested and developed a custom ROM for).
Essentially this discussion is pointless since we don’t need the performance a chip consisting only of A72 cores could potentially offer. Even the Snapdragon 801 as a last generation high-end chip can not be maxed out yet with normal applications / games.
Verdict: We will have some very exciting chipsets popping until 2016 and the Mediatek Helio X20 / MT6797 so far is my favorite among all of them since I believe it will be the most efficient chipset out there, and again, efficiency (in my eyes) is the most important thing right now.
I’m with you in this debate.
Just look at the amount of comments when it’s about SoCs.
I understand the passion behind it but the real deal here is power efficiency.
You can put the most powerful CPUs but we are still talking about raw power measured through benchmarks, a scenario that we won’t see in real life also because batteries won’t allow it.
Soon we might need more power and we can’t stop that process but right now the focus is efficiency until the battery technology will evolve.
Just to expand a little on what you said realjjj …
The 2x A72 cores are about as fast as 4x A57 cores anyways. According to this info from AnandTech: http://www.anandtech.com/show/8957/arm-announces-cortex-a72. So, performance-wise, the SD810 is not gonna come close to the Helio x20. And I love the fact that they have gone for a good albeit not super GPU this time. But I bet the T880 can handle any games you throw at it. ALso good thing MT is not using only 2 cores again but 4 cores this time which should add a whole lot more in GFX performance too.
Theoretically, 4x T880 is about 160% faster than 2x T760.
Big litle is ARMs bullshits & worst ever waist of silicon. Not to get me wrong I am not against micro multi purpose helping – offloading cores.
Truth is a little different…
Task migration is costy process & doing it 2-3 times is something that doesn’t pay of for small fast tasks as most are on smartphones. The A72’s will have healthy advantage in performance compared to A57’s but due to 15~20% bigger clock rate with same DTP. Everyone in the industry is pursuing a performance /cost metrics & they don’t give a shit about power efficiency or green computing for that matter. For example A72 @ 1GHz will consume about same amount of power (if not even less) as A53 @ 1.5GHz will performing similar. Magical barrier for efficient performing on silicon is around 1.0~1.1GHz, after that power consumption rises almost exponentially.
Cat 9 is pretty pointless for now , not much support for it. Apple for example has just cat 4 in it’s latest.
Maybe worth noting is the 20+20 carrier aggregation , don’t think MTK had carrier aggregation before this.
They are not going all out (no DDR4, somewhat limited video encoding, no 4k screen support) but maybe that’s good since it enables a better price.
The GPU should be ok but not much out there on it since it was announced at the same time as the A72. ARM claimed “The new Mali-T880 GPU delivers 1.8X the graphics performance of today’s Mali-T760 based devices”. So in theory it won’t match the Galaxy S6 but won’t be far behind.
It’s not really aimed at SD820 but it should be plenty cheaper so will enable better pricing.
Cat 9, because FOOCHER PROOFEENG.
I’m voting in favor of the Helio X20. Good thing to see that Mediatek is now taking the graphics serious in the Helio. The Mali T880 MP4 should be a sweet upgrade from the PowerVR G6200 rogue with cool visuals.
It’s probably be much cheaper as well. Gotta admire how far MediaTek has come in these past few years.
According to this info from AnandTech:
http://www.anandtech.com/show/8957/arm-announces-cortex-a72,
The Mali T880 is 1.8x faster than the T760 which makes it 80% faster overall within the same power envelope.
MT used only 2 cores in the T760 MP2 in MT6752 and it performed well. Which means that the T880 MP4 should be 160% faster theoretically.
Oh and just to point out one thing … the 2x A72 in the upcoming Helio x20 gives performance output of about a quad A57. Considering that the A57 is 1.9x faster than and A15 and the A72 is 3.5x faster than the A15 too. The A72 core is only 0.03x short of being 2x faster than the A57 (1.9 x 2 = 3.8) which is cool.
So basically, MT is giving us an SoC with 8x A53 and 4x A57. Now thats a real heavyweight SoC in the making coupled with the heavyweight T880 GPU IP.
I cant wait for this to surface in smartphones … haha.
In best case T880 will be 13% faster than T760 cluster /MHz.
Mali 760 mp8 190.4 gflop in exynos 7420 over power the adreno 430 288 core 324 gflop in benchmark , i think mali t880 mp4 700mhz equal to adreno 420
It’s MP 6 T760 in Exunos & compute theoretical performance GFlops doesn’t fit well in design flows. Adrenos have lots of those like for instance not begin capable to process any kind of texture surface transform as they should (you can see this in Antutu’s 3D test [lot’s of water]) & no suport half float precision operations (16 bit) that are important as they are faster & used by Android UI for rendering. MALI’s don’t have any significant designs flows. When it comes to drivers both are horrible (you can check this on Dolphine emulator forums).
Exynos 7420 t760 have 8 core, Where comes from your information??.
My opinion from benchmark results between exynos 7420 and sd810 on gsmarena gfx 2.7 t-rex 1080p off screen and gfx 3.0 Manhattan 1080p off screen and basemark x ,You should note Samsung Galaxy S6 has a QHD screen – 1,440 x 2,560px – which is around 80% more pixels than a 1080p screen, all sd810 devices in the market have 1080p screen antutu benchmark not give you real deference..
Anandtech review for exynos 5433 T760 mp 6 Give Different results between it and sd805 adreno 420 but accuracy benchmark give mali T760 mp 6 the lead and from t 628 mali gpu has better driver than any gpu on android platform. Especially after the latest updates of mali driver .
Ok it’s MP8. It’s a same story with gfx as with Antutu (3D) it’s only more representative in Antutu (refractions on water use texture surface transform).
Reed that review some time ago & it whose more about Samsung’s 20 nm proces, it’s probably best most comprehensive one in last two year’s or so.
About driver’s quality I already told you where to look! This is rather old article but nothing have changed much in a mean time.
Url:
https://dolphin-emu.org/blog/2013/09/26/dolphin-emulator-and-opengl-drivers-hall-fameshame/
Actually Power VR ones got worse & the rest is the same…
I agree with you but not totally mali gpu still best driver on android platform than adreno and power VR ,the comparison between t760 exynos 5433 and adreno 420 SD 805 give the mali gpu driver 50% performance over adreno gpu driver.
You obviously didn’t read linked article.
I read it but 1_ it’s old article abut two years ago 2_ dolphin emulator open GL specialist it’s not represent the total performance 3_ Dolphin emulator team said mali driver is not that bad 4 _ the article didn’t give the different by point , And the team able to solve one of the problem by themselves and send all the problems and bugs to ARM developer forums . I said the benchmark give 50% advantage to mali driver,that the different between bad and the horrible
Jes but freedreno works quite good & ARM did bad never tried to help development of open driver’s for MALI… The thing with Dolphine is that it’s an emulator & most advanced one on platform… The thing is when you emulate & use EGL 3.0 (3.1) every flow in driver’s really shows. ?
50% advantage in what?
Told you to look 3D mark results as relatively comprehensive ones
Ok let me invite you to this article http://www.anandtech.com/show/8718/the-samsung-galaxy-note-4-exynos-review/9
Read it and tell me what is the thing make t760 less computing power in GFX bench 3.0 ALU,alpha blending and fill rate tests give about same performance adreno 420 on 3D mark 1.2 unlimited, basemark X 1.1 and GFX bench 3.0 Manhattan and t-rex
Simply different architecture.
Adrenos can reach much more Gflops but that doesn’t translate good in real graphical performance.
When it comes to OCL & HSA implementation I believe ARM has an edge as the old consortium member.
MALI’s stil have significant advantage (when used) of half precision (16 bit) calculation mode that Adrenos lack.
It appears that parallelism inter connection joints on large single Adreno clusters is done badly but still this approach gives significant power savings compared to in this case 6 MALI separate clusters.
This is also relatively old article, about month afterwards Qualcomm catch up on driver overheads. Both companies still have not good enough driver divisions.
I’m not computer engineer can you explain the different in architecture and why adreno 420 better power saving
Every new Adreno is optimized one large single cluster on silicon. This approach is costs a lot but it gives best results as you see when you add more clusters they don’t scale ideally every other than first added 60 up to 70% of performance of first one. The gap between large cluster approach & couple of clusters is you see around 25% in performance /W.
About arhitekturu I really can’t explain it to you as Qualcomm removed all ATI’s documentation around two years ago. Person who probably would have know most about it that it’s not a direct employee of Qualcomm would probably be a Rob Clark.
Adreno 420? Forget it, Adreno 330 at most.
That’s if Mali t880 @850 Mhz performance 1.8 x t760 @ 650 Mhz that’s means 1.5 x in same Mhz (4×1.5=6 ) and we know ‘s Mali t760 mp6 in exynos 5433 give about the same performance of adreno 420 .
Mediatek and Kirin are using 4-core versions. Mali-T880 does about 28,9 GFlops per core, it means that 4-core version should do about 120GFlops (slightly below). Adreno 330 does 129,6-166,5GFlops (Snapdragon 800-801).
http://kyokojap.myweb.hinet.net/gpu_gflops/
You are right, but compute theoretical performance GFlops doesn’t fit well in design flows .visit this website http://www.anandtech.com/show/8718/the-samsung-galaxy-note-4-exynos-review/9 you will find exynos 5433 mali T760 mp 6 about the same performance of adreno 420 in sd805 , Mali 760 mp8 190.4 gflop in exynos 7420 over power the adreno 430 288 core 324 gflop in benchmark http://www.gsmarena.com/samsung_galaxy_s6-review-1227p5.php
Bear in mind that GFlops mean quite a lot under heavy loads. GFXBench benchmarks are pretty light, more powerful cores in Adreno GPUs are more future proof.
Are you kidding me GFX 3.0 manhattan is the most powerful 3D benchmark.you obviously didn’tsee all the article Mali equal to adreno in all benchmark heavy or light .
Most powerful among benchmarks but 3D games are more and more complicated.
I hope Mtk bring more power to their SOC in graphic department to give the consumer flagship performance, With low cost. And the most thing I worried about it the efficiency of mali gpu in the next exynos it may be use t880 mp8
It doesn’t look like it – future Mediateks are still going to have rubbish GPUs. Next Exynos – probably Mali-T880 MP8. Next Mediatek – Mali-T880 MP4.
T880 MP4 should have a lot more powerful than T760 MP2, not a mere 160%.
On the other hand A72 isn’t 1.9 times faster than A57. It’s 1.9 times power efficient at same performance level, if A57 manufactured 20nm while A72 manufactured 16nm FinFET. Some of it comes from that.
A72 roughly 22% consumes less power while delivering 5% more performance than A57 at same node if we believe Huawei.
A72 isn’t 1.9 faster in same mhz ,arm cortex a72 benchmark results from arm it 16-56 % faster depending on the different work load and 18-30% more efficient ,and about 50%multi threads and memory speed than Intel core M.
And on benchmark news the results of sd620 1.8 Ghz a72 geekbench single core performance abut 1512 point it more than both exynos 7420 and sd810 results.
PowerVR`s GPU`s are allot better than Mali, but ARM seams to be pushing them out of the game for some reason. The VR6200 in the 6595 is a great GPU, but lots of games have been having problems with PowerVR GPU`s recently, I think it`s a trick by ARM to push their own product Mali to developers with the promise of best optimization(and it really is). Remember hardware cannot perform much faster if the software is not there.
I think MALI driver’s are getting better wile imagination degraded it’s driver quality. I doubt their is anything else to it except possible bigger fees Imagination is demanding for their GPU’s. The last one is just a speculation. When I discussed with some of guys from Imagination tech about Warrior cores they where confident MTK will continue with their GPU part’s even I told this will happen.
The Mali T880 will compete with high end series 6xt or even series 7. The G6200 is a nice GPU indeed, but lower class.
I like it.
If Mediatek could be able to release it in short term it will kill SD810.
Mediatek has said that it’s shipping to device makers in Q4 so it’s not all that soon.
Hehe, that’s not likely. It’ll probably be out late as usual, like the MTK6595.
Mediatek cannot compete with Adreno in the GPU department, but will offer plenty of performance and being a newer product it will offer plenty in power efficiency. I expect this chip to wipe the floor with the S810, except in the GPU department, there Qualcomm will keep a 15-20% lead. Most people forget that DDR4 has higher latency than DDR3, so those extra Mhz won`t offer more performance, maybe 5% tops. Latency is what matters in RAM.
I just wish that Mediatek would come up with a 2xA72 and 4xA53(at around 1.8Ghz) and put a MP760MP4 in there and sell it as a mid-range chip around 250-300$ phones. That for me would be a killer chip.
I believe they will do mid-range chips with A72 cores, but probably not before 2016. I for myself am looking forward to see when the first tablets with their MT8173 will pop up and how it will perform. After all this appears to be the worlds first A72 chipset already available to manufacturers.
Who wants to compare Helios 20 vs S810 end of the year?!
Than you will have to compare Helios 20 vs S820!
where are phones with 6795? and why compering SD810 (which is out for few month) with 6797 while 6795 isn’t out yet. Does everyone forgot 6595? That supossed to be killer SoC too, where is now? just in MX4? Mediatek still need time to catch up SD in high end SoCs. And don’t forget Mediatek dosen’t share kernel sources
Lenovo, Coolpad, Siswoo and Zoppo all released devices with the 6595. I agree MTK hasn’t quite caught up to SD in terms of high end SoC’s but they are gaining a lot of ground with each new SoC they release.
Yes, but only Meizu worth mentioned because those phones didn’t sell really well.
The Vibe X2 sold like crazy in India.
Nice looking phone I have but one for my sister!
Nice looking phone I have buy one for my sister!
ooh t880mp4 almost equals adreno 330 from 2 year old s800!
Well done!
Why not just put mp16 or some top PowerVR? yes, it will be something like 10$ more expensive but it is worth it!
I guess you are confusing T880 with another GPU. It will be as fast as Adreno 430.
If Arm’s figures are to be believed then this would give the GPU around 170 GFlops of computer power which is around the same as the Adreno 420 found in the SD805. The Adreno 430 is supposed to have between 320-390 depending on clock speed so waaaaay behind the 810. However it seems that MT have finally woken up and realised they needed a decent GPU and I for one am happy with what they have gone for. Just stick this is a nice Vivo phone and my next phone choice will be simple! lol
Totally agree with you (as usual).
This GPU with this clock and number of cores is behind the adreno 430 ( pretty much behind) regardless other’s opinion since GFLOPS can’t lie.
But I’m really happy, it’s a good beginning in my opinion.
I think they couldn’t add more GPU cores due to lack of room but if this project of 10 cores fails (which I think it will) they might return to 8 cores plus a bigger GPU.
I think a lot of people forget that this GPU from MT is a quantum leap forward for them, as is the A72 cores. We’re talking about a company that’s always used the slowest cores (A7 & A53). They never used A57 or A15 cores in any mobile SoC and here they are readying a chip that is running Arm’s latest and greatest cores ahead of everyone else! This isn’t the MT of old and I’m really liking what I’m starting to see from them this year. Regarding what you said about the CPU cores, everyone ridiculed them when they first announced octacore chips and now everyone’s doing it. Lastly and this is important since it’s a lot of peoples arguments against 8 cores. Programs/apps don’t utilise them fully because until recently they didn’t exist so why would someone write a program for something that didn’t exist? As it gets more common I’m sure coding will improve to implement this fact and if 10 cores take off maybe the same can happen for that too.
I’m saying kudos to mediatek, I’m happy to see the path they are taking and how finally they are going to make true high end without forgetting of efficiency.
I didn’t mean on mocking them at all, I’m not confident about the 10 cores and I’ll be happy to be proved wrong. Even today, speaking on real usage, there’s no real big difference between 4 and 8 cores. That’s where my opinion came from. That’s all
I get what your saying and I know you weren’t mocking them. Your opinion of MT is the same as mine and I’m not convinced of 10 cores either, although it’s great bragging rights I guess and allows them to break 70000 on Antutu without using to many power hungry A72 cores.
I recommend you to have a look at the CorePilot datasheet on the Mediatek website. Applications nowadays indeed benefit from multiple cores even if they are not optimized. Of course they could benefit more if they were optimized, but modern CPU management software takes a lot of work away from app developers.
Yep, going to be interesting indeed. Especially when you think about the fact they managed to get out more of the slowest cores available (with the MT6732 / MT6752) than their competitor Qualcomm (comparing to the SD410 / SD615). Their CorePilot technology is one of their secrets to success, and if they manage it to handle the tri-cluster architecture properly and as efficient as it managed their previous 64-bit chipsets, we will see quite some dropped jawbones there.
That will probably never happen. Most of software that is capable of MP won’t use more than 2 cores. Even if they can use more cores performance scaling is far from linear. Tasks that are generally suitable for using many general purpose cores are done much more efficient on other forms of hardware (GPU’s, ASIC’s, DSP’s FPGA’s).
I personally think FPGA’s will be a real future of computing. Looks like Qualcomm understand that, Intel certainly did. It’s not like general purpose cores are going anywhere as they will always be needed it’s just that there won’t be much more of them from what we have today. Still I believe its possible to huck up more cores to work as ST based on simultaneous MP principal & based on hardware implementation of it. Sometimes like this would certainly have sense especially with small power efficient cores. We can see how complicated out of order architectures didn’t actually menage to deliver good performance per wat compared to simpler in order ones. It would be possible to menage SMP pretty efficiency on small cluster scale 2~4 cores & with in order cores benefits would be great. Unfortunately we are far from this getting a reality. Every other mumbo jumbo is just a plane marketing & before all big litle (two clusters). Again just to say 3 clusters is just utterly stupid.
Adreno 430 has 320-390 GFLOPS, and it’s beated by Mali-T760 MP8 which has mere 190 GFLOPS.
I missed that battle.
Which casino in Vegas did they fight?
Casino Anandtech.
How is it beaten? I read a comparison between the 2 GPU’s and it generally favoured the Adreno 430. The Mali beat it in some tests and lost out in others, there could be lots of reasons for this of course, mainly software optimisations though. Doesn’t matter how powerful a device is if the software isn’t up to par. Besides, what has the T760 got to do with this when we’re talking about the T880??
I gave T760 as example to emphasis the fact GFLOPS doesn’t mean always better performance.
According to ARM, T880 is around 80% faster than T760.
If T880 MP4 really is around 170 GFLOPS, and T760 MP8 is around 190 GFLOPS and if GFLOPS is really matter they should have more comparable to Adreno 330 which is not the case we know in fact.
http://images.anandtech.com/graphs/graph9146/73778.png
http://images.anandtech.com/graphs/graph9146/73780.png
Proof please
in terms of GFLOPS T880 mp4 is something about adreno330 not 430
Generally, 3dmark confirms such data, adreno330 devices are superior to all the mediateks in terms of graphic performance
Adreno 330 is up to 166 GFLOPS.
Adreno 430 is up to 389 GFLOPS.
Mali-T760 MP8 is about 190 GFLOPS.
So if we are comparing them with that aspect T760 MP8 should more close to Adreno 330 and crushed by Adreno 430, but in the reality T760 MP8 faster than Adreno 430 and crushing Adreno 330.
http://www.anandtech.com/show/9146/the-samsung-galaxy-s6-and-s6-edge-review/6
Yeah Adreno 330 is powerful than all Mediatek chips’ GPUs, but it’s irrelevant. They are weaker because Mediatek’s intend was make them cheap.
According to ARM, T880 will be 80% faster than T760.
Because it just doesn’t work this way, Qualcomm is paying a bunch for GPU development, why do you think their chips have stalled in CPU performance? Adreno as a company will always make the best GPU’s in the mobile segment and the Snaprdragon 800 will be a relevant chip for at least another 2 years. Mediatek know they can`t compete with that on the node they are using at the moment with offering Mali cores and PowerVR GPU`s are more expensive and have worse compatibility. Mediatek is aiming at the mid to mid-high segment where the good sales are at, Qualcomm and Samsung can battle it out on the top with GPU performance, but the fact remains that in performance terms MTK chipsets are solid and headsets with them offer pretty much the same performance in general for more than twice less of what the competition is asking.
Adrenos are actually based on old tech, they all have big problems when any kind of texture surface transform is involved, they lack half precision mode. Mali on the other hand don’t have any problems at all. ARM should probably build bigger cluster as a base instead of rising possible MP number.
But they perform when it matters and score big in tests, even the Adreno 305 can run almost any game flawlessly at 720p, which is basically a low-end GPU design. My Adreno 320 handles anything at 1080p without hiccups. Haven`t tried the Mali T760 yet, but I`m reading good things about it, if you can put more than 2 cores in there. Remember that no one besides Qualcomm know what actually is in the Adreno products, they give you only a few basic specs, but nothing about how the architecture works.
If you trow in a cup of water you will see how Adrenos perform. A305~306 performance in EGLS 3.0~+ is horrible. People don’t make games that will need more than usual gpu power as they would be able to sell them, point is to push that limit higher.
Hope for your sake you have heard for Rob Clark. Their whose some technical papers left of from ATI but Qualcomm made them disappear around two years ago. Architectural fundamentals remain the same never the less including the same problems.
What do you mean PowerVr has worse compatibility? They just choose Mali because maybe as you said it’s cheaper option. The Powervr goes very well in android.
& MTK again falls badly. Even I think two big cores are enough for real usage scenarios, putting second cluster of small cores is simply utterly stupid. The MALI T880 MP4 won’t be enough even for satisfying performance on FHD. If they cut off second A53 cluster (higher clocked one) & went with T880 MP6 this would be totally different story & it would rise price per SoC about 2$ making it a wonderful higher part mid range SoC to wish for. Seams they will never learn. Even S810 is a disaster comparing this to it is simply childish.
You are exaggerated too much. It will run smoothly on even QHD displays.
That will probably be like you are saying if you don’t expect to play more than candy crush saga on your device. On the other hand even synthetic GPU tests don’t mean much for current actual games we all want that future games match graphic lv of those tests. At the end even T760 MP12 in Galaxy S6 is not enough for QHD based on same expectations.
I get it now. S6 has T760 MP8 not MP12. And it’s not sweating running on QHD.
You know M1 Note has just T760 MP2 and FHD display. Nobody complains about it’s performance.
You are right about T760 MP8, my bad. But that doesn’t change the fact nor scores.
T760 MP2 is actually good enough but on a 720p display.
Their is nothing even remotely capable to deliver good performance for QHD nor it will be their for a long time. Even I hate Apple I must admit they are right when it comes to performance vs resolution metrics & I understand why they hesitated to go up from 720p for long time.
Personally I don’t get caught up in the whole MTK vs Qualcomm debates. I have used great SoC’s from both companies and have used bad SoC’s from both companies. I have also used some excellent Kirin and Exnyos chips. I just love that they all are pushing each other to innovate even more.
Looks like you missed all fun with Texas instruments and Broadcomm. You know how they say; only the good day young. ?
Well, Qualcomm has been the leader for a long time and Mediatek did “bad” chips for a long time. Now they suddenly went from zero to hero, launching one amazing chipset after another. Lots of people have problems accepting this since, well, we humans tend to develop certain feelings (call it being haters or fanboys). Anyway, nothing wrong doing some comparison and being proud for what they are doing. After all this means more intense competition, and competition is what drives the market.
I am aware of the history between the two I just don’t get caught up in choosing one side over the other.
I think the Helio X20 will be a little faster than the Snapdragon 810, but not significantly. Its main advantages will be efficiency and camera performance. I guess that’s what Mediatek want to focus on – and I like that. Top notch performance really isn’t what we need, especially when last generation Qualcomm chipsets still can’t be maxed out with todays applications and games.
No, it ill be a hell of a lot faster on the CPU side.
In theory A72 at 2.5GHz would be some 40% more or less faster than the Galaxy S6 and that one is plenty faster than the 810.
The SD620 with A72 at just 1.8GHz should easily beat the SD810. and more or less match the Exynos 7420.
If ARM meets it’s targets with A72.it will be pretty great.
That applies to chipsets that only use A72 cores. You forget that Mediatek does not use all those cores for a performance benefit only, but mainly due to efficiency reasons. The A72 cores won’t be used all the time. They will come to use if a complicated task needs to be done very quickly to be able to revert back to slower cores as fast as possible. By doing that you can increase energy efficiency significantly. The big.LITTLE architecture and also this new Mediatek tripple-core architecture never have been invented for a performance benefit. Energy efficiency always is priority no.1 with those. If you want to get maximum performance you would go for x cores of the same kind. That’s what Mediatek did with the MT6752 for example and it is the reason why it is so much more powerful than the Snapdragon 615.
That’s one weird way of looking at it and it’s wrong.
The small cores are used on tasks that don’t saturate them. So they don’t hit max load and they can deal with the task. There is no real loss in performance vs using the bigger core.When the small core gets a task that would saturate it so it would take longer, the load goes to the bigger ones.
Any loss in perf is mostly because of the time lost switching not because you use the small cores but ideally that would be negligible.In real life that does get a bit tricky and it’s not always handled well enough.
The goal for bigLITTLE is to save power without impacting performance not to sacrifice performance to save power.You actually can clock the big cores higher for short bursts because the small cores allow for a bit more TDP room.
BigLITTLE serves the same goal as dynamic voltage and frequency scaling where the core clocks and voltages are dialed down or up to adjust to the current need. BigLITTLE extends that and adds further power gains at a small die area cost by using cores that are best fit for the task instead of clocks that are best fit for the task.
Not to mention that SD810 also has little cores.
Now it is true that this chip has only 2 big cores but in mobile you don’t really see scenarios where 4 big cores would be pushed to high load at the same time and usually muti-core perf is TDP limited so you can’t really push 4 big cores at 100% for more than a few seconds anyway.
You simply didn’t get what I said. I never claimed there would a performance loss.
Just to expand a little on what you said realjjj …
The 2x A72 cores are about as fast as 4x A57 cores anyways. According to this info from AnandTech: http://www.anandtech.com/show/8957/arm-announces-cortex-a72. So, performance-wise, the SD810 is not gonna come close to the Helio x20. And I love the fact that they have gone for a good albeit not super GPU this time. But I bet the T880 can handle any games you throw at it. ALso good thing MT is not using only 2 cores again but 4 cores this time which should add a whole lot more in GFX performance too.
Theoretically, 4x T880 is about 160% faster than 2x T760.
lol ok let me repeat what you said, just in case you decide to edit
“That applies to chipsets that only use A72 cores. You forget that
Mediatek does not use all those cores for a performance benefit only,
but mainly due to efficiency reasons. The A72 cores won’t be used all
the time. They will come to use if a complicated task needs to be done
very quickly to be able to revert back to slower cores as fast as
possible. By doing that you can increase energy efficiency
significantly. The big.LITTLE architecture and also this new Mediatek
tripple-core architecture never have been invented for a performance
benefit. Energy efficiency always is priority no.1 with those. If you
want to get maximum performance you would go for x cores of the same
kind. That’s what Mediatek did with the MT6752 for example and it is the
reason why it is so much more powerful than the Snapdragon 615.”
Now maybe you can clarify what you said.
But i did what i could to explain it in easy to understand terms so i’m done.
Oh well, is that so. Me too then. No need to edit anything. Please show me where I wrote something about a performance downgrade, because I can not find that part. Anyway, your explanation above is right, I perfectly know how those architectures are working. But again, you simply didn’t understand what I wrote, so your explanation doesn’t proof my words wrong at all and neither of our claims is wrong.
You claimed that the X20 will be “a little faster than the Snapdragon 810”, i replied that it will be a lot faster only for you to claim that “That applies to chipsets that only use A72 cores.”.
Let’s all pretend we’re computer engineers with a doctorate in microchip design and know what we’re talking about so we can look cool on the internet.
Jesus, you really have no other problems, don’t you? 😀
So what is the definition of “faster” and a “a little faster”? Depends on the point of view obviously. So lets get things straight:
– The Snapdragon 810 is a big.LITTLE chipset with 4x A53 and 4x A57
– The MT6797 is a tripple cluster chipset with 4x A53, 4x A57 and 2x A72
– The basis of the MT6797 pretty much equals the Snapdragon 810 (except frequencies) but adds two A72 cores on top
– The A53 cores are the most power efficient cores but also the slowest. A57 is faster but needs more power. A72 is supposed to be a lot more faster but in reality probably not all that much more than A57 cores.
– The A53 cores of the MT6797 feature a lower clock speed than the ones on the Snapdragon 810. The A57 cores are similar plus the additional A72 cores result in a much higher “raw” processing power.
– Core Pilot 3.0 will manage the whole thing, making sure every core is used in the most efficient way possible which includes consideration of temperature, clock, voltage and core load while assigning tasks.
– The overall performance gain of this is about 15% going after Benchmark results (71k vs 61k in Antutu) which obviously don’t say much but are the only thing we can refer to right know.
So yes, the performance gain is there (I never denied that) but it isn’t anything groundbreaking. And I for myself never expected it to be groundbreaking, since (as I said before) big.LITTLE architectures and also Mediatek’s tri-cluster architecture (which is based on the idea behind big.LITTLE) never have been built for the sole purpose of getting the biggest performance boost possible. The main target is efficiency. Efficiency = ratio between power consumption and performance.
What I wanted to say with my claim “that applies to chipsets that only use A72 cores”: If you want to get the biggest performance gain possible, you will never go for such an architecture as a chip manufacturer since efficiency always means a certain degree of performance cut-down. The Allwinner A80T is a very good example for this. This chipset performs awful but becomes quite a power house as soon as you disable the big.LITTLE management and let the chipset do everything with its big cores and only use the little cores if the big cores are maxed out. Of course that also results in a cut-down on battery life of about 68% (personal experience on the Teclast P98 Air which I tested and developed a custom ROM for).
Essentially this discussion is pointless since we don’t need the performance a chip consisting only of A72 cores could potentially offer. Even the Snapdragon 801 as a last generation high-end chip can not be maxed out yet with normal applications / games.
Verdict: We will have some very exciting chipsets popping until 2016 and the Mediatek Helio X20 / MT6797 so far is my favorite among all of them since I believe it will be the most efficient chipset out there, and again, efficiency (in my eyes) is the most important thing right now.
LOOOOOOOOOOOOOOOOOOOOOOOOOOOOL
I cant help laughing bro. What you have here is a mix of geeks, nerds and also computer engineers so yes,some people actually know what they are talking about. Not everyone here are just phone fans.
Some are more advanced than that 🙂
At least They are no hurting anyone.
Look at those who give medical advices. And there are lot of people taking those advices seriously, sometimes more than doctors.
I’m with you in this debate.
Just look at the amount of comments when it’s about SoCs.
I understand the passion behind it but the real deal here is power efficiency.
You can put the most powerful CPUs but we are still talking about raw power measured through benchmarks, a scenario that we won’t see in real life also because batteries won’t allow it.
Soon we might need more power and we can’t stop that process but right now the focus is efficiency until the battery technology will evolve.
Truth is a little different…
Task migration is costy process & doing it 2-3 times is something that doesn’t pay of for small fast tasks as most are on smartphones. The A72’s will have healthy advantage in performance compared to A57’s but due to 15~20% bigger clock rate with same DTP. Everyone in the industry is pursuing a performance /cost metrics & they don’t give a shit about power efficiency or green computing for that matter. For example A72 @ 1GHz will consume about same amount of power (if not even less) as A53 @ 1.5GHz will performing similar. Magical barrier for efficient performing on silicon is around 1.0~1.1GHz, after that power consumption rises almost exponentially.
Big litle is ARMs bullshits & worst ever waist of silicon. Not to get me wrong I am not against micro multi purpose helping – offloading cores.
Cat 9 is pretty pointless for now , not much support for it. Apple for example has just cat 4 in it’s latest.
Maybe worth noting is the 20+20 carrier aggregation , don’t think MTK had carrier aggregation before this.
They are not going all out (no DDR4, somewhat limited video encoding, no 4k screen support) but maybe that’s good since it enables a better price.
The GPU should be ok but not much out there on it since it was announced at the same time as the A72. ARM claimed “The new Mali-T880 GPU delivers 1.8X the graphics performance of today’s Mali-T760 based devices”. So in theory it won’t match the Galaxy S6 but won’t be far behind.
It’s not really aimed at SD820 but it should be plenty cheaper so will enable better pricing.
Cat 9, because FOOCHER PROOFEENG.
I’m voting in favor of the Helio X20. Good thing to see that Mediatek is now taking the graphics serious in the Helio. The Mali T880 MP4 should be a sweet upgrade from the PowerVR G6200 rogue with cool visuals.
It’s probably be much cheaper as well. Gotta admire how far MediaTek has come in these past few years.
PowerVR`s GPU`s are allot better than Mali, but ARM seams to be pushing them out of the game for some reason. The VR6200 in the 6595 is a great GPU, but lots of games have been having problems with PowerVR GPU`s recently, I think it`s a trick by ARM to push their own product Mali to developers with the promise of best optimization(and it really is). Remember hardware cannot perform much faster if the software is not there.
According to this info from AnandTech:
http://www.anandtech.com/show/8957/arm-announces-cortex-a72,
The Mali T880 is 1.8x faster than the T760 which makes it 80% faster overall within the same power envelope.
MT used only 2 cores in the T760 MP2 in MT6752 and it performed well. Which means that the T880 MP4 should be 160% faster theoretically.
Oh and just to point out one thing … the 2x A72 in the upcoming Helio x20 gives performance output of about a quad A57. Considering that the A57 is 1.9x faster than and A15 and the A72 is 3.5x faster than the A15 too. The A72 core is only 0.03x short of being 2x faster than the A57 (1.9 x 2 = 3.8) which is cool.
So basically, MT is giving us an SoC with 8x A53 and 4x A57. Now thats a real heavyweight SoC in the making coupled with the heavyweight T880 GPU IP.
I cant wait for this to surface in smartphones … haha.
I think MALI driver’s are getting better wile imagination degraded it’s driver quality. I doubt their is anything else to it except possible bigger fees Imagination is demanding for their GPU’s. The last one is just a speculation. When I discussed with some of guys from Imagination tech about Warrior cores they where confident MTK will continue with their GPU part’s even I told this will happen.
In best case T880 will be 13% faster than T760 cluster /MHz.
T880 MP4 should have a lot more powerful than T760 MP2, not a mere 160%.
On the other hand A72 isn’t 1.9 times faster than A57. It’s 1.9 times power efficient at same performance level, if A57 manufactured 20nm while A72 manufactured 16nm FinFET. Some of it comes from that.
A72 roughly 22% consumes less power while delivering 5% more performance than A57 at same node if we believe Huawei.
The Mali T880 will compete with high end series 6xt or even series 7. The G6200 is a nice GPU indeed, but lower class.
Mali 760 mp8 190.4 gflop in exynos 7420 over power the adreno 430 288 core 324 gflop in benchmark , i think mali t880 mp4 700mhz equal to adreno 420
It’s MP 6 T760 in Exunos & compute theoretical performance GFlops doesn’t fit well in design flows. Adrenos have lots of those like for instance not begin capable to process any kind of texture surface transform as they should (you can see this in Antutu’s 3D test [lot’s of water]) & no suport half float precision operations (16 bit) that are important as they are faster & used by Android UI for rendering. MALI’s don’t have any significant designs flows. When it comes to drivers both are horrible (you can check this on Dolphine emulator forums).
Exynos 7420 t760 have 8 core, Where comes from your information??.
My opinion from benchmark results between exynos 7420 and sd810 on gsmarena gfx 2.7 t-rex 1080p off screen and gfx 3.0 Manhattan 1080p off screen and basemark x ,You should note Samsung Galaxy S6 has a QHD screen – 1,440 x 2,560px – which is around 80% more pixels than a 1080p screen, all sd810 devices in the market have 1080p screen antutu benchmark not give you real deference..
Anandtech review for exynos 5433 T760 mp 6 Give Different results between it and sd805 adreno 420 but accuracy benchmark give mali T760 mp 6 the lead and from t 628 mali gpu has better driver than any gpu on android platform. Especially after the latest updates of mali driver .
A72 isn’t 1.9 faster in same mhz ,arm cortex a72 benchmark results from arm it 16-56 % faster depending on the different work load and 18-30% more efficient ,and about 50%multi threads and memory speed than Intel core M.
And on benchmark news the results of sd620 1.8 Ghz a72 geekbench single core performance abut 1512 point it more than both exynos 7420 and sd810 results.
Ok it’s MP8. It’s a same story with gfx as with Antutu (3D) it’s only more representative in Antutu (refractions on water use texture surface transform).
Reed that review some time ago & it whose more about Samsung’s 20 nm proces, it’s probably best most comprehensive one in last two year’s or so.
About driver’s quality I already told you where to look! This is rather old article but nothing have changed much in a mean time.
Url:
https://dolphin-emu.org/blog/2013/09/26/dolphin-emulator-and-opengl-drivers-hall-fameshame/
Actually Power VR ones got worse & the rest is the same…
I agree with you but not totally mali gpu still best driver on android platform than adreno and power VR ,the comparison between t760 exynos 5433 and adreno 420 SD 805 give the mali gpu driver 50% performance over adreno gpu driver.
You obviously didn’t read linked article.
I read it but 1_ it’s old article abut two years ago 2_ dolphin emulator open GL specialist it’s not represent the total performance 3_ Dolphin emulator team said mali driver is not that bad 4 _ the article didn’t give the different by point , And the team able to solve one of the problem by themselves and send all the problems and bugs to ARM developer forums . I said the benchmark give 50% advantage to mali driver,that the different between bad and the horrible
Jes but freedreno works quite good & ARM did bad never tried to help development of open driver’s for MALI… The thing with Dolphine is that it’s an emulator & most advanced one on platform… The thing is when you emulate & use EGL 3.0 (3.1) every flow in driver’s really shows. 😉
50% advantage in what?
Told you to look 3D mark results as relatively comprehensive ones
Ok let me invite you to this article http://www.anandtech.com/show/8718/the-samsung-galaxy-note-4-exynos-review/9
Read it and tell me what is the thing make t760 less computing power in GFX bench 3.0 ALU,alpha blending and fill rate tests give about same performance adreno 420 on 3D mark 1.2 unlimited, basemark X 1.1 and GFX bench 3.0 Manhattan and t-rex
Simply different architecture.
Adrenos can reach much more Gflops but that doesn’t translate good in real graphical performance.
When it comes to OCL & HSA implementation I believe ARM has an edge as the old consortium member.
MALI’s stil have significant advantage (when used) of half precision (16 bit) calculation mode that Adrenos lack.
It appears that parallelism inter connection joints on large single Adreno clusters is done badly but still this approach gives significant power savings compared to in this case 6 MALI separate clusters.
This is also relatively old article, about month afterwards Qualcomm catch up on driver overheads. Both companies still have not good enough driver divisions.
I’m not computer engineer can you explain the different in architecture and why adreno 420 better power saving
Every new Adreno is optimized one large single cluster on silicon. This approach is costs a lot but it gives best results as you see when you add more clusters they don’t scale ideally every other than first added 60 up to 70% of performance of first one. The gap between large cluster approach & couple of clusters is you see around 25% in performance /W.
About arhitekturu I really can’t explain it to you as Qualcomm removed all ATI’s documentation around two years ago. Person who probably would have know most about it that it’s not a direct employee of Qualcomm would probably be a Rob Clark.
Adreno 420? Forget it, Adreno 330 at most.
That’s if Mali t880 @850 Mhz performance 1.8 x t760 @ 650 Mhz that’s means 1.5 x in same Mhz (4×1.5=6 ) and we know ‘s Mali t760 mp6 in exynos 5433 give about the same performance of adreno 420 .
Mediatek and Kirin are using 4-core versions. Mali-T880 does about 28,9 GFlops per core, it means that 4-core version should do about 120GFlops (slightly below). Adreno 330 does 129,6-166,5GFlops (Snapdragon 800-801).
http://kyokojap.myweb.hinet.net/gpu_gflops/
You are right, but compute theoretical performance GFlops doesn’t fit well in design flows .visit this website http://www.anandtech.com/show/8718/the-samsung-galaxy-note-4-exynos-review/9 you will find exynos 5433 mali T760 mp 6 about the same performance of adreno 420 in sd805 , Mali 760 mp8 190.4 gflop in exynos 7420 over power the adreno 430 288 core 324 gflop in benchmark http://www.gsmarena.com/samsung_galaxy_s6-review-1227p5.php
Bear in mind that GFlops mean quite a lot under heavy loads. GFXBench benchmarks are pretty light, more powerful cores in Adreno GPUs are more future proof.
Are you kidding me GFX 3.0 manhattan is the most powerful 3D benchmark.you obviously didn’tsee all the article Mali equal to adreno in all benchmark heavy or light .
Most powerful among benchmarks but 3D games are more and more complicated.
I hope Mtk bring more power to their SOC in graphic department to give the consumer flagship performance, With low cost. And the most thing I worried about it the efficiency of mali gpu in the next exynos it may be use t880 mp8
It doesn’t look like it – future Mediateks are still going to have rubbish GPUs. Next Exynos – probably Mali-T880 MP8. Next Mediatek – Mali-T880 MP4.
I like it.
If Mediatek could be able to release it in short term it will kill SD810.
Mediatek has said that it’s shipping to device makers in Q4 so it’s not all that soon.
Hehe, that’s not likely. It’ll probably be out late as usual, like the MTK6595.
Mediatek cannot compete with Adreno in the GPU department, but will offer plenty of performance and being a newer product it will offer plenty in power efficiency. I expect this chip to wipe the floor with the S810, except in the GPU department, there Qualcomm will keep a 15-20% lead. Most people forget that DDR4 has higher latency than DDR3, so those extra Mhz won`t offer more performance, maybe 5% tops. Latency is what matters in RAM.
I just wish that Mediatek would come up with a 2xA72 and 4xA53(at around 1.8Ghz) and put a MP760MP4 in there and sell it as a mid-range chip around 250-300$ phones. That for me would be a killer chip.
I believe they will do mid-range chips with A72 cores, but probably not before 2016. I for myself am looking forward to see when the first tablets with their MT8173 will pop up and how it will perform. After all this appears to be the worlds first A72 chipset already available to manufacturers.
this is the second place that I~ve seen say displays limited to 1600p instead of 4k. the mali gpu they are using supports 4k displays
Who wants to compare Helios 20 vs S810 end of the year?!
Than you will have to compare Helios 20 vs S820!
where are phones with 6795? and why compering SD810 (which is out for few month) with 6797 while 6795 isn’t out yet. Does everyone forgot 6595? That supossed to be killer SoC too, where is now? just in MX4? Mediatek still need time to catch up SD in high end SoCs. And don’t forget Mediatek dosen’t share kernel sources
Lenovo, Coolpad, Siswoo and Zoppo all released devices with the 6595. I agree MTK hasn’t quite caught up to SD in terms of high end SoC’s but they are gaining a lot of ground with each new SoC they release.
Yes, but only Meizu worth mentioned because those phones didn’t sell really well.
The Vibe X2 sold like crazy in India.
Nice looking phone I have buy one for my sister!
ooh t880mp4 almost equals adreno 330 from 2 year old s800!
Well done!
Why not just put mp16 or some top PowerVR? yes, it will be something like 10$ more expensive but it is worth it!
I guess you are confusing T880 with another GPU. It will be as fast as Adreno 430.
If Arm’s figures are to be believed then this would give the GPU around 170 GFlops of computer power which is around the same as the Adreno 420 found in the SD805. The Adreno 430 is supposed to have between 320-390 depending on clock speed so waaaaay behind the 810. However it seems that MT have finally woken up and realised they needed a decent GPU and I for one am happy with what they have gone for. Just stick this is a nice Vivo phone and my next phone choice will be simple! lol
Because it just doesn’t work this way, Qualcomm is paying a bunch for GPU development, why do you think their chips have stalled in CPU performance? Adreno as a company will always make the best GPU’s in the mobile segment and the Snaprdragon 800 will be a relevant chip for at least another 2 years. Mediatek know they can`t compete with that on the node they are using at the moment with offering Mali cores and PowerVR GPU`s are more expensive and have worse compatibility. Mediatek is aiming at the mid to mid-high segment where the good sales are at, Qualcomm and Samsung can battle it out on the top with GPU performance, but the fact remains that in performance terms MTK chipsets are solid and headsets with them offer pretty much the same performance in general for more than twice less of what the competition is asking.
Adrenos are actually based on old tech, they all have big problems when any kind of texture surface transform is involved, they lack half precision mode. Mali on the other hand don’t have any problems at all. ARM should probably build bigger cluster as a base instead of rising possible MP number.
Totally agree with you (as usual).
This GPU with this clock and number of cores is behind the adreno 430 ( pretty much behind) regardless other’s opinion since GFLOPS can’t lie.
But I’m really happy, it’s a good beginning in my opinion.
I think they couldn’t add more GPU cores due to lack of room but if this project of 10 cores fails (which I think it will) they might return to 8 cores plus a bigger GPU.
I think a lot of people forget that this GPU from MT is a quantum leap forward for them, as is the A72 cores. We’re talking about a company that’s always used the slowest cores (A7 & A53). They never used A57 or A15 cores in any mobile SoC and here they are readying a chip that is running Arm’s latest and greatest cores ahead of everyone else! This isn’t the MT of old and I’m really liking what I’m starting to see from them this year. Regarding what you said about the CPU cores, everyone ridiculed them when they first announced octacore chips and now everyone’s doing it. Lastly and this is important since it’s a lot of peoples arguments against 8 cores. Programs/apps don’t utilise them fully because until recently they didn’t exist so why would someone write a program for something that didn’t exist? As it gets more common I’m sure coding will improve to implement this fact and if 10 cores take off maybe the same can happen for that too.
I’m saying kudos to mediatek, I’m happy to see the path they are taking and how finally they are going to make true high end without forgetting of efficiency.
I didn’t mean on mocking them at all, I’m not confident about the 10 cores and I’ll be happy to be proved wrong. Even today, speaking on real usage, there’s no real big difference between 4 and 8 cores. That’s where my opinion came from. That’s all
I get what your saying and I know you weren’t mocking them. Your opinion of MT is the same as mine and I’m not convinced of 10 cores either, although it’s great bragging rights I guess and allows them to break 70000 on Antutu without using to many power hungry A72 cores.
Yep, going to be interesting indeed. Especially when you think about the fact they managed to get out more of the slowest cores available (with the MT6732 / MT6752) than their competitor Qualcomm (comparing to the SD410 / SD615). Their CorePilot technology is one of their secrets to success, and if they manage it to handle the tri-cluster architecture properly and as efficient as it managed their previous 64-bit chipsets, we will see quite some dropped jawbones there.
I recommend you to have a look at the CorePilot datasheet on the Mediatek website. Applications nowadays indeed benefit from multiple cores even if they are not optimized. Of course they could benefit more if they were optimized, but modern CPU management software takes a lot of work away from app developers.
Adreno 430 has 320-390 GFLOPS, and it’s beated by Mali-T760 MP8 which has mere 190 GFLOPS.
I missed that battle.
Which casino in Vegas did they fight?
Casino Anandtech.
How is it beaten? I read a comparison between the 2 GPU’s and it generally favoured the Adreno 430. The Mali beat it in some tests and lost out in others, there could be lots of reasons for this of course, mainly software optimisations though. Doesn’t matter how powerful a device is if the software isn’t up to par. Besides, what has the T760 got to do with this when we’re talking about the T880??
I gave T760 as example to emphasis the fact GFLOPS doesn’t mean always better performance.
According to ARM, T880 is around 80% faster than T760.
If T880 MP4 really is around 170 GFLOPS, and T760 MP8 is around 190 GFLOPS and if GFLOPS is really matter they should have more comparable to Adreno 330 which is not the case we know in fact.
But they perform when it matters and score big in tests, even the Adreno 305 can run almost any game flawlessly at 720p, which is basically a low-end GPU design. My Adreno 320 handles anything at 1080p without hiccups. Haven`t tried the Mali T760 yet, but I`m reading good things about it, if you can put more than 2 cores in there. Remember that no one besides Qualcomm know what actually is in the Adreno products, they give you only a few basic specs, but nothing about how the architecture works.
If you trow in a cup of water you will see how Adrenos perform. A305~306 performance in EGLS 3.0~+ is horrible. People don’t make games that will need more than usual gpu power as they would be able to sell them, point is to push that limit higher.
Proof please
in terms of GFLOPS T880 mp4 is something about adreno330 not 430
Generally, 3dmark confirms such data, adreno330 devices are superior to all the mediateks in terms of graphic performance
Adreno 330 is up to 166 GFLOPS.
Adreno 430 is up to 389 GFLOPS.
Mali-T760 MP8 is about 190 GFLOPS.
So if we are comparing them with that aspect T760 MP8 should more close to Adreno 330 and crushed by Adreno 430, but in the reality T760 MP8 faster than Adreno 430 and crushing Adreno 330.
http://www.anandtech.com/show/9146/the-samsung-galaxy-s6-and-s6-edge-review/6
Yeah Adreno 330 is powerful than all Mediatek chips’ GPUs, but it’s irrelevant. They are weaker because Mediatek’s intend was make them cheap.
According to ARM, T880 will be 80% faster than T760.
Hope for your sake you have heard for Rob Clark. Their whose some technical papers left of from ATI but Qualcomm made them disappear around two years ago. Architectural fundamentals remain the same never the less including the same problems.
That will probably never happen. Most of software that is capable of MP won’t use more than 2 cores. Even if they can use more cores performance scaling is far from linear. Tasks that are generally suitable for using many general purpose cores are done much more efficient on other forms of hardware (GPU’s, ASIC’s, DSP’s FPGA’s).
I personally think FPGA’s will be a real future of computing. Looks like Qualcomm understand that, Intel certainly did. It’s not like general purpose cores are going anywhere as they will always be needed it’s just that there won’t be much more of them from what we have today. Still I believe its possible to huck up more cores to work as ST based on simultaneous MP principal & based on hardware implementation of it. Sometimes like this would certainly have sense especially with small power efficient cores. We can see how complicated out of order architectures didn’t actually menage to deliver good performance per wat compared to simpler in order ones. It would be possible to menage SMP pretty efficiency on small cluster scale 2~4 cores & with in order cores benefits would be great. Unfortunately we are far from this getting a reality. Every other mumbo jumbo is just a plane marketing & before all big litle (two clusters). Again just to say 3 clusters is just utterly stupid.
What do you mean PowerVr has worse compatibility? They just choose Mali because maybe as you said it’s cheaper option. The Powervr goes very well in android.
Summer, winter, autumn we don’t have that in Africa be specific with period please
& MTK again falls badly. Even I think two big cores are enough for real usage scenarios, putting second cluster of small cores is simply utterly stupid. The MALI T880 MP4 won’t be enough even for satisfying performance on FHD. If they cut off second A53 cluster (higher clocked one) & went with T880 MP6 this would be totally different story & it would rise price per SoC about 2$ making it a wonderful higher part mid range SoC to wish for. Seams they will never learn. Even S810 is a disaster comparing this to it is simply childish.
You are exaggerated too much. It will run smoothly on even QHD displays.
That will probably be like you are saying if you don’t expect to play more than candy crush saga on your device. On the other hand even synthetic GPU tests don’t mean much for current actual games we all want that future games match graphic lv of those tests. At the end even T760 MP12 in Galaxy S6 is not enough for QHD based on same expectations.
I get it now. S6 has T760 MP8 not MP12. And it’s not sweating running on QHD.
You know M1 Note has just T760 MP2 and FHD display. Nobody complains about it’s performance.
You are right about T760 MP8, my bad. But that doesn’t change the fact nor scores.
T760 MP2 is actually good enough but on a 720p display.
Their is nothing even remotely capable to deliver good performance for QHD nor it will be their for a long time. Even I hate Apple I must admit they are right when it comes to performance vs resolution metrics & I understand why they hesitated to go up from 720p for long time.
Personally I don’t get caught up in the whole MTK vs Qualcomm debates. I have used great SoC’s from both companies and have used bad SoC’s from both companies. I have also used some excellent Kirin and Exnyos chips. I just love that they all are pushing each other to innovate even more.
Looks like you missed all fun with Texas instruments and Broadcomm. You know how they say; only the good day young. 😆
Well, Qualcomm has been the leader for a long time and Mediatek did “bad” chips for a long time. Now they suddenly went from zero to hero, launching one amazing chipset after another. Lots of people have problems accepting this since, well, we humans tend to develop certain feelings (call it being haters or fanboys). Anyway, nothing wrong doing some comparison and being proud for what they are doing. After all this means more intense competition, and competition is what drives the market.
I am aware of the history between the two I just don’t get caught up in choosing one side over the other.
where are phones with 6795 . . . ? and why we are comparing SD810 (which is out for few month) with 6797 (which would be out at the year end) while 6795 isn’t out yet. 6797 should be compared with SD820(it would be like with like comparison). Does everyone forgot 6595? it was supposed to be killer SoC too, where is now? just in MX4? Mediatek still need time to catch up SD in high end SoCs. And don’t forget Mediatek dosen’t share kernel sources
Yes that’s Mediatek!
6595 arrived too late for 64 bit lollipop so most manufacturers ignored & jumped to 6752 as releasing 32 bit phone dominated by 64 bit competitors would look foolish even though we still have little to show for it.
this is the second place that I~ve seen say displays limited to 1600p instead of 4k. the mali gpu they are using supports 4k displays
Summer, winter, autumn we don’t have that in Africa be specific with period please
Heiii…..
Recent is lollipop era, and it optimize for 64bit environment. Mediatek play very well in 64bit soc, and quallcomm is very terrible.
Quallcom too lazy to develop 64bit processor, krait is mature and has a limit to 805 (it just trick to increase clock speed).
About SD810 vs mtk6797, they are equal to compare, bcoze they produce in same trchnology node, 20 nm. Just who is the winner.
And dont forget about Screen on time, produce heating in long used
I use mi3 and tcl m2u, so far i satisfi tcl , mi3 is very sucksuck.
where are phones with 6795 . . . ? and why we are comparing SD810 (which is out for few month) with 6797 (which would be out at the year end) while 6795 isn’t out yet. 6797 should be compared with SD820(it would be like with like comparison). Does everyone forgot 6595? it was supposed to be killer SoC too, where is now? just in MX4? Mediatek still need time to catch up SD in high end SoCs. And don’t forget Mediatek dosen’t share kernel sources
Yes that’s Mediatek!
6595 arrived too late for 64 bit lollipop so most manufacturers ignored & jumped to 6752 as releasing 32 bit phone dominated by 64 bit competitors would look foolish even though we still have little to show for it.
Heiii…..
Recent is lollipop era, and it optimize for 64bit environment. Mediatek play very well in 64bit soc, and quallcomm is very terrible.
Quallcom too lazy to develop 64bit processor, krait is mature and has a limit to 805 (it just trick to increase clock speed).
About SD810 vs mtk6797, they are equal to compare, bcoze they produce in same trchnology node, 20 nm. Just who is the winner.
And dont forget about Screen on time, produce heating in long used
I use mi3 and tcl m2u, so far i satisfi tcl , mi3 is very sucksuck.
Nobody is talking about the phones where this Helios will be integrated into? Do we already know, because I am for one very interested in the capabilities they are promising: slow motion recording (like iPhone finally?) , native3d2.0, etc.
So : which phones are coming out with MT6797 ? 🙂
Someone know what is “Native 3D 2.0”??
Someone know what is “Native 3D 2.0”??
Quite irrelevant comparison, when the mediatek soc will be released, the competition will be snapdragon 815 and 820
Quite irrelevant comparison, when the mediatek soc will be released, the competition will be snapdragon 815 and 820
I would rather wait for the 16nm A72.
I would rather wait for the 16nm A72.
so in conclusion is it better to buy the one with the mediatek or snapdragon ?