ZolaIII, 11 May 2018Idle is better on a MT8176 because its not a three cluster.
Most software today (including on... moreI don't think it comes from tri-cluster...scheduler is able to run a single A53 from tiny cluster...
In my opinion problem comes from modem (and maybe from the way the Rom use it)
In some places I loose about 3% in the night.
In others, I loose about 10-12%! Horrible! In these place, modem have to switch between antennas.
In airplane mode I loose about 2%.
Mt8176 doesn't have lte modem
Mediatek improves modem on X30 (mips based), P23, P30, P60.
I don't test these ones
There are many apps that can use more than two cores for ages. I'm notre a dev, but I think It depends of scheduler and system/librairies too. LITTLE cores configs use more cores.
For exemple when you use a camera on the old mt6592 (octo cortex A7), it uses 3 cores, 1 more core set at half frequency (to be prepare for other task?)...it works with most camera app, not only stock.
With Snapdragon 600, it use 1 core.
That's why in real use mt6592 isn't really slower than S600 despite tiny A7 (except GPU) and far better than quad A7 like s400/mt6582
Others apps can, webbrowsers on multiple pages loading...
If you have to reload a dozen of webpages on X20, it works faster than mt8176 (except on Chrome that doesn't use A72 for any reason, all other browsers including Chrome dev/beta/canary does...strange exception in scheduler ?)
But some other apps do what they want, for exemple Basemark OS refuses to use cortex A72 on X20 (and refuse to use video decoder on rk3288 too). Fortunately there aren't many.
Anonymous, 11 May 2018You don't see the video about 652 and X25?
Like you don't read what gsmarena write about gpu ... moreIdle is better on a MT8176 because its not a three cluster.
Most software today (including on Android) is SMP & only SMP on two core's that's why you don't see the difference between SoC's that have more than two big cores. Only if you have real multi threading which isn't the case on any mobile OS you can have some use from more cores other than that task switching between really CPU intensive tasks can be a sole situation where other two big cores can be useful but most people don't play two games at the same time.
I did a couple of things on this one with S650 & also S652:
I don't do anything else then schedulers & timers, in kernel & heuristics/tuneables. While writing settings whose relative easy fixing scheduling logic in kernel whose a much longer & harder adventure. Every vendor today on their SoC's crosses the sustainable leakage (a lot) and when the SoC crosses the marginal temperature and overheats the error rate is increased, it needs time to lose the heat & has higher power consumption than normal. Trick is in preventing it to get to that point in the first place by tuning CPU schedule logic & applied saint relative limits for both used lithography and maximal sustainable frequency limit on given block (CPU, CPU cluster, GPU...). Both vendors and especially OEM's are very bad at this. Also clever usage of the hotplug can help you a lot in saving significant amount of power that can be used for other things as GPU. You sow the idle difference between the dual & triple cluster MTK SoC's even mid cluster their is based on the two A53's & they don't suck much in idle, additional big core's consume much more in idle & besides power rails are per all cores on the cluster so frequently scaling is shared. Switching off (by hotplug) addition big cores that aren't used anyway can allow significant additional headroom for GPU, end effect is of course lot less GPU throttling.
ZolaIII, 11 May 2018I ain't defending anyone. I told you how much migration between two clusters cost and that is ... moreYou don't see the video about 652 and X25?
Like you don't read what gsmarena write about gpu on p60 and s660...
You're sure of things then you don't go and see if in some case you're wrong.
Ok not common cache...Coherent System Interconnect, my bad! I'm wrong
Sure, tri-cluster will disappear because dynamIQ is better...
But it allows what standard big.LITTLE can't! Link with a third cluster.
Tri-cluster allows the best sustained multithreaded perfs in 20nm...others use 16nm to do better.
On CRC test (Elephant stress...I used it for sustained CPU perfs for my 10minutes test), X20 even destroy Snapdragon 820 by 80%...
810 fully stops all his Cortex A57 in less than 2 minutes!
The lower frequency I see on X20's A72 is 1.8Ghz
Ok it's not very usefull.
GPU throttle a bit (about 30% in long term on GFX-Bench in about 25-30minutes)
To be honest, day-life isn't very different of 2+4 (I test mt8176 on Pad too)
But not very different of Galaxy S7 too (outside gaming, and once more, benchmarks said 200%, real-games say 40 or 50%...I get 27fps on real-racing when S7 gets 40fps - FullHD test)
I'm curious to hear what 650 you test...I see one startings to switch between 1.4 and 1.8Ghz in 3 minutes! (I thought it was an Alcatel but it must be a mistake, I can't find one with S650...or it was an Alcatel with S652, don't remember)
X20 is really underrated.
Pleasant to use like flagship even if it isn't
The only think that could be better : power efficiency...due to 20nm (strange : IDLE seem to be better on mt8176 despite 28nm????)
Anonymous, 11 May 2018And migration on X20 isn't terrible! Corepilot use common cache on A72+A53, like dynanIQ does... moreI ain't defending anyone. I told you how much migration between two clusters cost and that is if core's are up & running, without any hotplug, with hotplug it's 200+ ms because core's & cache (per each & every lv) have pos they have to go thru. Core Pilot tm is a heterogeneous property task scheduler & a hotplug which never had a future because future is open source supported by 90% of industry (including pretty much every other vendor). Common cache? I would advise you not to speak about things you have no idea because nonsense like that is just ridiculous. HMP sched is a first generation, EAV is second (&based on HMP) tho never used, hibrid EAV variants are currently used. I did a lot of work with HMP and it's still a best solution if tuned properly. I told you methodology to test throttling. For instance S650 can push over 20 minutes on top both GPU & CPU utilisation and frequency without any throttling. Telling how X20 can't push even 10 minutes & on CPU side alone actually tells a lot.
ZolaIII, 11 May 2018Now you are lying. Kirin 950 whose best HiSilicon design because they took a lot effort in roo... moreQualcomm doesn't make profit on soc, because 80% of their profit are made on patents! That's how they unfairly destroy competitors!
Take a device that cost 300dollars with Helio P60...
Mediatek will give a soc that cost 15dollars! Must pay tsmc, arm...
Qualcomm will earn 15dollars, 5% of device costs, without giving any hardware! Only allow to use cdma/lte frequency
That's what abuse of dominance is! That's why Mediatek isn't able to sell good soc like X30! That's why Intel, NVIDIA, st-micro, Texas instruments, broadcom quit!
How can you defend Qualcomm?
ZolaIII, 11 May 2018Now you are lying. Kirin 950 whose best HiSilicon design because they took a lot effort in roo... moreAnd migration on X20 isn't terrible!
Corepilot use common cache on A72+A53, like dynanIQ does but with A75 and A55.
My X20 is often as fast as Snapdragon 820 or Exynos...
It even beats 820 on CRC and HTML5 test
You speak about devices you obviously never used!
ZolaIII, 11 May 2018Now you are lying. Kirin 950 whose best HiSilicon design because they took a lot effort in roo... moreI've got X20!
I can run its 10 cores during 10minutes with only 6% performances down on vernee apollo lite!
That's not my test : but here is one! X25 vs 652 throttling. X25 is less stable than X20 due to higher frequency but still far better than 652!
After that, you can't say that I'm lying!
Anonymous, 11 May 2018On Galaxy S throttling is about the same on both Adreno and exynos version Throttling reach s... moreNow you are lying. Kirin 950 whose best HiSilicon design because they took a lot effort in rooting it to the silicon. However best SoC of the same generation whose S650 especially for gaming. Does it throttle? Yes but not much 1.4GHz for a pair of A72's with 550 MHz on a A510 is sustainable. How ever it whose a cheap one & had a smaller power consumption than Kirin 950 so it whose better choice. X20 whose a worse on of the trio mentioned as it throttled the most S652, 653 ware not any better than X20 as they couldn't sustain more than 1GHz on quad A72 cluster. Those SoC's are almost tree years old now and midrange didn't progress at all regarding the performance as while A73's are more efficient they aren't faster at all & GPU's only got on protein diet (paird with faster RAM and bumped a bit clock's as A512 which is actually a A510). MTK triple cluster whose horrible thing because the latency for migration between clusters whose (& still is) high (120 to 200ms) on the new DynamIQ its down to 65~90ms. DynamiQ is a single dynamic variable cluster. ARM even proposed one big core with up to 7 small ones but that is ridiculous. Two are just fine and enough. MTK will remain to be trolled because they are a cheap basterd's and if you think how they are much cheaper than QC you are wrong because Qualcomm makes a rather minimal profit on SoC's while MTK's only profit is from SoC's. MTK improved in some aspects recently but they still have long way to go. I don't love QC, whose much bigger fan of both Texas Instruments and Broadcomm but QC or better say CAF is all what's left of open source this day's. Fun is fun but I like to get my phone's into shape and repair bug's that OEM's simply don't. At the end sustainable power consumption on the regular phones without fancy cooling is about 2.5W and with it around 3, 3.5W top. Most successful design regarding gaming in the recent years whose S835 which S845 is the less efficient but it can be tuned to become better actually on both you can achieve zero GPU throttling. Two big cores +four small one's is still the best typology & will remain so.
ZolaIII, 10 May 2018You have bunch of long run benchmarks on Anandteach. It's natural that big GPU's that are also... moreOn Galaxy S throttling is about the same on both Adreno and exynos version
Throttling reach sometime 50% on GPU!
In real game it's even worst, sometime, in 15 minutes T760mp8 on Exynos 7420 can fall under T760mp2 of mt6752! Ok it's full HD vs QHD but it's not normal! It's real test, not benchmarks!
Most testers only use benchmarks and it's lying.
Huawei with kirin 950 provide 880mp4 that doesn't throttling, it was one of the best gaming machine of the generation...even on real-racing 3 (renderered in full HD on allait machines), it get better framerate than Exynos or snapdragon
Now they do like others! Throttling GPU for benchmarks!
Propaganda of Mediatek????
Most western websites, most western users do Propaganda against them.
Many still think that x20 CPU throttles and that's a lie! 650, 652, 805, 808 throttle far far far more, nobody talk about it!
There was propaganda against octocores when mt6592 was released! Now everybody do octo A53
There was propaganda against tri-cluster and config with 2 Big cores...now Arm propose this kind of config with dynamIQ or Samsung does tri-cluster on 7885 and rumors talk about 2+6 on Snapdragon 700 serie.
They show dual-camera at mwc before the others, nobody talks!
They show AI on X20 at mwc before the other, nobody talks!
Vp9 support before Qualcomm, nobody talks!
Distant wireless charging, nobody talks!
In my country, websites about Android prefere talking about Apple than talking about Mediatek! They choose the P25 version of Meizu Pro 7 and compare it to flagship, not any test of X30!
Mediatek is one of the most trolled company on web!
Anonymous, 09 May 2018Have you test it? No! Read what Gsmarena write! About this F7 About R15 and R15Pro Read mo... moreMediatek should do something ,so people know p60 is different from the older one
DroidBoye, 10 May 2018I want you to focus more on the price point that this device with Helio P60 is being targetted... moreYou can also get P20 Lite for 15,990 PHP with beautiful looks (If performance does not matter to you). Also, cameras are comparable, same Notch display, comparable battery life, and it has Fast Charger! (with USB C)
Anonymous, 10 May 2018Please test. Install fps meter tool like gamebench and mesure in real games...it's often the ... moreYou have bunch of long run benchmarks on Anandteach. It's natural that big GPU's that are also power hungry throttle down still Snapdragons throttle far less than MALI's. Take your time & see independent expert reviews on Anandteach again for Kirin 970, Equnos 9810, Snapdragon 845 and 835 if you dig a bit harder you can also find ones for Apple A SoC's. They all still fall very much regarding sustainable performance behind Adrenos. Don't fall for MTK propaganda materials. Imagination Power VR is still best licensable GPU even their also lost a bit of their magic with determination of driver's quality. Chinese behind an US investment fund bought American Imagination (which is still in US) after US government forcedly split its MIPS CPU business (still mostly in UK) and they ware forbidden by all means from aqureing MIPS as US wants to undercut Chinese long run Academic work on MIPS architecture. Sad story actually. RISC V is a cure for illness. We will see what will happen with Imagination all do I don't have high hopes. If you want to test throttling of a midgrade GPU's (A510, A512, G72 MP3 & similar) in the real available gaming conditions I advise you to use Xenowerk (it's solid, stable uses EGLS 3.0 same as 90% of available games do & it's free) because it has extended settings & it's about as much as they can push maximum anyway on maximum settings with 60 FPS caps limit. Talks about EGLS 3.1 with testilation (Android extension) or Vulkan are unreasonable as we simple aren't there yet and because the hardware isn't capable enough on abridged consumer device for developer's to hassle with potential 5~10% of market especially on platform where people aren't keen paying for anything.
Anonymous, 09 May 201820000 what? Bananas?bananas are little costly, 20000 bananas will cost around 30000
GSM Team, Humble request to you. Can you please include the specific "subjective" comments from the long-term review to your phone reviews going forward? I think the purely objective reviews (like the current one) have outlived their useful life. It is the "Frustrations/Niggles/Annoyances" section in the long-term review which gives a real insight into whether a smartphone is suitable for a person or not. I know this is going to put you in a tight spot - Accusations of being "biased", "bought by X/Y/Z company" and so on. But I think you guys have earned a good reputation for being unbiased over a long period of time, that you can confidently proceed with this.
PS: The reason why I ask for it - I bought a Meizu M1 Note 3 years back, purely based on the excellent GSMArena review once it was available in India. But my experience with that phone was absolutely Terrible. I am guessing that the long-term review would have captured the issues with that phone easily, instead of the pure lab-test based one devoid of subjective observations.
ZolaIII, 10 May 2018Actually Adrenos as still a old ATI design in it's heart & good in FP calculations and not... morePlease test.
Install fps meter tool like gamebench and mesure in real games...it's often the same story :
huge difference in benchmark, small in real games (of course with same settings).
Once more, Gsmarena doesn't notice differences in real games.
About Imagination Technologies, it's a very sad story...unfortunately, they weren't a big competitor in soc for Android, Intel disappears, Exynos use Mali for many years, Mediatek use them sometime (but with very bad sales of X30...)...And as Apple stop using them, they can disappear. Now a chinese tech companie buy them. Chinese want to developpe their own soc, wait and see...
Future : American patent troll vs Chinese company with State's financial help! What a wonderfull world! I'm a stupid european with a fully open market in the middle of that!
It's amazing how powervr can handle heavy resolution even with a small amount of cores.
Even a 2 cores 6250 can handle many games till 2560x1600.
1 core PowerVR=2 or 2.5 cores Mali
But they carry a big weakness for the future...Tesselation is really bad!
In my opinion, it's not a mystery if GFXBench doesn't test tesselation and AnTuTu 7 isn't released on iOS
About GPU frequency, I'm agree...but it's strange that nearly all 800 serie's GPU throttle heavily in benchmarks with low frequency. (see GFX-bench Manhattan 3.1 on-screen and Manhattan 3.1 Long Term).
As thermal and consumption are linked...
Some demoes at MWC 2018, 660 at left, 835 at right :
DroidBoye, 10 May 2018I agree about the graphics performance, it's not competitive with Adreno. Very nice info about... moreFrom the IP owners. P6 is also function/algorithm reprogrammable while QDSP's aren't. CEVA has more powerful designs then Tensilica (CEVA-MX CEVA-NeuPro) regarding NNU (more luts & multipliers) but those are also more of a special purpose (AI NN) than general purpose DSP's and also less scalable but still good backed up with software frameworks and reprogrammable. All do it whose a tuff choice at the end Tensilica had & still has a wining design regarding me even Q6 is only a P6 optimised design.
ZolaIII, 10 May 2018Their is no dedicated AI on MTK P60, it uses Tensilica P6 DSP for the NNU purposes similar as ... moreI agree about the graphics performance, it's not competitive with Adreno. Very nice info about the AI chip, where did you get these? I want to know how it work compared to Qualcomm DSP, Google Pixel, and Huawei NPU.
Anonymous, 10 May 2018It's not the same level at all...
Mediatek use DSP for AI before Qualcomm, it's already the c... moreActually Adrenos as still a old ATI design in it's heart & good in FP calculations and not so much regarding texture fill rate that's why they enchanted the texture unit on Adreno 6 series & more because they ware afraid of Imagination than MALI.
On the contrary game's are mostly about texturing capacity as that after all determine the level of details they can push. That's why you see a large performance gains on the GPU's when you pair them with faster RAM. Benchmarks are just exaggerated with more taxing textures commercial engine based scene's that are there in the first place to tax hardware, so they do reflect the games of tomorrow. It's not a same if one GPU can push the same limited frame rate while keeping the utilisation lower along with lower frequency to another one keeping the same frame rate while being much more utilised & on the top frequency all the time. The Adreno 5 series are almost 2x efficient to G72, Adreno 6 pushes this to 2.4x level. We will see if ARM continues aggressive to push MALI generations every year in the two to three years they me by catch with Adrenos.
I know a rather lot about Tensilica P6, I am D&R, Synopsis, Cadence subscriber and I ever reviewed it some year and a half ago.