Boy that took long! Considering how green LG is behind its ears for SoC development we should cut them some slack.
I wonder the most powerful and/or commercial successful SOC do not rely on it.
It makes sense in a controlled environment or if you have silicon in spare. Right it is not the case especially with the resolution being push absurdly high, with cache amount still being pretty low, etc.
Samsung's Exynos based smartphones and phablets all have big.LITTLE configs. If they're NOT commercially successful in the Android world then who then would be?
For tablet Allwinner has the A80T as a high end big.LITTLE SoC for tablets and a 4+4 A7 A83T for mainstream tablets http://www.allwinnertech.com/en/news/compnews/452.html . Allwinner is right behind Apple in tablet market share.
It creates issue wrt software, it could be fine in a controlled environment like iOS but on Android I think it create more issue than it solves.
It won't create issues if the implementation is decent enough. There aren't any reported problems so far from implementations from Mediatek, Allwinner and others I've heard. If there's something shaky on the sw platform then it's not necessarily big.LITTLE at fault there.
I hope Nvidia stick to its in house core (so that if those cores are any good), I also hope they do not fall for the "mandatory four core" mantra that seems to plague embedded devices.
Current Tegra K1 successor is codenamed Erista laid out under 20SoC@TSMC. It'll contain according to rumors a 4*A57+4*A53 big.LITTLE config and a Maxwell "grandchild" GPU. If there's going to be a Denver (custom CPU) core variant no idea.
With its ranking system definitely the guys behind Antutu got something right.
Now what mediatek did made and make sense, the market has a whole took a wrong direction imho.
If you consider how ridiculously low the die area for each A7 core is it's not a bad decision after all. They could just have done that "little.LITTLE" right.
But it is true wrt to resolution, we are way past dimishing returns and actually crazy resolution on mobile devices create a lot of problems both hardware (crazy requirement) and software (size of the content being delivered to the devices, etc.).
I mean I laugh when I read reviewers saying that 200ppi is causing them some issue, how not sharp it is... how about eating its own BS. Even at 200ppi a phone (or tablet for that matter) is still the most sharp screen one has access to, end of the argument.
Why should resolution be a problem with the rather crazy fillrates today's GPUs have? Even a G62x0 Rogue is enough to even fit a hypothetical 8k screen.
Well they might be indeed but it is not needed, I expect their krait 400 and more to still be top of the line even after lollipop release.
What I don't get is why they decided to engage in the race to the bottom with Mediatek and the like. They did when they embraced ARM Cortex A7 quad core configuration in place of their own in-house design. They gave a lot of credence to competition and the below 4 cores SOC might suck. Now they are in tougher spot to differentiate themselves as competition is catching up.
Chosing low end A7 ARM cores for their low end/budget SoCs was a very conscious choice by QCOM and I don't think it'll ever change in the future. Custom CPU cores are far more expensive than standard CPU IP from ARM and for what they get for each low end SoC the added costs for a low end variant of a custom CPU is not worth it.
Besides if you look over A15 cores from ARM you might come up with quite a few disagreements; A7 cores however have one hell of a perf/mm2, perf/$ and perf/mW ratio. Quite hard to beat.
WRT to MTlio well it doesn't exist I thought it would be obvious that I was using my nickname, I should have gone with MTliolio would have been easier
Samsung's Exynos based smartphones and phablets all have big.LITTLE configs. If they're NOT commercially successful in the Android world then who then would be?
It looks like to me that Samsung uses Qualcomm chips every time they get the chance on, they introduce a Snapdragon based S5 for example.
Apple still does not consider it.
For tablet Allwinner has the A80T as a high end big.LITTLE SoC for tablets and a 4+4 A7 A83T for mainstream tablets http://www.allwinnertech.com/en/news/compnews/452.html . Allwinner is right behind Apple in tablet market share.
Imo it is a cheap marketing point, I'm not sure having 2 clusters of CPU can't actually prove detrimental to performances.
It won't create issues if the implementation is decent enough. There aren't any reported problems so far from implementations from Mediatek, Allwinner and others I've heard. If there's something shaky on the sw platform then it's not necessarily big.LITTLE at fault there.
THere are no serious reviews anyway. Heterogeneous computing makes thing tough for the OS how and where to pin down a process or thread?
Now most apps on mobile are made with bottom specs in mind, how snappy or reactive phones are might often have nothing to do with CPU performance at large and even less with the number of cores.
Current Tegra K1 successor is codenamed Erista laid out under 20SoC@TSMC. It'll contain according to rumors a 4*A57+4*A53 big.LITTLE config and a Maxwell "grandchild" GPU. If there's going to be a Denver (custom CPU) core variant no idea.
If you consider how ridiculously low the die area for each A7 core is it's not a bad decision after all. They could just have done that "little.LITTLE" right.
Why should resolution be a problem with the rather crazy fillrates today's GPUs have? Even a G62x0 Rogue is enough to even fit a hypothetical 8k screen.
Chosing low end A7 ARM cores for their low end/budget SoCs was a very conscious choice by QCOM and I don't think it'll ever change in the future. Custom CPU cores are far more expensive than standard CPU IP from ARM and for what they get for each low end SoC the added costs for a low end variant of a custom CPU is not worth it.
I hope Denver doesn't suck and they are used in more than one SoC but let see Nexus is going to be thoughtfully reviewed soon enough.
Besides if you look over A15 cores from ARM you might come up with quite a few disagreements; A7 cores however have one hell of a perf/mm2, perf/$ and perf/mW ratio. Quite hard to beat.
It seems that ARM overshot it with the A15 and they were too power hungry for the process available for cheap. ARM was left with nothing in between the A7 and A15 for a long while. The comparing the A9 (even the latest revision) to both A15 and A7 is a tad unfair as both are newer design.
There is nothing inherently wrong with cores offering more solid serial performance, something were wrong with the A15, in an alternative world ARM could have release both A7 & A17 2 years ago.
Touche I fell for it I'd still prefer a quad A7 config at 1.5GHz over a dual A9 config at 1.2GHz. The key is in the perf/mW part of the first
It is not cut that straight, there are overall few reviews as pointed out earlier. SOme fast CPU can more than often burn more in spike but overall less than a low power CPU.
Anyway I agree with you here I would take the A7 too (or even more the A53) over the the old A9 or newer A17 @1GHz. THough the choice might not be that clearly cut either.
It would be worth it to have in depth review of device like the Galaxy S4 mini and mega 6.3, both using a 1.7GHz x2 Krait 300 + adreno 305 Snapdragon 400, against other devices using more popular Snapdragon 400.
But looking at the screen size and battery size, and going by review that are not as good as I would want, it appears that actually the difference ain't that bad, especially once you account for the fact that in all cases (HTC816, Xperia t2, LG G2 mini or Xperia M2) it is the x4 A7 SOC that runs at the lowest speed (either 1.2 1.4 or 1.6 GHz). EDIT
there is also more RAM in Samsung phone 1.5GB, then there are also software considerations)
To be fair and using the Galaxy mega 6.3 and the Xperia t2 ultra as ref: the choice would be more like that:
x2 A17 @1.7GHz
x4 A7 @1.4GHz
The first chip I think comparing battery life and battery size and taking in account for screen size (6 vs 6.3) might consume more but far from a night and day difference.
Now if it were x2 A17 @1.4GHz vs x4 A7 @1.4GHz I think the difference would be even lesser, and real world performances (for the ref Krait are wider CPU than A17).
At the same speed the A17 offers better IPC, support for full speed SIMD computation, etc. I would favor the first set-up.
The thing is Quad-core is almost a market requirement to pretend to be above the lowest end, but ultimately it proves nothing above the technical merit of SOC that would rely on reasonable OoO CPU design.
EDIT
Further dreaming about a world where Qualcomm came with a successor for the A9 not the A15, the 64bit successor would be about to be released.
I would take a dual Cortex-A7 at 1.2GHz, because that's what is in the ZTE Open C (Snapdragon 200). But I'm concerned the battery capacity might be too low (1400mAh). So I'm happy camping on the outside of the "smartphone revolution", casually taking a glimpse sometimes.
I sort of would want low end performance and high end battery life, more text reading than video editing. But performance still matters as sadly these days your browser need to execute a few billion instructions to render badly written web pages and ultimately show you a few kilobytes of text you wanted to read .
Other rambling : is Wifi 5GHz getting common? (it seems to on laptops/desktops sort of, with the new 802.11ac standard, though 11n 2.4GHz-only still will be common).
That's a big problem since you need a second antenna, have it wired.. But 2.4GHz is very crowded so if I were to have a home wifi network and esp. desktop and smartphone often communicating, I would like it 5GHz.
THe Snapdragon 200 ain't bad going by the Moto E reviews, I wonder if under average usage we would fail to notice the difference with its big bro with 4 cores.
Now 1400mAh sounds indeed low the Moto E does pretty good but it runs on a 1980mAh battery.
It looks like to me that Samsung uses Qualcomm chips every time they get the chance on, they introduce a Snapdragon based S5 for example.
Apple still does not consider it.
Samsung has to buy 3rd party SoCs either way, since while manufacturing for Apple SoCs they probably don't have enough foundry capacities to cover their own entire demand just with Exynos SoCs.
Imo it is a cheap marketing point, I'm not sure having 2 clusters of CPU can't actually prove detrimental to performances.
To performance in a sterile fashion certainly NOT. But for perf[/b]/W[/b] (if done right) which is THE most important metric in this market definitely yes. You can run each quad CPU cluster always only at the same frequency. Having two different clusters even of the same kind of cores from ARM gives the luxury that you can run the active cores of cluster A at 500MHz and the active cores of cluster B at 1.5GHz as an example if needed.
THere are no serious reviews anyway. Heterogeneous computing makes thing tough for the OS how and where to pin down a process or thread?
Now most apps on mobile are made with bottom specs in mind, how snappy or reactive phones are might often have nothing to do with CPU performance at large and even less with the number of cores.
For the above case there's nothing the underlying sw has to do to gain the additional efficiency. Unfortunately the 6592 doesn't exactly belong to the above example.
I hope Denver doesn't suck and they are used in more than one SoC but let see Nexus is going to be thoughtfully reviewed soon enough.
I'm personally all for wider IPCs and less cores; Apple just went for 3 ultra phat Cyclone cores in A8X.
It seems that ARM overshot it with the A15 and they were too power hungry for the process available for cheap. ARM was left with nothing in between the A7 and A15 for a long while. The comparing the A9 (even the latest revision) to both A15 and A7 is a tad unfair as both are newer design.
There is nothing inherently wrong with cores offering more solid serial performance, something were wrong with the A15, in an alternative world ARM could have release both A7 & A17 2 years ago.
There's no magic wand for anyone out there; everyone learns from its own mistakes
It is not cut that straight, there are overall few reviews as pointed out earlier. SOme fast CPU can more than often burn more in spike but overall less than a low power CPU.
Anyway I agree with you here I would take the A7 too (or even more the A53) over the the old A9 or newer A17 @1GHz. THough the choice might not be that clearly cut either.
It would be worth it to have in depth review of device like the Galaxy S4 mini and mega 6.3, both using a 1.7GHz x2 Krait 300 + adreno 305 Snapdragon 400, against other devices using more popular Snapdragon 400.
But looking at the screen size and battery size, and going by review that are not as good as I would want, it appears that actually the difference ain't that bad, especially once you account for the fact that in all cases (HTC816, Xperia t2, LG G2 mini or Xperia M2) it is the x4 A7 SOC that runs at the lowest speed (either 1.2 1.4 or 1.6 GHz). EDIT
there is also more RAM in Samsung phone 1.5GB, then there are also software considerations)
To be fair and using the Galaxy mega 6.3 and the Xperia t2 ultra as ref: the choice would be more like that:
x2 A17 @1.7GHz
x4 A7 @1.4GHz
The first chip I think comparing battery life and battery size and taking in account for screen size (6 vs 6.3) might consume more but far from a night and day difference.
Now if it were x2 A17 @1.4GHz vs x4 A7 @1.4GHz I think the difference would be even lesser, and real world performances (for the ref Krait are wider CPU than A17).
At the same speed the A17 offers better IPC, support for full speed SIMD computation, etc. I would favor the first set-up.
The thing is Quad-core is almost a market requirement to pretend to be above the lowest end, but ultimately it proves nothing above the technical merit of SOC that would rely on reasonable OoO CPU design.
EDIT
Further dreaming about a world where Qualcomm came with a successor for the A9 not the A15, the 64bit successor would be about to be released.
If you're going for a low end SoC than a small sea of A7 or A53 makes more sense. All a manufacturer needs to do is to get the cache coherencies and cluster configs right in order to give the CPU cores the varying frequency flexibility they need to save more power.
If standard ARM cores wouldn't be bound to have all the same frequency within each cluster, we wouldn't most likely have this debate
I have to say I agree with Lolio..all this marketing BS with tons of low power cores is resulting in products with sub-optimal performance hitting the market. And as he mentioned, even Qualcomm was forced to do it because the market demanded it (take even the current Snapdragon 400 series, they went from dual Krait to quad A7, which is a regression in single thread performance). It seems like it is the easiest way to market specs to consumers. After all, the average consumer dosent have any idea how a A7 compares to a A15 or Krait 400. But "Octa Core" just sounds good to them!
But just want to add that its not that the A7 is bad or anything. Its reasonably good, and very power efficient to boot. I do not notice any significant difference in day to day performance between my Galaxy S3 with quad A9's at AFAIK 1.4 ghz vs my Redmi 1s with quad A7 at 1.6 ghz. But I would have preferred something like a dual Krait config instead (the S3 was sold with this config in the US market actually). I got a Sony Xperia Z2 just a few days back (got a deal too good to pass up) and I have to say, Krait 400 at 2.5 ghz really blows anything out of the water. And battery life is great too. I hope Qualcomm's custom 64 bit core is as good versus the competition.
Oh yes, I would argue that a lot of people change smartphones every year in fact. By long term, I meant more than a year. I have a few stories from friends and some people on forums that their phones (Micromax, Karbonn, etc, which are rebranded Chinese phones basically) die after about a year..year and a half of usage. And some reports of receiving lemons or experiencing hardware issues (I know this happens even to the big brands, but here it is seemingly a lot more common with the Chinese phones). In comparison, I have a Samsung Galaxy S from 2010 which is still working. And my S3 from 2012 is still going strong too.
I don't even recall which SoC your Redmi contains but I think it's a MTK6589; things like wifi and/or GPS signal could be lacklustering depending how old the SoC version really is. After a certain point Mediatek fixed quite a few things and integrated the wifi/GPS stuff into the SoC..
As torrentranz mentioned, its a Snapdragon 400, but clocked at a rather high 1.6 ghz. Its not a bad phone at all, and has some nice specs. The issues are more with the software, especially RAM limitations in MIUI, but the phone also does heat up a fair bit..and has relatively poor battery life ( presumably due to the higher clocked and probably higher voltage CPU)
Interesting..I didn't know it was on HPM..always thought it was LP. Any idea about their cortex A53 chips? The product pages do not have any info (MT6732 and MT6752).
Qualcomm's Cortex A53 SoCs (Snapdragon 410, 610 and 615) are on on 28LP as per Anandtech.
IMHO QCOM will bounce back to custom CPU cores in due time for =/>mainstream SoCs of the future. Smaller players not only have a hard time finding as many resources for custom core development, but in their majority avoid high end SoCs exactly because they sell at very low prices.
Your disagreements are mostly with ARM's own design decisions and not pure marketing tendencies per se.
I disagree "overall" and I don't don't think APple is a good example of adding more cores, they've been as conservative as can be, now they had the silicon budget and reason for people to upgrade.
Yet Apple stuck to 1 and 2 cores longer than anybody else. They also went though most the possible means to improve "real world" performances (within indeed the constraints of their power budget), they have (it seems /undisclosed) a great cache hierarchy and memory sub-system with a L3.
Truth be told I'm not sure that this 3rd core will had much in real world usage as I would think that throttling could trigger fast if all three cores are working at the same time.
My pov is that Apple decisions lately are a mixed bag that is hidden by the fact that their late adoption of a bigger screen has been a huge driver for a big part user base to refresh, it is a one shot move imo.
I won't do a review of what I think was good decision and what was bad here it is a bit OT but I'm not sure including a 3 cores was that great an ideav /will sell any device vs the other improvements coming with the devices.
I disagree "overall" and I don't don't think APple is a good example of adding more cores, they've been as conservative as can be, now they had the silicon budget and reason for people to upgrade.
Yet Apple stuck to 1 and 2 cores longer than anybody else. They also went though most the possible means to improve "real world" performances (within indeed the constraints of their power budget), they have (it seems /undisclosed) a great cache hierarchy and memory sub-system with a L3.
Truth be told I'm not sure that this 3rd core will had much in real world usage as I would think that throttling could trigger fast if all three cores are working at the same time.
My pov is that Apple decisions lately are a mixed bag that is hidden by the fact that their late adoption of a bigger screen has been a huge driver for a big part user base to refresh, it is a one shot move imo.
I won't do a review of what I think was good decision and what was bad here it is a bit OT but I'm not sure including a 3 cores was that great an ideav /will sell any device vs the other improvements coming with the devices.
The point you are missing is that Apple unlike Mediatek or the others are:
1. Designing only high end SoCs where more expensive custom cores make only sense.
2. Have the resources to develop custom cores.
Mediatek and the likes are selling their SoCs at single digit $ figures and have miniscule profit margins with very high volumes. Apple has insane profit margins and insane volumes and for high end products only.
Now if Apple wouldn't have the resources for custom CPU cores they'd be chosing what from the ARM CPU IP product pallette right now?
You also are forgetting that Apple has their own OS which is a further luxury no one else has.
IMHO QCOM will bounce back to custom CPU cores in due time for =/>mainstream SoCs of the future. Smaller players not only have a hard time finding as many resources for custom core development, but in their majority avoid high end SoCs exactly because they sell at very low prices.
Your disagreements are mostly with ARM's own design decisions and not pure marketing tendencies per se.
What exactly do you mean by "Mainstream SoCs"? Oh and I dont see them returning to their own cores before 16FF, so late 2015 at best. Yes of course, there is no reason for players like Mediatek to develop custom cores of their own. Sidenote: I remember Samsung announcing aw hile back that they would develop cores of their own..did anything come of that?
Not exactly..ARM's design decision aside..there is no good reason to go 8XA7 rather than 2XA15+2XA7, or 8xA53 rather than 2xA57+2xA53.
I'm not even trying to keep up. Too much of a hassle. I'm not in the market for a new phone or tablet anyway right now, I'm mainly on this forum because of general interest in mobile SoCs.
What exactly do you mean by "Mainstream SoCs"? Oh and I dont see them returning to their own cores before 16FF, so late 2015 at best. Yes of course, there is no reason for players like Mediatek to develop custom cores of their own. Sidenote: I remember Samsung announcing aw hile back that they would develop cores of their own..did anything come of that?
Mediatek's own "high end" SoCs are mainstream at best compared to their competition. Oh and don't scare me with Samsung; it took them more than one attempt to get big.LITTLE right; a 64bit custom core sounds like a damn tall order compared to that
Not exactly..ARM's design decision aside..there is no good reason to go 8XA7 rather than 2XA15+2XA7, or 8xA53 rather than 2xA57+2xA53.
Oh yes there is: die area and cost. When your majority of SoCs are sold at single digit $ figures you don't have the luxury for extravagant design choices. Mediatek's current "high end" SoC for smartphones is a 4xA15+4xA7 config and they'll cover their "mainstream" with 8*A7 configs.
MT6795 is IMHO an 8*A53 config as the 6752. The first is supposed to clock up to 2.2GHz and is most likely a PowerVR, while the latter up to 2.0GHz with a T760.
Mediatek's own "high end" SoCs are mainstream at best compared to their competition. Oh and don't scare me with Samsung; it took them more than one attempt to get big.LITTLE right; a 64bit custom core sounds like a damn tall order compared to that
Agreed..I was not disputing that. You mentioned that Qualcomm's custom 64 bit core would be ready for future mainstream SoCs. Was just asking what you considered mainstream SoCs.
Of course it isn't easy..but if there is anyone else apart from Qualcomm who has the resources and the volume to consider doing a custom core, it is Samsung (AMD is also apparently making one but who knows if they'll follow through). Since we haven't heard any news..was wondering if anyone had any info (Nebuchadnezzar maybe?)
Oh yes there is: die area and cost. When your majority of SoCs are sold at single digit $ figures you don't have the luxury for extravagant design choices. Mediatek's current "high end" SoC for smartphones is a 4xA15+4xA7 config and they'll cover their "mainstream" with 8*A7 configs.
Yes those do play a part of course. But the difference between a 2xA15+2xA7 config and a 8xA7 config are not that significant IMHO. From what I've been able to find, one A7 core takes about one-fourth (25%) to one fifth (20%) the die area of an A15 core. Lets split the difference and take it as 22.5%. So 2xA15 + 2xA7 would be 245% the area of a single A15. And 8xA7 would be 180% the area of an A15, i.e. the A15 config is ~35% larger. As per ARM, the area of an A7 core on 28nm is about 0.5 mm2. So as per this, the A15 config would be 5.4 mm2 and the A7 config would be 4 mm2. Hardly significant.
The main thing is that the single threaded performance of the A15 would be far higher and more noticeable in use. Well worth the minor increase in die size I would say. But its all a marketing game. I doubt the average consumer will notice on a spec sheet if its MT6592 or 6595 in a phone as both are "Octa core".
MT6795 is IMHO an 8*A53 config as the 6752. The first is supposed to clock up to 2.2GHz and is most likely a PowerVR, while the latter up to 2.0GHz with a T760.
I am reasonably sure that 6795 is Cortex A57+A53. There would be little point in making two Octa core A53 SoCs with a different GPU. The press release and product page for 6795 mention "True Octa Core" and HMP, which implies big.LITTLE. Also in the product roadmap its shown as the successor to 6595 so thats another giveaway - http://www.gizbeat.com/4739/new-mtk-soc-review-mt6795-mediatek-soc-roadmap/
However, all the info about it online points to it being 28nm. I dont think anyone else is attempting A57 on 28nm so power efficiency will definitely be a question mark.
Erinyes,
As I'm on the run in the hospital expecting our 2nd angel, I need to get back to a computer for proper research. I recall 1/6 A7 to A15 die area but aren't sure.
Erinyes,
As I'm on the run in the hospital expecting our 2nd angel, I need to get back to a computer for proper research. I recall 1/6 A7 to A15 die area but aren't sure.
Erinyes,
As I'm on the run in the hospital expecting our 2nd angel, I need to get back to a computer for proper research. I recall 1/6 A7 to A15 die area but aren't sure.
But what ARM engineers did show me at a recent analysts' conference is that relative performance of the Cortex-A7 and A15 differs by a factor of two or three depending on workload and implementation. Now, when you add that to the fact that the Cortex-A7 occupies about one quarter or one fifth of the die area and consumes one quarter to one fifth of the power, things become interesting (see chart below).
Samsung's material shows a A7 core taking 1/6 the die area and roughly 1/6 the power usage, at a significantly lower top frequency than the A15.
When paired with a 1/4 of the L2 cache of the A15, the die area scale is then 1/5.
Assuming the same cache proportions; a 2xA15+2xA17 configuration would use 11.4mm2 for cores and cache, while a 4xA7+4xA7 configuration would use 7.6mm2.
I'd still prefer a quad A7 config at 1.5GHz over a dual A9 config at 1.2GHz. The key is in the perf/mW part of the first 
