Low-cost emerging market SoC/phone discussion

[Ailuros]

Here's what seems to be a great quality smartphone with a MT6595:

http://www.pandawill.com/lenovo-vib...hd-screen-2gb-16gb-glonass-golden-p93566.html

No wonder it won a design award; finally a smartphone that doesn't look like a gazillion others. The only other thing I'd want on that one is a larger battery.

A friend had bought a THL5000 with a MT6592 and a 5000mAh battery; a battery lifetime to drool over :smile:

 

[Deleted member 13524]

No wonder it won a design award; finally a smartphone that doesn't look like a gazillion others. The only other thing I'd want on that one is a larger battery.

A friend had bought a THL5000 with a MT6592 and a 5000mAh battery; a battery lifetime to drool over :smile:

Details are still a little bit scarce, but I'll probably be pitching that Lenovo to my wife.
The GSMArena coverage on it mentions it supports some kind of modularity through "Vibe Xtension Layers" that you plug in the back of the smartphone.
One of them is a JBL speaker/stand and the other is an additional battery that increases the thickness by 5mm and adds a 1725mAh battery (for a total of 4000mAh).
Given how the smartphone is only 7.3mm thick, 12.3mm with the extra battery isn't that much of a problem:

Even less if you can just take the battery off when you need the phone to be lighter/thinner.

 

[Mariner]

Lenovo do produce some very good quality phones. My only previous complaint about them was that I think the styling was generally so-so. The Vibe X2, on the other hand, looks to be a very stylish phone and the initial 'hands-on' reviews seem to indicate it feels very good in the hand as well.

Just a pity it doesn't have a MicroSD slot and a larger battery.

 

[Ailuros]

Lenovo do produce some very good quality phones. My only previous complaint about them was that I think the styling was generally so-so. The Vibe X2, on the other hand, looks to be a very stylish phone and the initial 'hands-on' reviews seem to indicate it feels very good in the hand as well.

Just a pity it doesn't have a MicroSD slot and a larger battery.

Zopo has a MT6595 phone now on presell for the 30/11/14 http://www.tinydeal.com/zopo-zp999-55-ltps-fhd-mtk6595m-android-44-8-core-4g-phone-p-139947.html

It looks like a gazillion other phones, but Zopo also has a quite good build quality, it's slightly cheaper than the Vibe, a slightly larger battery and a MicroSD slot. The LTPS display with sapphire crystal also speaks for it (I think it's from Sharp?). What's somewhat turning me off is the 5.5" display. 5" are more than enough in my book *cough*

 

[Erinyes]

2MB on the A15s, 512KB on the A7s, for 5430. Haven't had time to look at 5433 properly. The A53 cluster on that chip neighbours some busy, SRAM-heavy logic that makes it harder to delineate the A53's L2 cells. From a top-level eyeball it's probably the same though.

I'm a little surprised it isn't 1 MB but I suppose any task that would need it would be switched to the A57's anyway.

Ok let's take it once more from the top:

Assuming an A7 costs you 1/4th in die area compared to an A15 and freely inventend figures for 2mm2 for each A15 and 0.5mm2 for each A7:

* 4xA15 + 4xA7 = 10 mm2
* 8xA7 = 4 mm2

First case delivers far more overall performance but also at the cost of significantly more die area and power consumption. If you're Mediatek, Rockchip or whoever else that wants to sell a lot of relatively small SoCs at very low prices, you'd obviously would avoid an as big core as A15 as much as possible.

You missed out the configuration I've been harping on, 2xA15 + 2xA7 = 5 mm2.

It would have much higher single threaded performance and this would be a lot more noticeable in the real world than 8xA7s.

I forgot to ask this question earlier but I am curious..would licensing costs play a big part as well? And does it cost say 4x to license a A15 vs an A7? Basically, does licensing 2xA15 + 2xA7 cost more than 8xA7? From what I understand, typical costs are in the region of cents per core so it shouldn't play a huge part anyway.

Cortex-A17 vs Cortex-A53 on Geekbench
The A17 is in another league, it should be compared against A15.

Yea A17 is way faster than A53. Thanks for that link, hadn't seen MT6752 results posted anywhere.

I am a bit disappointed by the performance of the A53 though. I'm seeing performance increases more in the 15-25% range vs A7 than the ARM claim of 40% - Cortex A53 vs Cortex A7

Adreno 306 is also quite a bust in the Desire 510. It barely matches the 305, forget surpassing it - GfxBench comparison (Look at offscreen results as the devices have different resolutions)

 

[Helmore]

ARM's claim of 40% more performance is probably on AArch64 code and aren't these benches on 32-bit code?

 

[Ailuros]

You missed out the configuration I've been harping on, 2xA15 + 2xA7 = 5 mm2.

It would have much higher single threaded performance and this would be a lot more noticeable in the real world than 8xA7s.

Mediatek has it in the tablet only MT8135. It'll still burn significantly more power than the 8xA7 config since in the latter the second cluster should rarely fire up except some synthetic nonsense like Antutu.

The MT8135 made it into Amazon's new line of low end tablets, one Teclast tablet no one wants to have and that's about it. I could swear that Chinese white box tablets integrated the MT6592 compared to the former quite a bit more.

The point is that MTK stayed away even for their current high end smartphone SoC the MT6595 from any A15 core.

I forgot to ask this question earlier but I am curious..would licensing costs play a big part as well? And does it cost say 4x to license a A15 vs an A7? Basically, does licensing 2xA15 + 2xA7 cost more than 8xA7? From what I understand, typical costs are in the region of cents per core so it shouldn't play a huge part anyway.

That's a question others are better suited to answer, but in Mediatek's example above they obviously have obtained licenses for A7, A15, A17 and recent 64bit ARM cores. For them it wasn't obviously a matter of licensing cost. For royalties it should obviously make a difference per core. The more recent the core and the higher the performance the higher the royalty should be.

Either way judging from Laurent's Geekbench link and Tottentranz's legitimate point about performance it's likelier that they have in their next 64bit "high end" smartphone SoC a 4+4 config then a 8*A53 config as I thought.

 

[Erinyes]

Mediatek has it in the tablet only MT8135. It'll still burn significantly more power than the 8xA7 config since in the latter the second cluster should rarely fire up except some synthetic nonsense like Antutu.

Yes..the point is they never offered it in a smartphone chip with an integrated baseband.

I disagree. Thanks to big.LITTLE, the A15 cores shouldn't fire up all that often either so power consumption should be in check. Even if it does consume more instantaneous power, it could consume less overall energy for the task as it would complete the task much quicker (Similar to Apple's strategy with A6/A7/A8).

The point is that MTK stayed away even for their current high end smartphone SoC the MT6595 from any A15 core.

Umm the A17 is in the same league as the A15. It made sense to use it as its a newer core.

That's a question others are better suited to answer, but in Mediatek's example above they obviously have obtained licenses for A7, A15, A17 and recent 64bit ARM cores. For them it wasn't obviously a matter of licensing cost. For royalties it should obviously make a difference per core. The more recent the core and the higher the performance the higher the royalty should be.

Oops my mistake, I meant to say royalties in my post, got it mixed up with licensing which obviously they've obtained for all the cores anyway.

 

[Ailuros]

Yes..the point is they never offered it in a smartphone chip with an integrated baseband.

I disagree. Thanks to big.LITTLE, the A15 cores shouldn't fire up all that often either so power consumption should be in check. Even if it does consume more instantaneous power, it could consume less overall energy for the task as it would complete the task much quicker (Similar to Apple's strategy with A6/A7/A8).

Bears the question then why MTK hasn't used the MT8135 or a variant of it for smartphones instead of the 6592. The 6592 gets away with 5" displays at around 2400-2500mAh batteries and it has a slightly better battery life than a quad A7 MT6589 at lower peak CPU frequencies.

"Much quicker" is relative to the frequency of each core used. In the unique Tegra 4+1 paradigm the companion core starts at a base frequency of around 50MHz, while on the fore mentioned 6589/6592 SoCs the A7 cores start at 500MHz. In other words the whole affair is far too complex to solve with a simple equation.

The point remains: Mediatek had already the A15 license and a big.LITTLE implementation for the tablet 8135 (not at any extravagant frequencies mind you for the A15 either despite it being tablet material), why didn't they use something similar for smartphones also?

Umm the A17 is in the same league as the A15. It made sense to use it as its a newer core.

If memory serves well the A15 in 8135 are R3. In order to make the circle though to my initial point the MT6595 wherever found right now in smartphones comes at quite higher device prices than former smartphone SoCs powered with MTK stuff. In other words MTK is growing its SoCs and asking probably higher prices for them.

 

[Mariner]

This is interesting:

http://www.gsmarena.com/htc_desire_820s_with_mediatek_chipset_goes_official-news-10127.php

Apparently the Snapdragon in the 'basic' Desire 820 has Octa-A53s but one cluster is clocked lower than the other. Interesting design choice, which ought to be better for battery life?

The Desire 820s has an Mediatek Octa-A53 chip which runs at higher clocks.

All else apparently being equal with the phones, it will be interesting to see how these two chips compare, both in performance and battery life. Different clocks, GPUs and modems but the rest of the hardware should be the same.

 

[Erinyes]

Bears the question then why MTK hasn't used the MT8135 or a variant of it for smartphones instead of the 6592.

As I've been alluding to all along..its easier to market an "Octa core" than a "Quad core". And add to that the fact that MT8135 is also more expensive thanks to the bigger CPU cores and more powerful GPU (and associated licensing/royalty costs)

The 6592 gets away with 5" displays at around 2400-2500mAh batteries and it has a slightly better battery life than a quad A7 MT6589 at lower peak CPU frequencies.

That's pretty much due to the fact that 6592 is on 28HPM vs 28LP for 6589.

If memory serves well the A15 in 8135 are R3. In order to make the circle though to my initial point the MT6595 wherever found right now in smartphones comes at quite higher device prices than former smartphone SoCs powered with MTK stuff. In other words MTK is growing its SoCs and asking probably higher prices for them.

AFAK R3 brought some improvements in power management but thats about it. I'm not sure what the relevance of that is though. True..the 6595 devices are more expensive. This is partly due to the fact that the SoC itself is more expensive and that the phones it is powering have generally better specifications than its other, less powerful SoCs. MTK is probably asking more money, but it would still be a fair bit lower than the likes of Qualcomm I would think.

Anyways I think we've discussed this to death already.

Moving on..from a link in LordEC911's post on another thread, I found reference to a new 28nm process from TSMC, 28 HPC (High Performance Compact). I hadn't even heard about this but it seems it entered volume production in September. Some details below:-

Compared with TSMC's 28LP, 28HPC provides 10% smaller die size and 30% lower power at all levels of speed, or over 20% speed improvement at the same power, through tighter process control, more efficient design solutions, and new process features

Link - http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&language=E&newsid=8761

I can see a number of players shifting to this from 28LP. For the benefits it offers, seems like its definitely worth it. TSMC says 10 products have already taped out and they expect 70 more by the end of 2014. Some mentions were Allwinner's octa-core A83 AP, Amlogic's S812 4K HEVC modem, with quad-core Cortex-A9 and Mali 450 GPU. MStar is using 28HPC for 4K UHD television processor and Spreadtrum is offering smartphone ICs using the technology.

 

[Ailuros]

As I've been alluding to all along..its easier to market an "Octa core" than a "Quad core". And add to that the fact that MT8135 is also more expensive thanks to the bigger CPU cores and more powerful GPU (and associated licensing/royalty costs)

Cost is the primary reason and 8*A7 are cheaper in about everything (be it royalties, power or whatever else).

That's pretty much due to the fact that 6592 is on 28HPM vs 28LP for 6589.

It's not that the only difference between them is just one more quad A7 cluster at higher clocks. They've also migrated from a single core SGX544@354MHz to a Mali 450MP4@700MHz amongst others. However humble the 450's may be in die area overall that's still twice the TMUs than the former and twice the frequency. HPM to LP might make a healthy difference; on estimate I'd estimate almost a half node's difference between them.

AFAK R3 brought some improvements in power management but thats about it. I'm not sure what the relevance of that is though. True..the 6595 devices are more expensive. This is partly due to the fact that the SoC itself is more expensive and that the phones it is powering have generally better specifications than its other, less powerful SoCs. MTK is probably asking more money, but it would still be a fair bit lower than the likes of Qualcomm I would think.

I've read somewhere some time ago that they plan to aim higher, which in essence makes sense since they cannot grow in sales volumes forever and they need to grow/expand for the future somehow.

Anyways I think we've discussed this to death already.

It helps me reconsider my so far conclusions and ideally track my mistakes.

Moving on..from a link in LordEC911's post on another thread, I found reference to a new 28nm process from TSMC, 28 HPC (High Performance Compact). I hadn't even heard about this but it seems it entered volume production in September. Some details below:-
Link - http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&language=E&newsid=8761

I can see a number of players shifting to this from 28LP. For the benefits it offers, seems like its definitely worth it. TSMC says 10 products have already taped out and they expect 70 more by the end of 2014. Some mentions were Allwinner's octa-core A83 AP, Amlogic's S812 4K HEVC modem, with quad-core Cortex-A9 and Mali 450 GPU. MStar is using 28HPC for 4K UHD television processor and Spreadtrum is offering smartphone ICs using the technology.

I had read about HPC a while ago in the german 3DC forum; interesting aspect that Allwinner uses it for its mainstream tablet SoC which is a 2*(4*A53) config.

 

[Deleted member 13524]

True..the 6595 devices are more expensive. This is partly due to the fact that the SoC itself is more expensive and that the phones it is powering have generally better specifications than its other, less powerful SoCs. MTK is probably asking more money, but it would still be a fair bit lower than the likes of Qualcomm I would think.

At least looking at the recent Zopo 999 and Lenovo Vibe X2 models with the 6595, they seem to be a bit more expensive than some models carrying the Snapdragon 801, like the ZTE Nubia Z7 Max/Mini - all other specs being equivalent.

 

[Ailuros]

The Meizu MX4 is still holding its presell "crown" (most expensive yet MT6595 device) at 480+ Euros.

 

[Erinyes]

It's not that the only difference between them is just one more quad A7 cluster at higher clocks. They've also migrated from a single core SGX544@354MHz to a Mali 450MP4@700MHz amongst others. However humble the 450's may be in die area overall that's still twice the TMUs than the former and twice the frequency. HPM to LP might make a healthy difference; on estimate I'd estimate almost a half node's difference between them.

I never implied that there was no other change. We were discussing battery life of MT6589 and 6592 and my point was simply that the difference is largely due to HPM.

I've read somewhere some time ago that they plan to aim higher, which in essence makes sense since they cannot grow in sales volumes forever and they need to grow/expand for the future somehow.

Maybe they do but they are positioned very well to ride the wave of growth that is happening in the low-mid range market. That segment has been growing much faster than the high end segment for the last few years. So even if they stay where they are..they will still do quite well!

I had read about HPC a while ago in the german 3DC forum; interesting aspect that Allwinner uses it for its mainstream tablet SoC which is a 2*(4*A53) config.

I think its a A7 based SoC rather than A53. This link seems to indicate so as well - http://www.prnewswire.com/news-rele...-octa-core-for-full-hd-tablets-273938041.html

Allwinner's roadmap shows their 64 bit A9X series is due only in Q4'15

 

[Ailuros]

I never implied that there was no other change. We were discussing battery life of MT6589 and 6592 and my point was simply that the difference is largely due to HPM.

There's no noticable difference in the majority of real time cases; if MTK would had gotten the cache stuff for "little.LITTLE" on the 6592 right, power might have been even in its advantage.

Maybe they do but they are positioned very well to ride the wave of growth that is happening in the low-mid range market. That segment has been growing much faster than the high end segment for the last few years. So even if they stay where they are..they will still do quite well!

Growth is always bigger where prices are lower and volumes higher. Note that Mediatek and the likes are only pushing for higher end CPUs lately, while still staying on the conservative side for GPUs. It's perfectly understandable since they still want to keep their SoCs as cost effective as possible. The G6200 the 6595 contains is not only a dual cluster Rogue, but also optimized for area while skipping the FP16 ALUs the 6230 has and FB compression.

I think its a A7 based SoC rather than A53. This link seems to indicate so as well - http://www.prnewswire.com/news-rele...-octa-core-for-full-hd-tablets-273938041.html

Allwinner's roadmap shows their 64 bit A9X series is due only in Q4'15

Ooops you're right.

 

[Deleted member 13524]

Huawei/HiSilicon announces new Kirin 620 mid-level SoC:

- 28nm
- 8x Cortex A53 @ 1.2GHz
- Mali 450MP4
- Integrated Cat4 LTE
- LPDDR3

This sounds like a jab at the Snapdragon 615 and Mediatek MT6752, except it's seemingly made by terrible choices.
An ancient OpenGL ES 2.0 GPU and eight low-power cores working at 1.2GHz. The point of those other SoCs is to have 4 clock-optimized and 4 area-optimized cores for some kind of little.LITTLE operation, but this one seems to have all of them low-clocked.
Performance should be equivalent or worse than a Snapdragon 410 but I guess they'll play with marketing numbers and unrealistic benchmarks.

The dumbest part is they already made a big.LITTLE chip already with the Kirin 920, using four rather low-clocked Cortex A15 (1.6GHz I think) and 4 low-clocked Cortex A7, along with a Mali T628.
A more than natural alternative to a lower grade chip would be to just cut everything in half: 2*A15 + 2*A7, along with an updated Mali T720/760, but no... they had to play the marketing numbers game.
The dumbest part is that this may not even be any smaller, cheaper, faster or more power efficient than their already existing 28nm Kirin 910 with 4* Cortex A9 and a Mali 450MP4.

 

[Kaarlisk]

The one measure that "got me" regarding these quad/eight core A7 SOCs was when, just for then fun of it, I benchmarked my current phone.
Turns out my 1.4GHz Scorpion is actually faster than a Moto G for web browsing. That phone cost $175 3 years ago...
So unless my phone breaks, I cannot justify buying anything that does not have at least dual cores at the performance level of dual 1.7GHz Krait.

 

[Deleted member 13524]

Turns out my 1.4GHz Scorpion is actually faster than a Moto G for web browsing. That phone cost $175 3 years ago...

Are you sure about that? I think many websites at some point tested the Snapdragon 400 in the Moto G against many dual-core Snapdragon S4 models with Krait 200 @ 1.5GHz and found the first to be generally faster in web browsing.
In no small part, this should be due to Moto G using a much better optimized WebKit browser together with a more recent version of Android.
That said, I'd find it hard to believe a single-core Scorpion would be faster than a higher-clocked dual-core Krait..

 

[Zaphod]

Due to a somewhat busted micro USB/charging port I picked up a Huawei G620S for €100 as a stopgap (tax and shipping included). Snapdragon 410 (4xA53), 5", 720P, NFC, 1GB RAM (KitKat is so much better than ICS/JB for multitasking), 8GB ROM, FM, BT 4.0, 8MP/2MP (surprisingly good), etc.

If this is the the present status and pace of advancement for lower end handsets I am quite amazed. I doubt 95% of my friends and family could tell the difference in speed and useability of this thing vs. a LG G2, Galaxy S4 or similar.

Of course it doesn't feel like a €350 handset in terms of build, but it's still pretty damn nice. Only drawback was the included laucher, but I put the stock 4.4 one on it from the market (improved speed and battery life) after which this thing performs quite admirably.