Samsung Galaxy S series rumours.....

[Deleted member 13524]

Here's the breakdown for Exynos (GT9500) or Snapdragon (GT9505) according to each country:

http://www.sammobile.com/2013/03/20...variant-of-the-galaxy-s-4-we-have-the-answer/


It seems to me that the Exynos version has no LTE so it generally goes for countries with less developed network infrastructures. Some countries will apparently sell both versions (it should depend on the provider, though).

We're yet to know where that dual-SIM version will appear, though. Maybe it's only China but hopefully it'll be available through merimobiles, pandawill, etotalk and the likes.

 

[Nebuchadnezzar]

Bill-of-materials analysis by IHS: http://press.ihs.com/press-release/...ries-236-bill-materials-ihs-isuppli-virtual-t

Surprisingly (or not) the Exynos/HSPA version is more expensive than the Snapdragon/LTE version.

http://press.ihs.com/sites/ihs.newshq.businesswire.com/files/2013-03-19_Samsung-S4-Table1.jpg

Korean news site reports that Samsung plans to ship 70% of the first 10mil units as Snapdragon variants: http://www.etnews.com/news/device/device/2739089_1479.html#ystfuv

They also talk about how the problem is intrinsically linked to the A15 high heat generation. Didn't really understand what they did about it as the translation is gibberish, maybe somebody can help out there with what they mean.

 

[Deleted member 13524]

If the big.LITTLE allows for all cores to be used at the same time, I don't get the 4+4 configuration instead of a 2+2 in a smartphone, within the next couple of years.

A net/chromebook could make sense with more intense browsing and productivity using keyboard + mouse, but in a smartphone I get the feeling that anything more than a 2* Cortex A7 + 2* Cortex A15 will do little more than just spend more power and give the marketing department nerdgasm material.

At 28nm in a smartphone, that Exynos 5410 will be throttling all the time because of the heat sensors, IMO.

 

[Helmore]

If that BOM is anything to go by, then Qualcomm will make around $60 for each Snapdragon Galaxy S4 sold. Not bad. I hadn't expected the Snapdragon 600 to only be $20, but what do I know? >.<

 

[tangey]

I've seen some rumours of a exynos5210, and its suggested usage in the S4 mini. Assuming their is any credence to this chip, it would be interesting to see if Samsung go back to mali400 for this lower class chip.

http://www.sammobile.com/2013/03/24/rumour-samsung-galaxy-s4-mini-to-feature-exynos-5210-processor/

 

[Ailuros]

I've seen some rumours of a exynos5210, and its suggested usage in the S4 mini. Assuming their is any credence to this chip, it would be interesting to see if Samsung go back to mali400 for this lower class chip.

http://www.sammobile.com/2013/03/24/rumour-samsung-galaxy-s4-mini-to-feature-exynos-5210-processor/

Yes but that article suggests a dual A15; even if that's the case it'll most likely be a 2+2 big.LITTLE config otherwise it wouldn't make sense for a mainstream smartphone. As for the GPU A15 with a mali400 would be a weird combination. There's nothing obligatory with ARM GPU IP to use 4 clusters for instance like in the T604 of the 5250 on the other hand.

 

[french toast]

Gsm arena has put some reviews and comparisons up of the galaxy s4.
First up is the batterylife stress test..which is one of the better ones out there, it includes standby time, talktime, Web browsing and video play back.
http://blog.gsmarena.com/samsung-galaxy-s4-battery-tests-are-done-heres-how-it-did/

Im pretty pleased with those results too be honest. .how samsung have managed to increase performance, display size/resolution, battery size. ..whilst decreasing dimensions and also increasing battery life..astonishing.

Im not a fan of the plastic construction. .but if this is the trade off then it's certainly worth it.

Also ive spotted a new screen mode in the display settings..which reads "professional RGB"
-looks promising.

Edit fixed.

 

[Helmore]

Gsm arena has put some reviews and comparisons up of the galaxy s4.
First up is the batterylife stress test..which is one of the better ones out there, it includes standby time, talktime, Web browsing and video play back.
http://blog.gsmarena.com/lg-optimus-g-pro-battery-tests-are-over-here-are-the-numbers/

Im pretty pleased with those results too be honest. .how samsung have managed to increase performance, display size/resolution, battery size. ..whilst decreasing dimensions and also increasing battery life..astonishing.

Im not a fan of the plastic construction. .but if this is the trade off then it's certainly worth it.

Also ive spotted a new screen mode in the display settings..which reads "professional RGB"
-looks promising.

Y̶o̶u̶'̶r̶e̶ ̶l̶i̶n̶k̶i̶n̶g̶ ̶t̶o̶ ̶a̶n̶ ̶a̶r̶t̶i̶c̶l̶e̶ ̶a̶b̶o̶u̶t̶ ̶t̶h̶e̶ ̶L̶G̶ ̶O̶p̶t̶i̶m̶u̶s̶ ̶G̶ ̶P̶r̶o̶.̶.̶.̶.̶.̶

EDIT: It's been fixed.

 

[Nebuchadnezzar]

Octa version:

GLBenchmark, incomplete. Also 2.7.0: http://www.nomobile.ru/upload/images/197388/samsung_galaxy_s4_645_16.jpg

Definitely lower than what 533MHz would have been if you do a direct comparison with the iPhone 5. Maybe it is indeed in the 4xxMHz range. Fillrate and triangle tests are very low compared to what it should be; again, taking the iPhone 5 into account.

From http://habrahabr.ru/company/nomobile/blog/174559/

All the reviews have been popping up with 1.8GHz clocks though, so that's curious.

Also 12h30 video playback on the Octa version versus 10h15 on the Snapdragon.

 

[tangey]

Octa version:

off-screen benchmark is 1.4-ish times the iphone5, which might point to the 480-490Mhz range. off-screen fillrate however is just about the same as the iphone5.

Mind you the on-screen textured triangles is sitting at 861M !, so it calls the rest of those figures into question.

Also 12h30 video playback on the Octa version versus 10h15 on the Snapdragon.

That seems weird, I wonder does that suggest different video decode IP ? I guess they might be using IMG's video decode in the Octa.

 

[Turbotab]

Octa version:

GLBenchmark, incomplete. Also 2.7.0: http://www.nomobile.ru/upload/images/197388/samsung_galaxy_s4_645_16.jpg

Definitely lower than what 533MHz would have been if you do a direct comparison with the iPhone 5. Maybe it is indeed in the 4xxMHz range. Fillrate and triangle tests are very low compared to what it should be; again, taking the iPhone 5 into account.

From http://habrahabr.ru/company/nomobile/blog/174559/

All the reviews have been popping up with 1.8GHz clocks though, so that's curious.

Also 12h30 video playback on the Octa version versus 10h15 on the Snapdragon.

In the video from that Russian website @ 2.14 to 2.18, you can see Quadrant listing:
Current Freq - 1600 MHz
Max Freq - 1800 MHz

http://www.youtube.com/watch?feature=player_embedded&v=3jMIo3Tykmg

If correct, that must be a single-core turbo value, the only single-threaded benchmark that IMO completes fast enough to avoid throttling would be Sunspider, also Samsung have always optimised their browser for fast sunspider scores on their Exynos procs, so we should see sub 600ms Sunspider scores on the Octa, if that turbo figure is correct and usng the Nexus 10 sub 700ms score on Chrome Beta as a benchmark.

What really confuses me is the Fillrate score is only 10% higher than the iPhone 5 clocked at 300 ish MHz, assuming the same number of ROP, is there some bottleneck in the PowerVR5 uArch that is limiting linear performance gains.

 

[Ailuros]

What really confuses me is the Fillrate score is only 10% higher than the iPhone 5 clocked at 300 ish MHz, assuming the same number of ROP, is there some bottleneck in the PowerVR5 uArch that is limiting linear performance gains.

The 543MP3 in the iPhone5 should be clocked at 325MHz and it gets in GLB2.7 fillrate offscreen 1757 MTexels/s while the Galaxy S4 gets 1986 MTexels/s; definitely a difference in fillrate efficiency but then again the iPhone5 runs under iOS while the S4 under Android, let alone all the other differences (like GPU system level cache size f.e.) that could be there.

The frequency should be at 480MHz (now where did I hear that number before ) and I'm actually pleasantly surprised with the 4.6k frames in GLB2.5. I don't want to think of its GLB2.7 score as it'll be rather dissapointing against an Adreno320 at the same frequency.

 

[ams]

Im pretty pleased with those results too be honest. .how samsung have managed to increase performance, display size/resolution, battery size. ..whilst decreasing dimensions and also increasing battery life..astonishing.

The battery life endurance testing of Galaxy S4 at GSM Arena looks quite respectable. That said, do note that the Galaxy S3 that GSM Arena tested used the Exynos 4 Quad SoC that was fabricated on a less advanced and less mature 32nm fabrication process (vs. the more advanced and more mature 28nm fabrication process for the S600 SoC used in the Galaxy S4). Also note that the Galaxy S4 has about 24% more battery capacity than the Galaxy S3.

GSM Arena tests battery life endurance in three tests: talk time, web browsing, and video playback. The talk time test has the screen off with the CPU processor in a low voltage state. The S600 [Krait] CPU uses asynchronous symmetric multi-processing with independent frequency and voltage scaling for each CPU core. This saves some power compared to a traditional variable symmetric multi-processing CPU architecture such as the Exynos 4 Quad [Cortex A9] CPU at idle. That could in part explain why the S600-based Galaxy S4 has 34% more talk time than the Exynos 4 Quad-based Galaxy S3, even though battery capacity has increased by "only" 24%. That said, moving forward, most variable symmetric multi-processing CPU architectures will make use of a power sipping battery saver core(s) such as demonstrated in big.LITTLE or 4+1, so these newer vSMP architectures will probably gain an upper hand in idle power consumption compared to aSMP architectures such as Krait.

In the web browsing battery life endurance test, most likely GSM Arena is loading one web page at a time. The S600-based Galaxy S4 has higher single threaded CPU performance vs. the Exynos 4 Quad-based Galaxy S3, so it can load and render web pages more quickly and then go to sleep in order to save power. This improved speed and improved time to sleep combined with newer software, newer fabrication process, and increased battery capacity would help to explain why web browsing battery life is nearly 65% more on the Galaxy S4 compared to the Galaxy S3.

Do note that web browsing battery life tests in general may depend on the quality and reliability of the internet connection at the time of the test, and GSM Arena's data has some unexpected results. For instance, even though the Galaxy S4 has nearly 12% more battery capacity than the HTC One, and even though both are using the same S600 SoC (with the Galaxy S4 clocked about 12% higher), the HTC One ends up getting nearly 15% better web browsing battery life than the Galaxy S4. Even more strange is some of the other web browsing battery life data in their expanded list. For instance, the HTC One X+ gets nearly 85% more (!) web browsing battery life than the HTC One X.

In the video playback battery life endurance test, the Galaxy S4 only gets about 2.5% better battery life than the Galaxy S3, even though it has much higher battery capacity in comparison, and even though the Galaxy S3 does not make use of any power sipping battery saver CPU core(s). Did GSM Arena use a higher video playback resolution for the Galaxy S4 compared to the Galaxy S3? Who knows. Still, all things considered, certainly very respectable endurance testing.

 

[ams]

That seems weird, I wonder does that suggest different video decode IP ? I guess they might be using IMG's video decode in the Octa.

No, I think it suggests that the power sipping battery saver CPU core(s) in Exynos 5 Octa are more power efficient during video playback vs. the aSMP Krait CPU cores.

 

[french toast]

The battery life endurance testing of Galaxy S4 at GSM Arena looks quite respectable. That said, do note that the Galaxy S3 that GSM Arena tested used the Exynos 4 Quad SoC that was fabricated on a less advanced and less mature 32nm fabrication process (vs. the more advanced and more mature 28nm fabrication process for the S600 SoC used in the Galaxy S4). Also note that the Galaxy S4 has about 24% more battery capacity than the Galaxy S3.

GSM Arena tests battery life endurance in three tests: talk time, web browsing, and video playback. The talk time test has the screen off with the CPU processor in a low voltage state. The S600 [Krait] CPU uses asynchronous symmetric multi-processing with independent frequency and voltage scaling for each CPU core. This saves some power compared to a traditional variable symmetric multi-processing CPU architecture such as the Exynos 4 Quad [Cortex A9] CPU at idle. That could in part explain why the S600-based Galaxy S4 has 34% more talk time than the Exynos 4 Quad-based Galaxy S3, even though battery capacity has increased by "only" 24%. That said, moving forward, most variable symmetric multi-processing CPU architectures will make use of a power sipping battery saver core(s) such as demonstrated in big.LITTLE or 4+1, so these newer vSMP architectures will probably gain an upper hand in idle power consumption compared to aSMP architectures such as Krait.

In the web browsing battery life endurance test, most likely GSM Arena is loading one web page at a time. The S600-based Galaxy S4 has higher single threaded CPU performance vs. the Exynos 4 Quad-based Galaxy S3, so it can load and render web pages more quickly and then go to sleep in order to save power. This improved speed and improved time to sleep combined with newer software, newer fabrication process, and increased battery capacity would help to explain why web browsing battery life is nearly 65% more on the Galaxy S4 compared to the Galaxy S3.

Do note that web browsing battery life tests in general may depend on the quality and reliability of the internet connection at the time of the test, and GSM Arena's data has some unexpected results. For instance, even though the Galaxy S4 has nearly 12% more battery capacity than the HTC One, and even though both are using the same S600 SoC (with the Galaxy S4 clocked about 12% higher), the HTC One ends up getting nearly 15% better web browsing battery life than the Galaxy S4. Even more strange is some of the other web browsing battery life data in their expanded list. For instance, the HTC One X+ gets nearly 85% more (!) web browsing battery life than the HTC One X.

In the video playback battery life endurance test, the Galaxy S4 only gets about 2.5% better battery life than the Galaxy S3, even though it has much higher battery capacity in comparison, and even though the Galaxy S3 does not make use of any power sipping battery saver CPU core(s). Did GSM Arena use a higher video playback resolution for the Galaxy S4 compared to the Galaxy S3? Who knows. Still, all things considered, certainly very respectable endurance testing.

You put a good break down across.
I would say nothing of the galaxy s4 is a night and day improvement, but it manages to improve by a noticeable amount in every area. .and it does this despite being thinner,narrower and lighter..perhaps slightly better built.

The battery life is very good..video didnt improve by much, but we have a larger, brighter 1080p display there...got to play some part.

Standby time is also there and I that is very good also, web browsing for that specification AMOLED screen is astonishing. .beating even high end smaller ips displays. .thats quite a turn around considering amoleds inefficiency displaying whites.

Yes kraits are asynchronous and very efficient on TSMC 28nm LP process..which must be quite mature by now...but put into the equation the extra performance on tap of the whole SOC...its a very well engineered ballanced phone Samsung had produced.

Lets not forget mature firmware improvements will likely yield even better battery life and performance.

On the screen front..here is a nice (but not exactly scientific) screen comparison from one of my go to phone sites. Phone arena.

http://www.phonearena.com/reviews/S...-Z-vs-One-vs-Galaxy-S-III-vs-Lumia-920_id3286

This is one area where samsung has fallen down in past galaxy phones (once you have got over seeing the first saturated colour/contrasty displays).

Amoleds offered amazing visuals with some pretty awefull tradeoffs. ..once ips panels made their wsy onto phones with better contrast, wider colour gamut and viewing angles...not to mention superior pixel densities. ..samsungs displays were starting to feel out of touch..seemed to me they would be unable to match lcds in tgeir traditional weak areas.

Samsung has obviously poured many millions ££ into improving it because aside from the htc one...it matches (or nearly matches) lcds in every area..whist still winning on contrast.

The new professional rgb mode is something fans and bloggers alike have been calling for to match the realism of ips..it looks like they have pretty much pulled it off.

Very impressive.

 

[Nebuchadnezzar]

That seems weird, I wonder does that suggest different video decode IP ? I guess they might be using IMG's video decode in the Octa.

...?

Video decoding is done on a separate IP block from the CPU or GPU, it's dedicated silicon for video decoding and encoding. Samsung's is called MFC / Multi-Function-Codec; generally it has been always the most capable hardware decoder in all SoC implementations. It could handle 1080p high profile 50mbps streams without a hitch where other vendors lacked the resources.

The Octa probably has a new revision. The battery advantage may come frm that, but most of it simply comes from the fact that video decoding is very CPU friendly, you can do the above with a single core locked at 500MHz, the A7's leakage and power characteristics in this use-case are a definite win over the Snapdragon.

The talk time test has the screen off with the CPU processor in a low voltage state. The S600 [Krait] CPU uses asynchronous symmetric multi-processing with independent frequency and voltage scaling for each CPU core. This saves some power compared to a traditional variable symmetric multi-processing CPU architecture such as the Exynos 4 Quad [Cortex A9] CPU at idle. That could in part explain why the S600-based Galaxy S4 has 34% more talk time than the Exynos 4 Quad-based Galaxy S3, even though battery capacity has increased by "only" 24%.

Irrelevant.

The S3 limits itself to a single core when the screen is off. Wether the Snapdragons are made to do this to I don't know, in either case the Snapdragons have no advantage in that use-case, and possibly a disadvantage if they're badly configured (This was the case in previous generations, it fired up all cores even in screen off).

In the video from that Russian website @ 2.14 to 2.18, you can see Quadrant listing:
Current Freq - 1600 MHz
Max Freq - 1800 MHz

I just checked Antutu and it's incorrectly reading out the frequency. It's reading the maximum frequency in the frequency table, not the actual maximum frequency per DVFS policy. I'm starting to get annoyed with the absolute technical incompetence of reviewers in the mobile space.

The chip may very well be running at 1600MHz in this case, Quadrant correctly reads out the max policy value.

 

[tangey]

I still think we can't take those glbenchmark numbers too seriously given the completely false triangle number it's reporting (nearly 1G)

 

[ams]

The S3 limits itself to a single core when the screen is off. Wether the Snapdragons are made to do this to I don't know, in either case the Snapdragons have no advantage in that use-case, and possibly a disadvantage if they're badly configured (This was the case in previous generations, it fired up all cores even in screen off).

There is no doubt that differences in implementation and process technology could explain the differences in talk time (rather than any inherent differences in aSMP vs. vSMP). That said, a traditional vSMP multi-core [ARM] CPU architecture will typically have a relatively large L2 cache (say, 1MB for quad-core Cortex A9, and 2MB for quad-core Cortex A15) that will need to be shared by all the CPU cores, irrespective of how many CPU cores are active. An aSMP multi-core [Krait] CPU architecture will typically have a relatively small L2 cache (say, 512KB) for each CPU core. So even with only one CPU core active, aSMP could have some power savings vs. traditional vSMP due to use of a smaller L2 cache for that one core (assuming that the 512KB L2 cache is not over saturated in the first place). A newer vSMP architecture with companion core(s) could negate that aSMP power savings by a) using a smaller L2 cache that is dedicated to the companion CPU core, and/or by b) using a much more power efficient companion CPU core.

 

[Nebuchadnezzar]

An aSMP multi-core [Krait] CPU architecture will typically have a relatively small L2 cache (say, 512KB) for each CPU core.

This is not the case: http://images.anandtech.com/reviews/SoC/Qualcomm/Krait/krait.jpg

http://www.anandtech.com/show/4940/qualcomm-new-snapdragon-s4-msm8960-krait-architecture

Asymmetric clocks between the cores don't mean dedicated L2 cache for each core.

 

[ams]

This is not the case: http://images.anandtech.com/reviews/SoC/Qualcomm/Krait/krait.jpg

http://www.anandtech.com/show/4940/qualcomm-new-snapdragon-s4-msm8960-krait-architecture

Asymmetric clocks between the cores don't mean dedicated L2 cache for each core.

Did Anand get that diagram from Qualcomm, and did Qualcomm re-work their CPU architecture when moving from dual-core Krait to quad-core Krait? According to NVIDIA, quad-core Krait has 512KB dedicated L2 cache for each CPU core, rather than a larger 2MB shared L2 cache. Xbit Labs makes the same claim too (albeit for the upcoming quad-core S800).