Samsung Exynos Thread (big.LITTLE Octa-core)

[III-V]

If you were going to carry a second battery to swap out when needed then you can almost as easily carry an external phone battery/charger like this one.

I carry one for all of my iGadgets and it isn't an issue. Never been reduced to being a "wall hugger" because of it, and I doubt S6 users will become wall huggers either.

Yep, this is definitely what I'd recommend. The only instance where non-swappable batteries really become an issue is when your charging port breaks (all too common with micro USB 2.0) -- otherwise, you'd be able to use an external battery charger.

Speaking of charging ports, I was trying to desolder one at work the other day, and you'd swear the damn thing was soldered with adamantium. I ramped the hot air station to 440C and the sucker still didn't come off.

 

[krumme]

First benchmarks are in:



www.gsmarena.com/samsung_mwc_2015-review-1217p5.php

Almost nailed the offscreen results with my prediction above. Curiously Exynos 7420's performance penalty for increasing the resolution from 1080P to QHD is slightly smaller than Exynos 5433/7410, probably a direct result of the increased memory bandwidth (LPDDR4). GFXbench Offscreen (1080P) tests show ~30% performance improvement while onscreen scores (QHD) are 35-40% better than Mali T760MP6 (Exynos 5433/7410 - LPDDR3).

One of the reasons for that insane AnTuTu score (other than the SoC itself) is the addition of Universal Flash Storage 2.0.

As expected and shows new Mali is very strong with right bandwith. Added benefit of less throttling comes on top of that.
Idle consumption and real world battery consumption is imo what is crusial now.
Keep posting !

 

[sciwizam]

More benchmarks, ones that show the benefit of UFS 2.0

http://www.phonearena.com/news/Firs...n-AnTuTu-and-outrageous-memory-speeds_id66653

The AndroBench memory speed test result shows a mind-boggling improvement in read and write speeds over the Galaxy S5! We'll let the numbers speak for themselves. Where last year's flagship scored 24.2MB/s (sequential write_, the Galaxy S6 edge scored 139.08MB/s! Where the Galaxy S5 scored 176.5Mb/s (sequential read), the Galaxy S6 Edge scored 314.87MB/s! And where the Galaxy S5 scored 13.7MB/s (random read), the Galaxy S6 edge scored 77.2MB/s! All that in just a year! We're extremely impressed with what the new UFS 2.0 storage has to show so far.

 

[lopri]

More S6/Edge Geekbench scores, presumably from MWC.

http://browser.primatelabs.com/geekb...=universal7420
http://browser.primatelabs.com/geekb...rch?q=SM-G925F

 

[krumme]

The s6 and new finfet process shows what samsung is capable of. That company have so steller and broad technology competences. Its incredible it took them so long to use it to protect the vital phone market. They were greedy and incredible shortsighted. It shows the weakness of the company. What they do is so predictable but they can execute the most fantastic. They should really use that far more agressive.

 

[Sweepr]

As expected and shows new Mali is very strong with right bandwith. Added benefit of less throttling comes on top of that.
Idle consumption and real world battery consumption is imo what is crusial now.
Keep posting !

The new scores make me wonder how Exynos 5433/7410's Mali T760MP6 would perform given the same bandwidth as Adreno 420. It's already faster in real games with much lower bandwidth (13.2GB/s vs 25.6GB/s).

More S6/Edge Geekbench scores, presumably from MWC.

http://browser.primatelabs.com/geekb...=universal7420
http://browser.primatelabs.com/geekb...rch?q=SM-G925F

The best Geekbench 3 MT score is now 5663. To put things in perspetive, that's almost twice the iPhone 6's Apple A8.

The top GFXBench scores for Exynos 7420 (Mali T760MP8) are now 59 FPS @ T-Rex offscreen and 26 FPS @ Manhattan Offscreen, faster than Snapdragon 810 and not that far from Tegra K1/Apple A8X inside a thin and light phone.
Edit: According to Andrei @ Beyond3D GPU clockspeed is now 700/773MHz and voltage dropped quite a bit compared to 20nm Exynos 5433/7410 (-300mV top frequency and -200mV acros the board).

http://gfxbench.com/device.jsp?benc...Samsung+Galaxy+S6+(SM-G920)&testgroup=overall

 

[krumme]

I am still amazed by the acceleration here. Cyclone aparently really put everyone's ass on the frying pan.
When qcom position a72 as midrange what will the future highend be? 820 must be really really fast. And Samsung own core must be the same to make sense. With cheap 801-5 28nm series flooding the market now the entire reason for eg atom is nill. That performance is soon history.
Cpu perf development have not been so interesting in years if ever. So many big players. Fantastic.

 

[Sweepr]

Some new comparisons using benchmark results from MWC 2015:

- Geekbench 3 Highest Scores - Multi-Core x Single-Core
Exynos 7420 (Samsung Galaxy S6): 5663 / 1526
Snapdragon 810 (HTC One M9): 4089 / 1276
Snapdragon 810 (LG G Flex 2): 4578 / 1292
Snapdragon 810 Reference Platform: 4884 / 1302
Exynos 5433/7410 (Samsung Galaxy Note 4): 5053 / 1417
Apple A8X (iPad Air 2): 4665 / 1829

GFXBench T-Rex Offscreen:
Exynos 7420 (Samsung Galaxy S6): 59 FPS
Snapdragon 810 (HTC One M9): 48.8 FPS
Apple A8 (iPhone 6): 43.1 FPS
Apple A8 (iPhone 6+): 45.1 FPS
Apple A8X (iPad Air 2): 72.5 FPS
Tegra K1 (Shield Tablet): 64.8 FPS

GFXBench Manhattan Offscreen:
Exynos 7420 (Samsung Galaxy S6): 26 FPS
Snapdragon 810 (HTC One M9): 22.8 FPS
Apple A8 (iPhone 6): 18.2 FPS
Apple A8 (iPhone 6+): 19.3 FPS
Apple A8X (iPad Air 2): 33.4 FPS
Tegra K1 (Shield Tablet): 31.4 FPS

- Expect better results from Exynos 7420 and S810 in the coming months (there aren't lots of submissions yet). Right now S810 has yet to catch up to last year's Exynos 5433 @ Geekbench 3 and Exynos 7420 is in a league of its own.
- Exynos 7420 is roughly 20% faster than Apple A8X @ MT and 20% slower @ single-thread.
- While Galaxy S5 roughly matched Apple's A7 graphics performance Exynos 7420 is 30-42% faster than the Apple A8 @ GFXBench, a very respectable advantage.
- Exynos 7420's GPU offers 85-90% the graphics performance of Tegra K1 inside a thin and small phone profile.

 

[Nothingness]

A French site has quickly tested throttling on S6: the 6th consecutive run of Geekbench shows -1% for ST and -4% for MT.

http://www.tomshardware.fr/articles/galaxy-s6-tests,1-55637.html

 

[Headfoot]

It's pretty crazy how quickly Samsung's "14nm" process is rolling out on the heels of the 20nm process. Regardless of how accurate the term 14nm is, it's still impressive that they've undoubtedly now surpassed TSMC

 

[dawheat]

A French site has quickly tested throttling on S6: the 6th consecutive run of Geekbench shows -1% for ST and -4% for MT.

http://www.tomshardware.fr/articles/galaxy-s6-tests,1-55637.html

Sounds impressive compared to the G Flex 2 obviously, but how long does it take to run each Geekbench test? Curious if we're talking about 6 minutes of load or 20.

 

[Nothingness]

Sounds impressive compared to the G Flex 2 obviously, but how long does it take to run each Geekbench test? Curious if we're talking about 6 minutes of load or 20.

I was wondering the same... Anyone can comment on how long it takes to run Geekbench on a fast smartphone?

 

[Arachnotronic]

I was wondering the same... Anyone can comment on how long it takes to run Geekbench on a fast smartphone?

It's really, really fast...much, much faster than on a PC.

Just ran it on my iPhone 6 Plus...took under a minute.

 

[North01]

GFXBench T-Rex Offscreen:
Exynos 7420 (Samsung Galaxy S6): 59 FPS
Snapdragon 810 (HTC One M9): 48.8 FPS
Apple A8 (iPhone 6): 43.1 FPS
Apple A8 (iPhone 6+): 45.1 FPS
Apple A8X (iPad Air 2): 72.5 FPS
Tegra K1 (Shield Tablet): 64.8 FPS

GFXBench Manhattan Offscreen:
Exynos 7420 (Samsung Galaxy S6): 26 FPS
Snapdragon 810 (HTC One M9): 22.8 FPS
Apple A8 (iPhone 6): 18.2 FPS
Apple A8 (iPhone 6+): 19.3 FPS
Apple A8X (iPad Air 2): 33.4 FPS
Tegra K1 (Shield Tablet): 31.4 FPS

Some of those numbers are wrong for the Galaxy S6, there seems to be a glitch on the GFX webpage. Here are links to the results:

T-Rex - 2510 frames (44.8 fps)

Manhattan 3.0 - 1095 frames (17.7 fps)

-----

For comparison

GFXBench Manhattan 3.1 Offscreen (version 3.1 tests GLES 3.1 features):

NVIDIA Shield Tablet (Tegra K1 32-bit) - 1427 frames (23.0 fps)

Galaxy Note 4 (Snapdragon 805) - 773 frames (12.5 fps)

Galaxy S6 (Exynos 7420) - 634 frames (10.2 fps)

Galaxy Note 4 (Exynos 5433) - 435 frames (7.0 fps)

 

[Sweepr]

Some of those numbers are wrong for the Galaxy S6, there seems to be a glitch on the GFX webpage. Here are links to the results

Something happened with the top submissions there, you can see that the scores from your link were obtained today (not yesterday) so they are not the same. Those scores are almost the same as Exynos Note 4 (and lower than GSMArena for the S6 results too) so I'd rather trust what I posted. Probably a website error.

 

[North01]

What do you mean? They are the top scores @ T-Rex / Manhattan 3.0 for each device.

I know, but when you click on the submitted scores to the results of the device (the links I provided), you don't see any result that correlates with those numbers.

We'll have to wait and see as more official results are added.

 

[Sweepr]

I know, but when you click on the submitted scores to the results of the device (the links I provided), you don't see any result that correlates with those numbers.

We'll have to wait and see as more official results are added.

Probably a website/database error. Edited my post above.
Scaling from Exynos 5433/7410 makes it clear that those scores are wrong.

 

[North01]

Probably a website/database error. Edited my post above.
Scaling from Exynos 5433/7410 makes it clear that those scores are wrong.

Possibly, but here are the top results for the new GFXBench 3.1 Manhattan benchmark (not many devices have been benchmarked yet).

 

[Nothingness]

It's really, really fast...much, much faster than on a PC.

Just ran it on my iPhone 6 Plus...took under a minute.

So that'd be 6 minutes of benchmark... Definitely not enough for testing throttling.

 

[Arachnotronic]

Sounds impressive compared to the G Flex 2 obviously, but how long does it take to run each Geekbench test? Curious if we're talking about 6 minutes of load or 20.

Here's the iPhone 6 Plus, for anyone interested.

http://forums.anandtech.com/showthread.php?t=2423179

 

[lopri]

I did not realize that I posted dead links in post #279. Here are the correct addresses.

http://browser.primatelabs.com/geekbench3/search?page=1&q=universal7420
http://browser.primatelabs.com/geekbench3/search?q=SM-G925F

 

[Fjodor2001]

So, just concluding that big.LITTLE is the shit these days in mobile. Funny so many doubted it was a good design decision. I wonder how long until we'll see it on laptops and desktop...

 

[Sweepr]

Possibly, but here are the top results for the new GFXBench 3.1 Manhattan benchmark (not many devices have been benchmarked yet).

Lots of submissions here (you need to select GFXBench 3.0 instead of 3.1):
T-Rex Offscreen, Manhattan Offscreen

I wouldn't read too much into those OpenGL ES 3.1 results yet, could be software related (benchmark doesn't play nice with Mali, bad/unoptimized/early drivers, etc.). This API was implemented with Lollipop so it's very recent.

 

[North01]

Lots of submissions here (you need to select GFXBench 3.0 instead of 3.1):
T-Rex Offscreen, Manhattan Offscreen

I wouldn't read too much into those OpenGL ES 3.1 results yet, could be software related (benchmark doesn't play nice with Mali, bad/unoptimized/early drivers, etc.). This API was implemented with Lollipop so it's very recent.

Nice! I'm glad to see those early results had just been an anomaly. I'll look forward to seeing results from the final product. Although, I wish they would release GFXBench 4.0 already (tests OpenGL 4.x and features such as tessellation). I'd be curious to see how these devices perform with Android Extension Pack.

 

[Exophase]

So, just concluding that big.LITTLE is the shit these days in mobile. Funny so many doubted it was a good design decision. I wonder how long until we'll see it on laptops and desktop...

It's fair to conclude that big.LITTLE makes sense for ARM's current CPUs. That's about it. No one has yet done a design with other CPU types, so we don't know for sure what's better (if such thing can really be ascertained). There are a ton of variables and tradeoffs to consider. ARM has markets where the little core is going to be interesting by itself, so it makes sense for them to do both cores - for others it might not be justifiable. And while big.LITTLE allows cores to be better optimized due to having a more restricted dynamic range, and it allows better utilization for some asynchronous workloads, it comes at a cost to area and overhead sharing/transferring data between clusters. So who knows.

Don't hold your breath seeing it in laptops or desktops. The main motivation for workloads that are common today is power savings, and the benefit is lower there.