x PhoneArena is hiring! Reviewer in the USA

Snapdragon 820 V3 scores high on benchmark test

Posted: , by Alan F.

Tags:

Snapdragon 820 V3 scores high on benchmark test
The Snapdragon 820 is apparently undergoing more testing and revisions by Qualcomm. Earlier this month, a Qualcomm executive said that 30 handsets that will use the new chipset, are being planned. Coming off the Snapdragon 810 overheating debacle, everyone will be looking intently at the chip, especially since they feature Qualcomm's own Kryo cores.

Back in July, the Snapdragon 820 was benchmarked by GeekBench which produced a single-core score of 714, and a multi-core score of 1942. Those aren't exactly high-end results. But the four Kryo cores were running at less than half the maximum 3GHz clock speed (1.46 GHz, to be exact). Another test of the Snapdragon 820 showed improvement, with a single-core tally of 1732, and a multi-core score of 4970.

That brings us to the test of the third version of the Snapdragon chip, which was recently conducted. This time, the chipset scored 2032 in single-core, and 5910 in multi-core. This beat out the Exynos 7420 powerhouse and the multi-core score of the Apple A9.

The Snapdragon 820 is just one of many new high-end chips that will be driving top-shelf handsets over the next year. One competitor is Huawei's Kirin 950 SoC. Last month, that chip scored 6096 for its multi-core test. There is also the deca-core CPU offered as part of MediaTek's Helio X20 chipset.

The Xiaomi Mi 5 has been rumored to be the first handset to be powered by the Snapdragon 820. The chip is expected to start shipping to manufacturers in December, with the first handsets to be powered by the SoC to start rolling out next March.


source: TechGrapple

68 Comments
  • Options
    Close




posted on 27 Sep 2015, 12:57 8

1. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


very respectable and great scores.... one big key is how much/long can the s820 sustain the performance without degrading too much? that is equally just as important if not more important then trying to hit a "higher peak performance" but then dropping like a rock lol after a sustained workload. would love to eventually see A9 vs S820 v3/v4 (what ever version s820 comes out in) performance degradation numbers/thermal throttling numbers. A9 seems to do actually pretty well in sustaining near peak performance after a sustained workload which means it can maintain near its benchmark scores almost indefinitely, just like the A8 can, seems to be a feather in Apple's hat when it comes to SoC, designing so that sustained performance is key

either way we are all in for a great treat next year. 820 is coming with some awesome stuff on the SoC, Anti malware IP blocks dedicated hardware blocks on the SoC, Qualcomm's Zeroth platform which is a form of cognitive computing and brings some very awesome interesting stuff, blazing fat CAT LTE, better ISP, etc

these scores make me assume that the 820 will have little performance degradation as Qualcomm is going that route judging by the scores

posted on 27 Sep 2015, 13:03 7

3. Bm888 (Posts: 390; Member since: 06 Jul 2015)


2016 will be a real battle of soc's a hightened one in cpu history of mobile devices and a diversity away from the usual suspects. ..ie QC

posted on 27 Sep 2015, 16:14 6

29. someone12 (Posts: 177; Member since: 28 Aug 2015)


samsung will win again with the M1

posted on 27 Sep 2015, 21:51 3

39. wargreymon (Posts: 679; Member since: 05 Nov 2013)


Please stop with being a fanboy. Whoever shows the best chip wins. Just randomly shouting a name because you like that company or is paid by them is just stupid, so put up some facts or shut up.

posted on 27 Sep 2015, 21:53 6

40. Nathillien (Posts: 208; Member since: 05 Oct 2014)


More interesting:
Smapdragon 820 - made by Samsung
A9 - made by Samsung
Exynos 7, 8 - made by Samsung

posted on 28 Sep 2015, 08:15

51. may_czos (Posts: 718; Member since: 22 Nov 2014)


Most A9's will be made by TSMC.

posted on 27 Sep 2015, 20:32 2

33. cheetah2k (Posts: 1527; Member since: 16 Jan 2011)


Interesting that the 3rd pic shows the fab is 16nm TSMC... I thought Sammy was fabricating these for Qualcomm??

Does this mean we'll be stuck with the inferior TSMC 16nm fab again?

Pass

posted on 27 Sep 2015, 22:25 2

43. iushnt (Posts: 1579; Member since: 06 Feb 2013)


The third pic is for huawei Kirin 850.

posted on 28 Sep 2015, 08:15

52. may_czos (Posts: 718; Member since: 22 Nov 2014)


TSMC's 16nm process is eual if not better than Samsung's 14nm.

posted on 28 Sep 2015, 11:33 1

55. rcl4444 (Posts: 40; Member since: 06 Oct 2012)


Really can't believe I'm about to give this comment the credit of a response, but....

Please, PLEASE tell us how TSMC's 16nm can possibly be equal or better than Samsung's 14nm?

Not withstanding the simple microarchecture mathematics of a smaller die size process = more processors/less power consumption, and that Samsung's FinFET manufacturing process has been well recognised as the most cutting edge at the moment.

So again I ask please explain to us all?

posted on 28 Sep 2015, 12:41

56. may_czos (Posts: 718; Member since: 22 Nov 2014)


14nm is just a number. Actual transistor density is pretty much identical in 14nm Samsung's, 16nm TSMC's and 20nm Intel's processes. They're virtually identical in terms of efficiency.
More - TSMC's process has better yield and it's one of the reasons why most of the Apple's A9 processors will be made by TSMC.

posted on 28 Sep 2015, 13:38 1

58. Techist (Posts: 234; Member since: 27 Jan 2015)


Incorrect. Samsung's is die size for the 14nm is slightly smaller than TSMC's 16nm, and they were first out of the gate. Intel has them both beat on those two fronts, however:

http://www.extremetech.com/extreme/195897-samsung-and-apple-team-up-on-14nm-chips-expected-in-2015

posted on 27 Sep 2015, 13:05 10

4. RebelwithoutaClue (Posts: 2763; Member since: 05 Apr 2013)


The A8 wasn't that flawless when it comes to sustained performance. ArsTechnica did a test and it did throttle quite soon (http://arstechnica.com/apple/2014/09/iphone-6-and​-6-plus-in-deep-with-apples-thinnest-phones/3/)

posted on 27 Sep 2015, 13:40 8

10. Tizo101 (Posts: 394; Member since: 05 Jun 2015)


thank you, now the iArmy can stop making noise about the only perfect thing on the iPhone.

posted on 27 Sep 2015, 13:58 1

15. vincelongman (Posts: 4164; Member since: 10 Feb 2013)


This is why I say Geekbench 3 isn't a good benchmark (although its still one of the best mobile benchmarks)

Geekbench 3 take my 2 year old Nexus 5 about 40 secs to finish
I can run it about 10 before thermal throttling starts kicking in

posted on 27 Sep 2015, 14:05 1

17. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


http://www.phonearena.com/news/Snapdragon-vs-Exynos-vs-Apple-vs-Atom-throttling-stats-confirm-810-is-the-worst-A8-is-top_id70945

i never said it was flawless, i said it was quite good at sustaining performance near its peak performance. A8 doesnt throttle any more then 15-20% from its maximum performance both on GPU and CPU side.also Can run GFXbench/Ice storm almost indefitnetly on the 6 plus/6(a8 devices) and the frame rate from when it started the test till the device dies doesnt deviate much from original FPS when the device started. the a8 was one of the better chips that can deliver sustained performance vs competition although a decent amount of that was contributed to running at lower screen resolutions, other half is Apple doing some optimizing/engineering towards sustained performance and not reaching for higher peak performance but then dropping fast, its a balance

posted on 27 Sep 2015, 14:22 2

22. RebelwithoutaClue (Posts: 2763; Member since: 05 Apr 2013)


What also makes a difference is that the speed of the A8 isn't totally depended on GHZ, so a 15% drop in frequency has less impact. But I wouldn't want to call 85% near peak performance. Still it is most definitely one of the best soc's when it comes to throttling (which was never my point anyway)

posted on 27 Sep 2015, 14:37

23. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


Why cant you call it around 85% of its peak performance? The A8 GPU and cpu do not deviate more then 15-20% of its peak performance in benchmarks and real world use

You can Run ice storm unlimited and other intense graphical benchmarks over and over and over and the FPS from when it started is not much differeny then after many multiple countless consecutive ran benchmarks to really put the SoC through heavy testing. A8 gpu doesnt give less then 8-10 fps then non throttled fps rating even after consecutive testing. I would say it is easily safe to say the A8 can maintain about 85% of its peak performance almost seemingly indefitnetly (until phone dies obviously)

But the A8 is old news now anyway, i was jus trying to make my point. I hope OEM's in the future stay with perfect balance of sustainability and also peak because both are very important in mobile world in my opinion as a perfect balance of both brings solid consistent performance to a device

posted on 27 Sep 2015, 20:44 1

35. cheetah2k (Posts: 1527; Member since: 16 Jan 2011)


The A8 is only dual core too. much less heat, more time at full speed, less time throttling... Also means each core has to do more work too - but on an iPhone the CPU can be lazy because the OS & GUI is very basic, not overly graphic intensive and only pushing 750P pixels to the screen...

Put an A8 on a 2K screen and lets see how well it handles..

posted on 27 Sep 2015, 22:23

42. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


Thats nonsense. The Power VR GX 6450 gpu in the A8 is on par with Adreno 420 in performance roughly, A8 can handle 2k just fine, see the iPad mini 4.

posted on 29 Sep 2015, 17:11

60. SebastianFox (Posts: 16; Member since: 17 Oct 2014)


except that a tablet SOC is not the same as a smartphone SOC,Higher TDP and higher wattage,If you want to talk like this Mali-T760 MP4 can handle 2.5K display*(It is not 2k,1080p is 2k,1440p is 2.5K son,MKBHD ruined your minds)
anyway,PowerVR GX6450 clocked at 533 MHz in A8 literally equals PowerVR GX6430 clocked at 600 MHz,Both GPUs share same architecture,Same number of cluster,Same technology but different sizes,One was inside the iPhone 6 backed up by 2 beasty cores,The other one found himself inside Asus ZenFone 2....When you check benchmarks for both,You'll just realize that PowerVR GX6xxx series is average..
No joke,But even Mali T760 MP4 is equal or even superior to GX6450,That's if we are talking about FP32 performance,Texture rate and mTrans/s/core,You won't even dare to put any PowerVR GPU with 6 Rogue clusters or less clocked at that low frequency...

posted on 27 Sep 2015, 13:15 13

6. TyrionLannister (unregistered)


Mobile SoCs aren't supposed to sustain peak performance. And no, the A8 and A9 are better than other SoCs but even they can't sustain their performances.

There is simply no heat sink in a phone, making sustained performance impossible. Even intel's mobile chip, the core M shows burst performance for a short time and comes back to sustained loads.

Most of the workload in a smartphone/tablet OS is of a bursty nature. Like opening an app or a webpage, taking a quick snap, firing flashlight, or doing almost anything.

Staying on top frequencies will drain your battery in a couple hours tops. Yes, the A8 and A9 can keep about 90% of their peak performance, Exynos 7420 can keep about 70% and these are actually more than enough for a smooth bursty workload.

Games are the only sustained workload I can think of. But even in games, there are only a few moments where there is too much happening on the screen and needs the full GPU power. Most of the game is not so demanding.

We don't want a super-throttling chip like the 810, which can't even keep 50% of it's performance. But in mobile world, the SoCs will always be of this kind, as the usage is bursty.

The situation is way different in desktop world though. They are made for constant work and have heat sinks/fans to keep them running. And also they don't have to worry about battery. Desktops can be used for hours of heavy gaming, video editing and such workloads that require consistent performance. And hence, the SoC's offer that.

There is a reason an Exynos 7420 can score as high a benchmark score as an intel i5, but the real performance won't even be close. As the i5 is built to sustain it and the exynos isn't. Both have their specific purposes and they fulfil them very well.

posted on 27 Sep 2015, 13:21 1

8. Podrick (Posts: 510; Member since: 19 Aug 2015)


Great post! Always get to learn a lot from you and some other guys.

posted on 27 Sep 2015, 13:34

9. DoggyDangerous (Posts: 784; Member since: 28 Aug 2015)


but despite every leak, we are keep getting devices with 808 and 810. I am pretty much bored with 820 rumors. When it will really hit the market? All of my favorite devices will be released that time and I cannot wait till October 2016 to get the 820 up grade of lumia 950 and LG G4 pro

posted on 27 Sep 2015, 13:40 2

11. TyrionLannister (unregistered)


You will get SD820 devices will be available in Q1 2016. I guess the first phone with SD 820 will be Xiaomi MI5 or LG G flex 3/G4 pro.

posted on 27 Sep 2015, 14:11 1

18. TylerGrunter (Posts: 1514; Member since: 16 Feb 2012)


I would not be surprised is someone manages to pull a phone with it by years end.
Probably in Qualcomm are working double hours to have it ready as soon as possible (meaning some phones before the holidays)

posted on 27 Sep 2015, 14:39

24. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


Yeah i would not want QC even releasing the 820 this year, i want them to perfect it as much as they can

posted on 27 Sep 2015, 13:56

14. TylerGrunter (Posts: 1514; Member since: 16 Feb 2012)


Just one correction, the throttling happens exactly the same in desktops.
What is different is how the CPUs are advertised there: both Intel and AMD normally provide the base clock (sustaned load) and then provide a "turbo" clock for the bursty loads. But max clocks can not be maintained either in the desktop.
In mobile they only provide the equivalent of the "turbo" clock, that of course it can not be maintained for long periods.
So the behaviour is the same, what is different is the marketing only.

posted on 27 Sep 2015, 15:00 1

26. TyrionLannister (unregistered)


Well if you're considering that, it's true. But Intel markets the case clock too. And only one core can do turbo.

posted on 27 Sep 2015, 21:44

38. perry1234 (Posts: 151; Member since: 14 Aug 2012)


I miss your blog man. Get back to writing / updating soon :P

posted on 28 Sep 2015, 05:24

49. TylerGrunter (Posts: 1514; Member since: 16 Feb 2012)


That´s also incorrect, there are sometimes two different turbo speeds, one for a single core (higher) and one for multiple cores. They only advertize the single core one though. Probably because it is the higher one.

posted on 27 Sep 2015, 14:12

19. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


Your right Tyrion, mobile SoC's arent meant to hold sustained performance for that long as most mobile workloads are short and burstry and are race to zero workloads. but, we are seeing a trend now with mobile SoC's and how OEM's are getting better at sustained performance with mobile SoCs, within past year or so we are starting to see this.

in my opinion though there always has to be a balance of sustainability and peak performance in a mobile chip though. what good would a device be with a SoC that has 2500 single core score and 7000 multi core score but after 30 seconds it starts to heat up a good amount and then peak performance drops like a rock.

for most people, they do bursty quick workloads, for some, they run games that can push the envelope and vice versa

i would say staying within 15-20% of peak performance is very good and extremely respectable, A8/a9 obviously cant sustain peak performance forever but they do offer very good sustained performance.

posted on 27 Sep 2015, 14:45

25. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


sorry i made a error, *race to sleep, not race to zero lol

posted on 27 Sep 2015, 20:41

34. zunaidahmed (Posts: 703; Member since: 24 Dec 2011)


Hmmm, as much as I am concerned, apple a9x could actually break the single core 2800 (a9 already does 2500 for core single core) multi core 8000 barrier, or even 10000 for multi core (if it has a quad core that is). 8000 is 100% do-able but 10000 is just my speculation. But hey, who say's not to have hope? Let's see what apple puts on the table on November.

posted on 29 Sep 2015, 17:39

62. SebastianFox (Posts: 16; Member since: 17 Oct 2014)


2800 Singlecore??maybe,8000 Multicore???no,I even doubt that A9x will reach 7000 multicore,A8X was not double performance as A8,It was 70% better not 100%

posted on 27 Sep 2015, 16:07

28. zeeBomb (Posts: 1875; Member since: 14 Aug 2014)


The lannisters, have sent their regards.

posted on 27 Sep 2015, 23:23 1

44. phonefanboy (Posts: 11; Member since: 06 Oct 2011)


Great post, without the fanboy dribble.

posted on 28 Sep 2015, 13:41

59. Techist (Posts: 234; Member since: 27 Jan 2015)


Well said!!

posted on 27 Sep 2015, 23:25 1

45. Macready (Posts: 874; Member since: 08 Dec 2014)


In reality, sustained PEAK performance is really academic, since:
A) no phone game or app requires full peak performance for long consecutive periods
B) the phone battery drains in no time

Sustained performance (in terms of thermal properties and drain) over a long period at half or little over half of peak will be more interesting for heavy gaming and things like editing.

posted on 27 Sep 2015, 13:00 1

2. patrickf3 (Posts: 32; Member since: 11 Feb 2015)


I just hope it can keep its cool.

posted on 27 Sep 2015, 13:07 1

5. Podrick (Posts: 510; Member since: 19 Aug 2015)


Nice. They can't afford to screw up this year too.

posted on 27 Sep 2015, 13:21 1

7. Ghost04 (Posts: 333; Member since: 03 May 2014)


The chip makers should concentrate more on power efficiency .
So, we can have 3-4 days of battery life instead of just having more Powerful CPU ....
A midrange phone is powerful phone enough to run most functions
I am really looking for phones with great and long battery life over the the ones with unnecessary powerful CPU.

posted on 27 Sep 2015, 20:22

32. Gemmol (Posts: 788; Member since: 09 Nov 2011)


Get windows phone, battery life is always better since the system run the same chips that android use but better.....look at htc one m8 for android and windows, same exact phone but different operating system and the windows version had a way longer battery life n other improvements

posted on 27 Sep 2015, 13:43 1

12. Derekjeter (Posts: 816; Member since: 27 Oct 2011)


Why are bench marks still important? They have nothing to do with real life usage. BS articles.

posted on 28 Sep 2015, 00:29

47. yoosufmuneer (Posts: 1498; Member since: 14 Feb 2015)


Why click on the article then?

posted on 27 Sep 2015, 13:51 2

13. carlemillward (unregistered)


Power efficiency, heat dissipation and lag-free performance is what I am interested in.

posted on 27 Sep 2015, 14:03 1

16. Carl3000 (Posts: 153; Member since: 11 Oct 2014)


820 won't overheat badly. 810 overheat terribly because big.LITTLE architecture. Constant switching of voltages from big and little cores too often was the reason. This guy did a YouTube video explaining it exactly and talked about how it affected battery drain, like why 2014 flagships last longer than their successor's using the 810.

posted on 27 Sep 2015, 14:13 6

20. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


Exynos 5433 at 20nm node is same thing as Snapdragon 810 at 20nm with same big.little setup and the 5433 doesnt have any overheating issues like the snapdragon 810. it isnt big.little, it is Qualcomm's implementation of it with their beefy GPU that generates alot of heat.

the 810 is not a good example of big.little, exynos 5433 and 7420 are much better examples of executing big.little right

posted on 28 Sep 2015, 08:17

53. may_czos (Posts: 718; Member since: 22 Nov 2014)


5433 doesn't have power hungry A57 cores that weren't meant for mobile devices.

posted on 28 Sep 2015, 12:46 1

57. miket1737 (Posts: 2383; Member since: 17 Mar 2013)


http://www.anandtech.com/show/8537/samsungs-exynos-5433-is-an-a57a53-arm-soc

it has 4+4 a57/a53 cores.

posted on 27 Sep 2015, 14:18 1

21. hound.master (Posts: 1044; Member since: 27 Feb 2015)


I don't think if Kirin and mediatek are even comparable since their gpu sucks.

posted on 27 Sep 2015, 21:59

41. wargreymon (Posts: 679; Member since: 05 Nov 2013)


The Kirin 950 has a GPU worth mentioning though, and could be a real contender.

posted on 28 Sep 2015, 08:18

54. may_czos (Posts: 718; Member since: 22 Nov 2014)


It couldn't. It has the same Mali-T880 MP4 GPU as Mediatek X20. Snapdragon 800 level.

posted on 29 Sep 2015, 17:15

61. SebastianFox (Posts: 16; Member since: 17 Oct 2014)


you mean that Adreno 330 = Mali T880 mp4,Go drunk,You are home

posted on 27 Sep 2015, 15:13 6

27. alex3run (Posts: 710; Member since: 18 May 2014)


Exynos 8890 is still unreachable.

posted on 27 Sep 2015, 17:28 1

30. Grkm3 (Posts: 5; Member since: 27 Sep 2015)


I can break 6500 with my note 5 in geekbench

posted on 28 Sep 2015, 07:06

50. TyrionLannister (unregistered)


Did you overclock it. As it's simply impossible to do at 2.1 GHz.

posted on 29 Sep 2015, 17:48

63. SebastianFox (Posts: 16; Member since: 17 Oct 2014)


It is simply possible...You can't oc latest big.LITTLEs,You can spend years searching in xda and you won't even get N4 Exynos OC,That is still possible,With the right kenrel and maintaining that frequency by using Performance cpu scheduler that can sustain performance with...CPU damage,This only shows how powerful big.LITTLE is..Nothing more nothing less

posted on 01 Oct 2015, 08:03

65. Grkm3 (Posts: 5; Member since: 27 Sep 2015)


You do know there are unlocked bootloder note 5s right? The kernel im using allows overclocking and underclocking and full voltage control.go spend 5 min in the note 5 xda development thread and look at all the kernels to oc

posted on 01 Oct 2015, 07:59

64. Grkm3 (Posts: 5; Member since: 27 Sep 2015)


Yes sky kernel unlocks up to 2 4Ghz i also got 450mb/read performance on nand.

Currently world record holder in androbench lol

Want to comment? Please login or register.

Latest stories