Samsung Galaxy S7: Snapdragon 820 vs Exynos 8890 flavors compared
1. tech2 (Posts: 3404; Member since: 26 Oct 2012)
Anyone know how the two performs in battery department ? Also, which of the two has a better GPU?
19. tyrionTheWise (unregistered)
Adreno has a boost clock, it can't sustain the performance for more than a minute or so. The boost clock will result in better benchmarks though.
41. Shocky (unregistered)
Nice, Qualcomm are basically cheating then.
45. tyrionTheWise (unregistered)
Not exactly. It will give you better performance in bursty GPU workload. Gaming doesn't qualify but suppose you're rendering a webGL based webpage, or an image via OpenGL. In those cases, this burst frequency will come in. Also, benchmarks will be higher. Apple A9 is also similar. It can't sustain performance for long.
Many desktop manufacturers do that, and it's a good thing as it helps in bursty workload. All Nvidia GPUs have a boost clock.
However, the adreno 530 is definitely not a better GPU than the mali T880MP12 for gaming. Gaming is not bursty.
48. Shocky (unregistered)
Still sounds like cheating to me, I don't see how a brief increase on clockspeeds will make any real world difference in your examples.
Benchmarks on the other hand will get a nice boost from this and that's unfortunate.
49. tyrionTheWise (unregistered)
Rendering an image is a small process, a faster GPU for a short amount of time will help in that. It's the 'race to finish' type of scenario.
52. Shocky (unregistered)
Images render pretty damn quickly on my Galaxy S6, would a small boost in GPU clockspeds really make it notability faster? I don't think it would.
I don't think Qualcomm believe that either.
57. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
It's most definitely not cheating. Pushing a GPU or CPU to a higher speed, even if just for a short time until thermals limit it, is a very standard feature. Nvidia has been doing that for years on their GPUs with GPU Boost. It's a feature that's so standard that I wouldn't be surprised if Samsung has this feature too, and if they don't, it's their own fault for not fully utilizing the chip.
58. Shocky (unregistered)
If your referring to desktop GPU's then that's not how GPU boost works, it's a sustained boost, not temporary.
I've never read anything refering to this on their mobile parts.
No Samsung does not have this feature, they all throttle eventually, however boosting the GPU to a frequency it can't maintain for more than a minute or more is cheating.
Based on a continuous looping test Exynos 8890 maintained near peak performance for around 15 runs of an intensive benchmark before throttling kicked in, that's pretty good and shows no evidence of what you're suggesting.
61. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
Yes, that is how it works. Boosts will be sustained until temperatures rise to a point where it is unsafe for the GPU to operate, then they will throttle to keep the GPU from damage. This is no different to what Qualcomm is doing, except phones reach that point much more quickly than laptops or desktops because they don't have active cooling. The GPU rises to a frequency past the rated clock speed, then goes back down when temperatures get too high. If cooling were better on phones, it would sustain that frequency, just as Nvidia GPUs with better coolers have better performance than similar GPUs with worse coolers.
If Samsung doesn't have that feature, then that's their own fault. It's a very effective way to boost performance for short GPU intensive tasks. There's nothing wrong with it.
62. DoggyDangerous (Posts: 783; Member since: 28 Aug 2015)
I prefer Qcom over others bcoz of wide developers support and custom ROMs. So, small throttling is acceptable to me.
65. Shocky (unregistered)
This is a little different to your standard throttling.
63. Shocky (unregistered)
It's completely different, the stock coolers on nvidia cards can sustain this performance, they generally run well within temp limits even with gpu boost, most of the cards are from 3rd parties with improved coolers anyway.
Do either of my GTX980 cards throttle? the answer is NO, I have all the stats displayed while a play, from temps, frequency and gpu usage I would know if they were throttling.
How can you even compare that with that Qualcomm are doing? You don't know what you're talking about. Go away.
In a laptop? sure, because heat can be an issue if the manufacturers were lazy, but desktop cards? No.
66. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
Reference cards can sustain performance above stock clocks, but can't get as high as non-reference designs due to temperature limits. Here's a comparison between various 980Ti models from different manufacturers with different coolers;
Notice how every single card gets better scores compared to the reference design. This is because they all have better coolers. The better the cooler, the more performance you get. That is how it works.
Qualcomm is using the same thing in their GPUs. However, it is less effective because phones don't have active cooling, and they can't sustain their performance for very long. If you compared a tablet and a phone running the 820, the tablet would score higher because it has more surface area to spread the heat, keeping the GPU cooler and sustaining that boost speed.
67. Shocky (unregistered)
Even when the stock cards reach 80C the cards don't drop to stock clocks, it just reduces the boost.
Non-reference usually don't have that issue due to better cooling.
It's not the same as what Qualcomm are doing as this can be sustained and partially sustained in worst cases, excluding user error.
What Qualcomm are doing can NEVER be sustained, I don't understand why you're even comparing the two. It's ridiculous.
/shakes head in disbelieve.
69. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
Of course they wouldn't go below stock at 80 degrees, Nvidia usually sets TJ Max to 92 degrees, that's when they go below stock. But the point is that the lower the temps, the higher the boost. Qualcomm is doing the same thing, it's just a worse case scenario because it's passively cooled. The 820 goes past stock clocks for a few minutes, then back down to stock when the temperature gets too high. It COULD be sustained longer with more surface area, but on a phone it can only be done for a few minutes.
This is NOT cheating. It's just boosting the GPU when temperatures allow it, which only happens to be a couple minutes for something as small as a phone. Completely fair game.
71. Shocky (unregistered)
80C is the target temperate for these cards and they rarely go beyond that, so generally they don't throttle stay above stock clocks.
The problem with Qualcomm doing this in a phone is the only time it will be able to acheive this is for a very short time after you start a game/benchmark and then it will immediately drop, it wouldn't surprise me is this boost causes the SoC to overheat and throttle a lot faster than it usually would.
This also invalidates benchmark results and gives users a very false image of what hardware can actually do.
75. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
It doesn't create a very false image. Any self-respecting reporter would run the test multiple times to get an average result. This negates the effects of a single boosted benchmark. PhoneArena did this, "To correct for random flukes, all tests were ran three times, with the scores that you see being the average from the runs."
Even with three consecutive runs, the 820 still managed to beat the 8890 in a few benchmarks. This can mean two things. Either Qualcomm's boosting is actually effective and can be sustained through long intensive tasks, or even at stock speeds it can still beat the Exynos. Either way, the same message is being given to the reader.
78. Shocky (unregistered)
You're giving them fair to much credit if you think that's true.
Do you think Phonearena care if the sources they link to provide results from a single or over multiple runs? of course they don't.
82. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
So your only argument at this point is the validity of the news source. I think my point has been made then.
98. Shocky (unregistered)
You've completely missed the point.
94. mitchytan92 (Posts: 82; Member since: 02 Mar 2015)
I don't think by "ran three times", they meant "ran 3 consecutive runs". I think they just ran and wait for the phone to cool down and ran again then taking the average score. It is unbelievable how both chipset can withstand a score of 130k+ if it is 3 consecutive runs.
70. Shocky (unregistered)
Non-reference cards are clocked higher dumbass, you can overclock the reference cards and get the same results.
73. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
Listed clock speeds don't matter at all. Almost every GPU regardless of cooler goes past their listed clocks with GPU Boost, even reference. However, the cards with better coolers go much higher than the reference.
80. Shocky (unregistered)
You really think a stock card will suddenly boost to speeds matching a non-reference card if you install a 3rd party cooler on there? No, it won't.
It's a bit more complicated than that.
83. Berzerk000 (Posts: 4275; Member since: 26 Jun 2011)
Actually, yes. That does happen. If you put a water cooler on a graphics card with the NZXT G10 or Corsair HG10 bracket, or do a custom loop, it's performance will increase significantly. Maybe not to the point of non-reference cards because those typically have better power delivery as well, but it will be a noticeable improvement.
104. zunaidahmed (Posts: 703; Member since: 24 Dec 2011)
Yes, it would boost a little higher, but no, it still can't match a lot of non reference cards. Non reference cards usually have a custom VBIOS, which has higher power limit than reference card, some even have better power design such as better power phases, more power supply pins, which all adds up to higher boost clock and even higher overclock(if you do overclock)
68. Shocky (unregistered)
One is sustainable and represents actual performance, the other has no hope in ever being sustainable.
Even in a tablet it won't be a able to sustain this for long enough , it may take a little longer to throttle but that's it.
91. XileForce (Posts: 4; Member since: 11 Mar 2016)
samsung had this feature on the exynos 7420. It would use 772mhz only under certain situations, such as not fully loaded, in order to reduce thermals. I would be very surprised if the 8890 didn't have a similar feature.
source: Galaxy S6 Kernel Developement
103. zunaidahmed (Posts: 703; Member since: 24 Dec 2011)
Actually, a better example would be Intel's mobile chipset, not desktop class GPU. They have higher boost clock but cannot maintain it for long, it really helps when opening a website or simple short photoshop rendering. You don't need the max power at all situation, but it's always there when you need it the most. BTW, iPhone can actually maintain the boost way longer than s6, it's been tested by a lot of reviewers, including Linus (linustechtip). Anyway, boosting clock speed is normal in everyway, the reason others are arguing with you, is because normally, nvidia factory clocks the GPU very low, so it can maintain boost speeds during the whole gaming session, try overclocking to the max, and then see how it boost, I have tried it, it keeps on jumping up and down which proves you are totally correct, no need to get into an argument with them. Desktop hardware has so much headroom, so that boost is always there, but that not the same for laptop (mobile) chipsets.
79. Shocky (unregistered)
That's the usual reason trolls change their name.
88. tyrionTheWise (unregistered)
No. I deleted my account for a week due to exams. I was stupid and didn't know it can't be recovered. So I had to make a new account as it still shows my name is in use. Went would I be banned anyway? I never even got a warning.
100. rubyonrails3 (Posts: 251; Member since: 01 Oct 2014)
My personal experience with Exynos 8890 feels same like SD820 as you say. my battery was 25% and phone was bit warm and when I played Need for speed No Limit it was like I'm playing on bad phone so many frame drops, but then I left phone and charged fully in that time I also Installed Game on Nexus 6p, iPhone 6s plus to check how it perform. iPhone was so smooth, same was Nexus 6p and now my S7 edge also performed similar.. so I suppose Exynos 8890 throttled cuz of heat now it was cool so it worked as expected. Beside If you run benchmark(Geekbench 3) Multiple times those score keeps dropping.
108. RisinT96 (Posts: 4; Member since: 30 Sep 2015)
Except most of the nvidia gpus can usually sustain the boost clock and not just boost for short periods of time.
Also like the Intel turbo boost.
95. mitchytan92 (Posts: 82; Member since: 02 Mar 2015)
I kind of hope that benchmark apps do take thermal throttling into account as that we can know the actual performance. Like maybe running 2 times as a warming up rounds and finally the actual test.
17. alex3run (Posts: 710; Member since: 18 May 2014)
Exynos has better GPU. It won't beat records in synthetics but will beat SD820 in real apps and games.
93. stanislav (Posts: 136; Member since: 22 Sep 2014)
Its really hard to belive it when the exynos has half the GFLOPS of Adreno 530
101. rubyonrails3 (Posts: 251; Member since: 01 Oct 2014)
Exynos have better CPU, SD820 have better GPU
21. Arch_Fiend (Posts: 2274; Member since: 03 Oct 2015)
Snapdragon for days.http://www.anandtech.com/show/
10120/the-samsung-galaxy-s7-review/4They only test the Sd820 version but the GPU results she help you understand why I say Sd820 has the better GPU.
They'll be testing the E8890 when given the chance so when Anandtech decides which soc/gpu is better we'll have a definitive answer.
24. Back_from_beyond (Posts: 56; Member since: 04 Sep 2015)
A Dutch website that tested the Exynos version said batterylife had improved substantially compared to the S6, Phonearena suggested the SD820's batterylife however wasnt that great. A proper comparison is needed, but sounds like the Exynos version might have the better batterylife.
29. darthmaul (Posts: 25; Member since: 04 Feb 2015)
Phonearena and battery life tests...mine is Exynos and lasted after the first charge for 20 hours, with almost 5 hours of screen on time, data on.
33. rd_nest (Posts: 1593; Member since: 06 Jun 2010)
32. rd_nest (Posts: 1593; Member since: 06 Jun 2010)
Check this image. GFXbench continuous test. For 1st time, Exynos showed better performance for first 10 iterations.
Now wait for Anandtech to do more thorough test.
42. Shocky (unregistered)
Faster for 15 minutes and then the throttling kicked in, I wonder how low Exynos 8890 would gone with another 20 minutes.
47. rd_nest (Posts: 1593; Member since: 06 Jun 2010)
It's not minutes, it's iterations.
And this is because metal frame in iPhone allows to dissipate heat more easily than glass.
Point being this is first time it happened that Exynos had better control for 'considerable amount of time' than A9.
50. Shocky (unregistered)
FYI - the test they're running lasts almost 1 minute exactly.
36. Punchy506 (Posts: 70; Member since: 10 Jan 2016)
The sd820 comes with very high android system and Android OS battery drain. The two together represent ~45% in battery stats. Check xda.
2. hafini_27 (Posts: 878; Member since: 31 Oct 2013)
Great to see Qualcomm gets back on track.
4. Chuck007 (Posts: 1054; Member since: 02 Mar 2014)
The Exynos 8890 is obviously the better optimized chip here.
Still don't mind having to stick with the Snapdragon 820 for my Galaxy S7 Edge though. Fingers crossed it doesn't overheat.
5. JMartin22 (Posts: 1954; Member since: 30 Apr 2013)
600 mms is too slow. Use the stock Browser on both versions, not Chrome, stock browser is more optimized than Chrome.
12. sorcio46 (Posts: 435; Member since: 27 Jul 2011)
Try with Snapdragon Browser on the S820, you will have better results :D
13. JMartin22 (Posts: 1954; Member since: 30 Apr 2013)
This is the first time I've ever seen performance year over year regress in web browsing performance with stock browser builds. I thought you just used Chrome on the SD820.
26. true1984 (Posts: 822; Member since: 23 May 2012)
its due to having 4 less cores. the browsers work better with multicore processor. at least thats what i've gathered
51. tyrionTheWise (unregistered)
Sunspider is a single core benchmark. This just suggests Samsung didn't optimize the SD 820 variant.