A ton of Snapdragon 820 details leak: 14nm, quad Hydra CPUs with 35% better performance, new Adreno 530 GPU

A ton of Snapdragon 820 details leak: 14nm, quad Hydra CPUs with 35% better performance, new Adreno 530 GPU
Chinese analyst, Pan Jiutang, took to Weibo (China's Twitter equivalent) earlier today, spilling the beans on the supposedly upcoming Qualcomm Snapdragon 820, which is expected to be announced on August 11th in Los Angeles. Dubbed MSM8996, Jiutang seems to have gotten his hands on internal slides outlining the Snapdragon 820, and the intel grab is quite meaty here.

We start off with the obvious: Qualcomm will be moving to a 14nm (FinFet) node with the Snapdragon 820, matching Samsung's Exynos 7420. Unlike the Snapdragon 810, however, the Snapdragon 820 will go back to "just" four custom, 64-bit CPUs — referred to as Hydra, not Kryo here — supposedly ensuring a "~35% performance improvement". On the graphics end, we're told to expect an Adreno 530 GPU and 40% performance improvement, along with over 30% better power efficiency. 

The SD820 will also provide support for dual-channel, LPDDR4 memory interfaces (up to 1866MHz), up to 28-megapixel cameras with programmable DSPs that will allow manufacturers to further differentiate their smartphones, 4K video at 60 frames per second, and even a dedicated, low-power sensor that will enable "emerging, always-on use-cases"

Finally, the source revealed a timeline slide, suggesting that the Snapdragon 820, as Qualcomm promised, is expected to see the light of day by the end of 2015. From what we're told, however, it won't be until 2016's Q1 that we see any super-phones flaunting the new super-processor. More as we get it.


source: Weibo via g for games

FEATURED VIDEO

81 Comments

1. Japultra

Posts: 130; Member since: May 16, 2014

What I want to know is how does this compare to the Exynos 7420?

7. aesthetics

Posts: 128; Member since: Oct 02, 2014

Equal speed on cpu side, but with more powerful gpu

11. alex3run

Posts: 715; Member since: May 18, 2014

You'd better wait for gaming tests before making such a conclusion. Adreno 530 either can give no real life advantage over Adreno 430 or can significally boost the performance especially for high-res screens. I'll wait.

43. neela_akaash

Posts: 1239; Member since: Aug 05, 2014

Heard that Xiaomi Mi 5 will be the first to use SD820. Wow it will be a hell of a device under $400..

61. dazed1

Posts: 771; Member since: Jul 28, 2015

LOL under 400? keep dreaming.

62. neela_akaash

Posts: 1239; Member since: Aug 05, 2014

Don't worry! It will be under $400.

72. marorun

Posts: 5029; Member since: Mar 30, 2015

Qualcomm custom core where always at least 30% faster than stock arm core. Big little architecture is a faillure because of the voltage issue when switching from one cluster to another ( this is what make the overheating issue btw and even the S6 suffer from this.. in fact if its would only use the higher end core ( and only 4 of them ) its would get even better performance but would loose battery life. Qualcomm Custom core where always better than stock core for battery usage as well. As for the GPU Adreno 530 will improve gaming thats for sure. But will its only improve on the framerate or on the quality thats not the chip itself but the dev thats need to use it fully.

35. GreekGeek

Posts: 1276; Member since: Mar 22, 2014

lol Where did you get that absurd conclusion?

45. arch_angel

Posts: 1651; Member since: Feb 20, 2015

haha i know right.

37. medicci37

Posts: 1361; Member since: Nov 19, 2011

Slightly more powerful gpu. Seems like there isn't going to be anything much better than the 7420 in the near future. So much for progress

39. Ruturaj

Posts: 1484; Member since: Oct 16, 2014

Not until Samsung makes chip with A72 cores and T860/880 GPU. Qualcomms 620 will have A72 but in 620 which means DDR3, slow GPU, built on 28nm or at best 20nm process. Mediatek is busy trying octa and deca core. Kirin use fewer gpu cores. So mostly Samsung will come out with better chip.

74. marorun

Posts: 5029; Member since: Mar 30, 2015

Exynos 5 serie was also supposed better on paper but got beat to pulp by same generation Qualcomm custom core chip. You dont seem to understand this so i leave you in your dream world. PS: the only high end SOC qualcomm failed is the only one they used non custom core. Every other where superior to other SOC of same gen for android.

73. marorun

Posts: 5029; Member since: Mar 30, 2015

Adreno 430 when not been trottled is a lots faster than the Mali crap in the exynos. Did you know the mali keep on trottling as well? ( less because of the 14 nm build ) A non trottled Adreno GPU will always be faster than a not trottled Mali of the same gen and same end ( high end vs high end ect ) Also the Adreno has better texture filtering and shader so if dev use them correctly they can make much better looking game ( thats wont work on mali or will work with lower quality.

25. vincelongman

Posts: 5560; Member since: Feb 10, 2013

Wait for actual phones and reviews Especially after Qualcomm claims of the 810 BTW this leak is fake IMO Qualcomm made it clear their custom cores were going to be called Kryo, not Hydra Also I'd expect Qualcomm to advertise the 820's support for their Zeroth Platformhttp://www.anandtech.com/show/9028/qualcomm-announces-snapdragon-820-and-zeroth-platform

28. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

Hydra may be the internal name of the cores while Kyro the advertize one. The move to 14nm should help a lot with the performance/overheating. The only thing that really worries me is the 35% performance improvement of the cores... If it's just that it means they are roughly A57 equivalents. So they went again for a medium core that would need high frequencies to perform and that could backfire.

33. vincelongman

Posts: 5560; Member since: Feb 10, 2013

Oh that could be true too Hopefully the improvements are bigger Qualcomm really need to set up their game Samsung using their own chips Intel/MediaTek dominating low, mid and upper mid range Nvidia dominating graphics/Android tablets I'd except another by year for Qualcomm if this rumor and the low gains are true

34. Berzerk000

Posts: 4275; Member since: Jun 26, 2011

How would they be A57 equivalents? It's a 35% bump compared to the 810, which uses A57. The 810 was behind the Exynos 7420 because of the throttling issues, but Hydra/Kyro would put Qualcomm ahead of Samsung again. Yes it would need higher frequencies, but it worked well for them in the 800 and 805, why not do it again?

36. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

A57 in are also in Exynos 7420, and they are pretty much 30% more powerful than the ones in SD810. So the 35% extra power is coming from MOSTLY from the node shrink, not the better cores == pretty much A57 equivalents. Why not do it again? I'm not saying they should not, just that it can backfire. The smaller the node you use the greater the leaking currents, the greater the leaking currents the more power a chip needs at high frequencies == it can backfire if the node doesn't play very well with high frequencies. FinFET helps to avoid the leaking currents, but it's not magic either ;) And the big issue about that approach is that Kyro will be playing against A72 cores, which have been already benchmarked and provide quite an improvement over the A57 cores.

38. j2001m

Posts: 3060; Member since: Apr 28, 2014

Yes, if Samsung just goes with a small change to the basic a72 cores (i.e. Making it on 14nm and not there own core they have a better cpu

40. Ruturaj

Posts: 1484; Member since: Oct 16, 2014

With new mali GPU cores they will have great GPU too.

46. Berzerk000

Posts: 4275; Member since: Jun 26, 2011

On Geekbench single core the One M9 scores 1209, and the Galaxy S6 scored 1440, that's only a 20% difference. So only 20% of that performance boost came from the shrink, 15% comes from the better cores. Also, keep in mind that it's a 35% performance increase with a 30% power efficiency improvement as well. They could very easily increase the power going to the chip to compensate for leaking and still have an increase in efficiency. Isn't A72 supposed to be out in Q3/Q4 of 2016? I think Qualcomm would already have a new flagship chip out by then. The 820 is supposed to win over OEMs for the first half of 2016 and make up for the mess that was the 810. The new chip in H2 2016 (830?) is supposed to compete with A72. Unless ARM moved up the projected dates for A72. Last time I checked they were aiming for devices with A72 cores to come out in August/September 2016.

57. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

It looks to me that you chose the best score of the One M9 and a regular-low for the Exynos. The HTC One ranges from 800 till 1250 in single core, I just used a 1050 as average:http://browser.primatelabs.com/geekbench3/search?utf8=%E2%9C%93&q=htc+one+m9 But the 30% is just my rough stimation, so no need to discuss it much further. A72 have been already implemented, check the Mediatek MT8173. I don´t know where you got your data from, but they could be used any time. In fact I was expecting the Note 5 to use them. My expectation is they´ll go againt the SD820 in the new Exynos for the galaxy S7 and in the Helio X20-X30 from Mediatek to name two different companies.

58. hound.master

Posts: 1044; Member since: Feb 27, 2015

There's rumors that says GS7 will be power by an SD variet as well as exynos SD 820 maybe. And also i don't believe that helio X20/X30 will be powerful/efficient or even comparable to exynos or SD and would have overheating if all cores be used simultaneously. I don't care about raw CPU power as long as it be efficient gpu and other elements are more important since have real usage and I'm against the octa core cause it makes a lot of great not talking about 810 i mean overall.

79. marorun

Posts: 5029; Member since: Mar 30, 2015

Anyway almost no games or apps use more than 2 core as they are ported from app store and build for dual core cpu. So i would also take a quad core with higher performance per core and better GPU over a load of core thats wont be used. ( beside in benchmark to brag lol )

78. marorun

Posts: 5029; Member since: Mar 30, 2015

SD810 in new Sony Tablet i saw had obver 1400 in single core. ITs the trottling thats lower the score nothing else. and when they say the % improvement they say VS the current A53 stock core used in the SD810 and those core are the same one as used in the new Exynos. Dont base yourself on Benchmark but on fact. anyway you will see custom core always win over stock one.

92. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

I perfectly know that the Sony Tablet gets 1300 in single core (1400 is it's just your delusional dreams):http://browser.primatelabs.com/geekbench3/search?dir=desc&q=Sony+SGP712&sort=score Throttling is part of a SoC performance, as far as I was aware till now, thanks for pointing out it's not: now we can safely put i7 80TDP Watts in smartphones safely because thermal constrains are not to be taken into account [/sarcasm] I hope you are wrong an it's not the improvement over A53 cores as you say, that would make the situation much worse, as Samsung and Nvidia are using A57 already Benchmark are the one of the best ways to measure the IPC in SoCs, so they are the FACT, just saying the name of a core means nothing as far as performance is IPC*frequency, and some cores have lower IPC but made it up with higher frequencies.

66. alex3run

Posts: 715; Member since: May 18, 2014

30% efficiency comes from 14 Nm node, the cores WILL eat more than A57-s because they are going to be pretty much faster (in single core perf).

77. marorun

Posts: 5029; Member since: Mar 30, 2015

You dont understand they will bring 35% improvement over stock A57 core wich make them in fact as good as A72 core whitout the same power drain. But yeah live in your Samsung dream world.

91. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

That's just your guess, you want to understand that 35% is the IPC improvement. I understand it different> 35% is the TOTAL performance improvement over the SD810, that includes node and frequency gains too. Because that's what the slide says: performance improvement, not "instructions per cycle improvements" You want to keep hoping and dreaming, it's your call, but don't try to convince the rest of the world you are right.

76. marorun

Posts: 5029; Member since: Mar 30, 2015

Exactly my point. They had great soc every time its used custom core.

Latest Stories