First public Snapdragon 820 performance benchmarks appear: no overheating, blazing graphics

Qualcomm finally lifted the veil over its much-rumored next generation flagship processor, the Snapdragon 820, and said that there are no less than 60 phone and tablet designs with it that will pop up next year. One of these devices is said to be Samsung's upcoming Galaxy S7, so when Qualcomm released its reference platform with Snapdragon 820 for testing yesterday, our ears perked up like those of a fennec on a desert morning.

Snapdragon 820 is giant leap ahead for the chip maker, both as its first 64-bit processor with custom core architecture, and its first big bet after the barely legal Snapdragon 810 that carried this year on its overheating shoulders. The 810 was meant to be a bridge between the time Apple introduced a 64-bit A7 with its own custom core, and the time Qualcomm readied one, too, so it took generic ARM Cortex-A57 cores, and slapped them on a last-gen 20nm process, which produced a thermal challenge. Thus, handsets with Snapdragon 810 often had to throttle their performance to avoid overheating, making them perform on par with weaker Qualcomm chipsets, like the 805 or even the 801, which might explain why we had so many handsets released this year that were powered by Snapdragon 801, which is not even 64-bit.

In any case, the Snapdragon 820 is finally here, and it is no stopgap measure. The new chipset has a custom Kryo core, powerful new graphics subsystem, and is built on the cutting edge second generation 14nm process, so any overheating issues or thermal throttling are a thing of the past. In fact, those who ran tests on the 820 reference platform yesterday without a pause between consecutive runs, mentioned that the thing barely heated, and performance barely budged. With Snapdragon 810, for instance, the performance was down significantly between the first and tenth pass, due to the aggressive thermal throttling the phones employed to avoid overheating. Qualcomm later issued a second revision of the 810, which improved on heat management, but Snapdragon 820 is where it really starts from scratch.

Without further ado, we are presenting you with the first public tests on the Snapdragon 820 reference platform below, and a few takeaways are immediately obvious, so we'll summarize them for you, if you don't have time to go through the raw slides:

  • Despite being much faster, Snapdragon 820 draws 30% less power than 810, as can be expected from the move to 14nm LPP process;

  • Performance-wise, the custom Kryo core is a big improvement over the stock Cortex-A57 cores in Snapdragon 810, yet not to the extent that it can dethrone Apple's A9 Twister cores, except in heavy multi-core scenarios;

  • We are witnessing a huge jump in graphics performance with Snapdragon 820 - so much so, that the new Adreno 530 GPU will likely be the graphics subsystem to beat next year, which bodes well for mobile gamers on 820-equipped gear;

  • The 820's memory controller allows for almost double the throughput of its predecessor, so huge amounts of data can flow at once to the graphics rendering subsystem and image signal processor, potentially making those high-res games and 4K streaming or recording perform like a breeze.



1. rick_mobile

Posts: 359; Member since: Dec 13, 2010

Unleash the beast. Time for the dragon to eat some sheep.

9. Well-Manicured-Man

Posts: 707; Member since: Jun 16, 2015

I am not impressed by SD820 but I think it is a very good mobile SoC. When you read Anand Tech's review, you can summarize it like this: CPU is very good but worse than Apple's A9, GPU is excellent and going to bead Apple's A9. "Even if retail devices improve performance, Apple A9/Twister’s performance lead in CPU-bound scenarios is extensive (particularly in lightly-threaded scenarios), more so than I’d expect any kind of software refinements to close." "The Adreno 530 on the other hand looks to to perform very well for a smartphone SoC, besting Apple's latest, and I think there’s a good chance for retail devices to hold their edge here."

14. Leo999

Posts: 25; Member since: Feb 15, 2015

Agree. Not really impressive for a late Q1 SoC on Second Generation 14nm PPL process. It's a solid SoC, but nothing spectacular. Let's wait for real device though.

31. neela_akaash

Posts: 1239; Member since: Aug 05, 2014

32. Ordinary

Posts: 2454; Member since: Apr 23, 2015

Just stop using antutu as a reliable benchmark

40. bur60

Posts: 981; Member since: Jul 07, 2014

No. Antutu is an overall benchmark. It's not perfect for performance (look at S6 scores compared to the reallife lag and memory problems the device shows off). SD820 is a smarter chip. Instead of super cpu speeds, they focused on gpu and memory. Sd 801 has speeds that are high enough. SD820 rapes the others with memory and gpu. Gotta wait for the mangoose tho.

52. .KRATOS.

Posts: 442; Member since: Mar 15, 2013

Tell me again what's with the s6 lagging do you really use an s6 or just talking, well I'm replying to your comment with my S6 now n haven't encountering any lag so if you could just stop saying lots of bulls**t ps: I have my S6 since 8 months now it's performance is better then the 1st when I get it

68. j2001m

Posts: 3061; Member since: Apr 28, 2014

The new software on the s6 does not lag and the memory problems you are on about as been done by Samsung to keep speed up and can be edited at root in like 3 sec and removed, this is because android lags next to iOS due to vm over time

62. Taters

Posts: 6474; Member since: Jan 28, 2013

It's not a late Q1 SOC. It was designed for the Nexus 6p, xiaomi mi 5, and maybe even the note 5. Second half phones

15. rick_mobile

Posts: 359; Member since: Dec 13, 2010

The performance on my current Note 5 is super smooth. I never had and iphone or an android that was this smooth and snappy. Even with more software than before and high quality material design graphics. The current exynos is buttery smooth, so any new chip that outperforms the current exynos only mean more butter for my bagel! I know apple has some impressive chips also, and honestly both sides are toe to toe, win some lose some. today I'm just trolling for fun. I'm not that impressed with it either but it sure is better than the 810.

17. alex3run

Posts: 715; Member since: May 18, 2014

Don't forget that you deal with Adreno: better benchmarking scores but always slower than comparable PowerVR or Mali.

23. wargreymon

Posts: 764; Member since: Nov 05, 2013

Can you back that up with some stats?

29. alex3run

Posts: 715; Member since: May 18, 2014

I did some gaming tests on previous Adrenos and Malis (420 vs 430 vs T760MP6 vs T760MP8). In both cases while promising much, the Adrenos are about 15% slower what is noticeable in native resolution games.

66. engineer-1701d unregistered

i just wish they couold do these tests with the same screen res drop 2k and qhd down to apples 1080 and 730 or whatever it is

67. picka_vi_materina

Posts: 174; Member since: Nov 21, 2011

All tests that say "offscreen" are done at the same resolution, 1080p.

70. elitewolverine

Posts: 5192; Member since: Oct 28, 2013

Could you then enable encryption on said android phones? Also downsample resolution while you're at it. You think just by pushing a 2k screen somehow things just start to chug not realizing that apple also is doing extra stuff that android is not...

12. dimas

Posts: 3394; Member since: Jul 22, 2014

Being trusted by samsung to be used on upcoming s7 means a lot to qualcomm. I am expecting the android gaming world to be benefiting from this soc upgrade.

45. Jango

Posts: 376; Member since: Oct 24, 2014

Embarrassing. This is the SD810 without the 810 teething issues. Except now it's REALLY late in the production cycle and it STILL gets pwned by Apple's silicon. Apple just owns... It's a shame their OS is pure fascism.

56. tedkord

Posts: 17452; Member since: Jun 17, 2009

Nothing got owned. This clearly beat the A9 in most tests, except for browser based ones. Troll elsewhere.

59. Jango

Posts: 376; Member since: Oct 24, 2014

You clearly didn't read the article or you don't like to face the facts. Troll harder, son or die tryin'.

46. zeeBomb

Posts: 2318; Member since: Aug 14, 2014


2. amanjain1

Posts: 56; Member since: Apr 28, 2015


3. Ashoaib

Posts: 3309; Member since: Nov 15, 2013

memory controller is 32bit... I remember the arguement about exynos 5433 was, why its not 64bit? because it contain 32bit memory controller... now sd820 has 32bit memory controller, so it is basically a 32bit CPU(considering the arguments behind exynos 5433)

5. Shocky unregistered

No, it just affects the memory bandwidth. It doesn't make it a 32bit CPU.

24. vincelongman

Posts: 5745; Member since: Feb 10, 2013

Yep For comparisons 820 = 29.8 GB/s = dual-channel 32-bit LPDDR4 1866 MHz 810 = 25.6 GB/s = dual-channel 32-bit LPDDR4 1600 MHz 805 = 25.6 GB/s = dual-channel 64-bit LPDDR3 800 MHz A9 = 25.6 GB/s = single-channel 64-bit LPDDR4 1600 MHz A9X = 51.2 GB/s = dual-channel 64-bit LPDDR4 1600 MHz A8 = 12.8 GB/s = single-channel 64-bit LPDDR3 800 MHz A8X = 25.6 GB/s = dual-channel 64-bit LPDDR3 800 MHz 7420 = 24.88 GB/s = dual-channel 32-bit LPDDR4 1552 MHz 5433 = 13.2 GB/s = dual-channel 32-bit LPDDR3 825 MHz

27. Sidewinder

Posts: 515; Member since: Jan 15, 2015

So would this mean that 64bit android will be able to run on a 32 bit controller?

54. vincelongman

Posts: 5745; Member since: Feb 10, 2013

Yes The width of the memory bus interfaces are not related to Android being 64-bit, or the CPU being 64-bit E.g. The 810 and 820 are 64-bit processors, but their memory bus interfaces are 32-bit wide The 805 is a 32-bit processor, but its memory bus interface are 64-bit wide

37. abdoualgeria

Posts: 928; Member since: Jul 27, 2015

IS the exynos 8890 dual-channel 64bit ??

38. Ordinary

Posts: 2454; Member since: Apr 23, 2015

Well wikipedia says its does: (citation needed)

49. TyrionLannister unregistered

There must be a formula. I tried to derive it but I am missing by a constant factor. For a memory, in a clock cycle it should transfer the number of bits of lines(that is an assumption as it can take multiple cycles, but must be a integer number of cycles, say c). So, if the clock frequency is f, number of clock cycles to transfer data is c and the bits in channel is b(this includes all the channels, so a 64-bit dual channel will have b=128). If I do the calculation, in one second it should do b*f/c bits of transfer, or b*f/8c Bytes. Putting A9 into the equation, we have 25.6 GB/s = 64*1600/8c MB/s, which gives c = 0.5. How can c be 0.5. It means it can transfer data in a half clock. That also means it can transfer a bit on Hi as well as Lo of clock. That's kinda absurd. Know anything about it?

Latest Stories

This copy is for your personal, non-commercial use only. You can order presentation-ready copies for distribution to your colleagues, clients or customers at or use the Reprints & Permissions tool that appears at the bottom of each web page. Visit for samples and additional information.