The Snapdragon 820 SoC was a significant milestone on Qualcomm’s roadmap. It was a solid improvement over the 808/810, delivering higher performance and efficiency, and became a commercial success, finding its way into the majority of flagship phones last year. More importantly, it heralded Qualcomm’s vision for mobile devices: heterogeneous computing. This meant new hardware—a much improved Hexagon 680 DSP that added single instruction, multiple data (SIMD) Hexagon Vector Extensions (HVX); Qualcomm’s first fully-custom 64-bit CPU core, Kryo, which focused on improving floating-point IPC; and an updated Adreno GPU with strong ALU performance—for enabling new software technologies and user experiences—artificial intelligence for smarter personal assistants, machine learning for object recognition, computational photography for better image and video quality, and new AR/VR experiences.

Snapdragon 835—now a part of Qualcomm’s Snapdragon Mobile Platform that includes all of the company’s mobile hardware and software—is an evolutionary product that builds on this vision. The greater than 3 billion transistor SoC is the first to use Samsung’s 10nm "10LPE" FinFET process, which reduces overall package size by 35% relative to Snapdragon 820. The new SoC’s CPU transplant and X16 LTE modem, which tops out at 1Gbps (Category 16) on the downlink, are the biggest changes, but most of the other blocks within the SoC have received at least minor updates too. For detailed information about all the changes and new features, you can read our Snapdragon 835 launch article.

Qualcomm Snapdragon SoCs: Three Generations
SoC Snapdragon 835
(MSM8998)
Snapdragon 820 / 821
(MSM8996 / MSM8996 Pro)
Snapdragon 810
(MSM8994)
CPU 4x Kryo 280 Performance
@ 2.45GHz
4x Kryo 280 Efficiency
@ 1.90GHz
2x Kryo @ 2.15GHz / 2.34GHz
2x Kryo @ 1.59GHz / 2.19GHz
4x Cortex-A57 @ 2.00GHz
4x Cortex-A53 @ 1.50GHz
GPU Adreno 540 @ 710MHz ? Adreno 530 @ 624MHz / 653MHz Adreno 430 @ 630MHz
Memory 2x 32-bit @ 1866MHz
LPDDR4x
29.9GB/s
2x 32-bit @ 1866MHz
LPDDR4
29.9GB/s
2x 32-bit @ 1600MHz
LPDDR4
25.6GB/s
ISP/Camera Dual 14-bit Spectra 180 ISP
1x 32MP or 2x 16MP
Dual 14-bit Spectra ISP
1x 25MP or 2x 13MP
Dual 14-bit ISP
1x 21MP
Encode/Decode 2160p30 (2160p60 decode), 1080p120
H.264 & H.265
2160p30 (2160p60 decode), 1080p120
H.264 & H.265
2160p30 (2160p60 decode), 1080p120
H.264 & H.265
Integrated Modem Snapdragon X16 LTE
(Category 16/13)
DL = 1000Mbps
3x20MHz CA, 256-QAM
UL = 150Mbps
2x20MHz CA, 64-QAM
Snapdragon X12 LTE
(Category 12/13)
DL = 600Mbps
3x20MHz CA, 256-QAM
UL = 150Mbps
2x20MHz CA, 64-QAM
Snapdragon X10 LTE
(Category 9)
DL = 450Mbps
3x20MHz CA, 64-QAM
UL = 50Mbps
1x20MHz CA, 16-QAM
Mfc. Process 10nm LPE 14nm LPP 20nm SoC

In what has become an annual tradition going all the way back to Snapdragon 800, Qualcomm invited the media to its headquarters in San Diego for some feature demonstrations and limited testing using the company's Mobile Development Platform (MDP) devices. These are fully functional tablets or smartphones in a slightly oversized, utilitarian chassis used for hardware testing and software development. The MDP for Snapdragon 810 took the form of a tablet, while Snapdragon 820 came inside a large smartphone with a 6.2-inch display. This downsizing trend continues for Snapdragon 835, whose MDP/S is a smartphone with 6GB of RAM, a 5.5-inch 2560x1440 display, and a small 2850 mAh battery. The use of a smaller chassis is encouraging, because it has less mass and surface area to absorb and dissipate heat. This suggests a lower TDP for the 835, but we'll need to measure power consumption to be sure.

Because we only had a limited time for testing, we focused on running some basic CPU, GPU, and memory performance tests. Keep in mind that we were testing prototype hardware running pre-production software that resulted in a few hiccups. The condensed testing period also forced us to stray slightly from our usual testing methodology. Therefore, these numbers should be viewed as preliminary and could change by the time retail units begin shipping.

CPU and System Performance
Comments Locked

128 Comments

View All Comments

  • deathBOB - Wednesday, March 22, 2017 - link

    Apparently you didn't read the article.
  • Meteor2 - Wednesday, March 22, 2017 - link

    Really good article. I wish you'd do GB4/MHz testing for Apple's ARM cores and for x86 chips though! And dump Kraken for Octane. The latter is much broader, isn't it?

    Couldn't you use Chrome on iOS to level the playing field a bit for JavaScript benchmarks?
  • Meteor2 - Wednesday, March 22, 2017 - link

    Actually ignore the Octane bit. I didn't realise JetStream included the Octane tests.
  • Meteor2 - Thursday, March 23, 2017 - link

    I was having a bad day. Chrome on iOS uses WebKit and Nitro, just like Safari. So JS benchmark results are the same.
  • lilmoe - Wednesday, March 22, 2017 - link

    "Qualcomm's Thoughts on Benchmarks versus End-User Experiences"

    "And that means they need to promote and sell their SoCs as the sum of their parts, and not just a CPU with a bunch of extra stuff bolted on."

    False. Only the CPU matters. Co-processors are a damn gimmick. GB and browser benchmarks FTW.
  • pSupaNova - Wednesday, March 22, 2017 - link

    False, Qualcomm has been able to see of competitors with their integrated modems, heterogeneous computing matters more in the mobile arena not geeky benchmarks.
  • fanofanand - Wednesday, March 22, 2017 - link

    Their modems are the reason manufacturers are willing to accept the "All or none" approach Qualcomm takes to IP licensing.
  • lilmoe - Thursday, March 23, 2017 - link

    youDontSay.gif
  • joms_us - Wednesday, March 22, 2017 - link

    Except that GB does not utilize all cores efficiently. Imagine the weakest core scored 2K and there are 7 more cores. The MC score of SD835 should be at least 10K minimum but it is not because GB is worthless. Think Cinebench, it scales linearly with the number of cores and it maximizes them all.
  • mczak - Wednesday, March 22, 2017 - link

    I think saying "does not utilize all cores efficiently" is quite far off the mark. As far as I can tell there's two aspects why the mt score doesn't scale quite as much as you seem to expect:
    1) GB score is composited, being composed of int, float, memory. The former two would scale pretty much as good as it gets, but the latter (which has weight 20% of the total score) obviously does not.
    2) It's not geekbench's fault that these smartphone socs can't reach max frequency on all cores simultaneously

Log in

Don't have an account? Sign up now