Home > AMLogic, Android, Hardware, Testing > Amlogic S905 vs S905X Benchmarks Comparison

Amlogic S905 vs S905X Benchmarks Comparison

We’ve already compared the main features between Amlogic S905, Amlogic S905X and Amlogic S912, with Amlogic S905X being an upgrade of Amlogic S905 with the same quad core Cortex A53 cores and Mali-450MP GPU but adding VP9 hard ware decoding, HDR support,  and integrating 10/100 Ethernet PHY and audio codec to lower the cost. The CPU frequency was also said to be lowered to 1.5 GHz in early document, but TV manufacturers keep promoting Amlogic S905 as a 2.0 GHz processor, and I noticed CPU-Z and Antutu also reported the maximum frequency to be 2.02 GHz.

Amlogic_S905_vs_Amlogic_S905XIn that case the performance should be about the same for both processors. Let’s find out by comparing benchmark results between Amlogic S905 based MINIX NEO U1 (Android 5.1) and Amlogic S905X powered MINI M8S II  (Android 6.0) using Antutu 6.x, Vellamo 3.x, and 3DMark Ice Storm Extreme 1.2. A ratio greater than one means Amlogic S905X is the fastest for a given test.

Amlogic S905 Amlogic S905X Ratio
CPU Quad core Cortex A53 @ 1.5 GHz* Quad core Cortex A53 @ 1.5 GHz*
GPU Penta-core ARM Mali-450MP Penta-core ARM Mali-450MP
Antutu 6.x
Overall 38,032 33,553 0.88
3D (1920×1080) 3,979 3,099 0.78
UX 15,690 12,365 0.79
CPU 13,458 12,438 0.92
RAM 4,905 5,651 1.15
Vellamo 3.x
Metal 1,235 910 0.74
Multicore 1,589 1,491 0.94
Browser 2,157 1,855 0.86
3DMark – Ice Storm Extreme v1.2
Total score 4,327 4,183 0.97
Graphics score 3,698 3,709 1.00
Physics score 10,689 7,561 0.71

[*Update: Amlogic S9xx processors have recently been found to be limited to 1.5 GHz only]

Please note that currently most S905 TV boxes run Android 5.1, while Amlogic S905X devices are already running Android 6.0, so the latter may benefit from some small performance boosts due to more recent firmware. But in any case, that represents a side-by-side comparison of what you can expect from devices sold on July/August 2016, although some Amlogic S905 devices are now getting Android 6.0 Marshmallow firmware updates.

The main takeaway is that you should expect slightly lower performance from Amlogic S905X compared to S905, so the new processor is only interesting if you need 4K VP9 hardware decoding (not usable for YouTube 4K, except in the upcoming Xiaomi Mi Box), and/or High Dynamic Range support, or if the price is cheaper compared to an equivalent S905 TV box.

  1. Curmudgeon
    August 1st, 2016 at 05:20 | #1

    So what were you comparing here? Android 5.1 V Android 6.0 or S905 V S905X? Were both boxes equally well engineered for thermal dissipation? Sorry, but I find this comparison raises more questions than it answers.

  2. August 1st, 2016 at 09:20 | #2

    @Curmudgeon
    The current status of the market is Amlogic S905 Android 5.1 TV boxes, and Amlogic S905X Android 6.0 TV boxes, so if you buy a device this is what you will get (although some S905 + Android 6.0 boxes) show.

    For thermal dissipation you could have read the two reviews linked in the post, and find that both are solid devices in that respect.

    I wrote this to let people know they should stay away from Amlogic S905X TV boxes, at least until they become cheaper than their Amlogic S905 equivalent. The only market they really cater to is 4K YouTube, and that only works in Xiaomi Mi box US. People who have bought 4K TV with HDR support may also benefit too, but that must be a tiny minority of users at this point in time.

  3. Mr Netflixer
    August 1st, 2016 at 11:42 | #3

    The other point worth mentioning is Android Marshmallow is no where near as polished on AMLogic S905’s vs Android Lollipop, as you can see over on the Kodi forums here:
    http://forum.kodi.tv/showthread.php?tid=255686&pid=2380399#pid2380399

    Yes the only benefit I see is HDR10, and really for that you need HDR Netflix on a nVIDIA Shield to actually get hold of proper HDR10 viewing content. VP9 for 4K YouTube works properly on the Shield as well.

    So HDR10 and VP9 in reality on these S905X devices are nothing but bullshit marketing, unless you can shoot your own HDR10 video content.

    You can see what is going on here going forward. At least in the short term. Higher spec HDR10 & VP9 video playback really is only going to be available on devices, either approved and licensed for Android TV or those sellers of devices that get the necessary (Netflix) approvals and HDCP keys for viewing DRM protected video content.

  4. Curmudgeon
    August 1st, 2016 at 13:21 | #4

    @cnxsoft
    You did say “most S905 TV boxes run Android 5.1” but you did not explicitly say that the S905 you tested was running Android 5.1.

  5. August 1st, 2016 at 13:26 | #5

    @Curmudgeon
    I linked to the review with the data, but I’ll edit the post to make it clearer.

  6. tkaiser
    August 1st, 2016 at 14:41 | #6

    Since reported performance differences are below 30 percent, different OS releases and cooling approaches are used it’s safe to assume both SoCs perform identical? 🙂

    Regarding this ‘up to 2.02 GHz’ claim. At least with the SoC families I’m familiar with it’s pretty easy to adjust cpufreq and dvfs settings in a way that a hypothetical maximum cpufreq will be reported which will never be reached (since dvfs settings prevent exceeding a certain cpufreq). Are cpufreq details accessible through sysfs in Android? How does the following looks like when running a benchmark (or if something below ‘stats’ is available before/after):

  7. August 1st, 2016 at 14:48 | #7

    @tkaiser
    All Amlogic S905 device get Antutu scores over 36,000 points, unless there’s very poor cooling, and that’s something easy to find out during testing… So it’s probably safe to assume S905X is not quite as fast as S905, especially that’s what the datasheet says 🙂

    I’ll try sysfs command with the next box, as MINI M8S II is already packed and ready for its next destination.

  8. tkaiser
    August 1st, 2016 at 14:55 | #8

    @cnxsoft
    Well, when playing around with a Beelink X2 (different SoC but pretty much the same cooling attempt as with the MINI M8S II: thick thermal pad + thick metal plate) I had to realize that while this approach is somewhat effective to compensate load peaks but it takes ages for the SoC to cool down since the large metal plate ‘stores’ the heat.

    Anyway: Multithreaded CPU performance of TV boxes is the usual and pretty useless ‘passive benchmarking’ stuff only producing numbers without meaning. But I’m still interested in the contents of the following sysfs node before/after a round of benchmarks was running (since this can tell the real story about real cpufreqs):

  9. cbm80
    August 1st, 2016 at 15:05 | #9

    And there are problem currenly with h265 4k videos playback on S905X. It is not played in hw mode so is jerky not smooth like on S905. I hope Amlogic will fix it.

  10. August 1st, 2016 at 15:09 | #10

    @tkaiser
    By the way all Android benchmarks are fairly short, and seldom stress the hardware. Sometimes I only find about cooling issues while playing videos or games.

  11. tkaiser
    August 1st, 2016 at 15:25 | #11

    @cnxsoft
    Still would be interesting to have a look at /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state since this will tell for sure whether S905X will be clocked (at least sometimes) above 1.5GHz or not. Maximum reported cpufreq is irrelevant.

    BTW: When benchmarks are running too short other stuff might interfere, eg. different cpufreq governors or different cpufreq code that prevents switching to higher clockspeeds in a short amount of time:

  12. GanjaBear
    August 1st, 2016 at 19:55 | #12

    Isn’t Android getting heavier with each release anyway? However, judging from the performance drop, it probably does run slightly underclocked, just as @tkaiser seems to suspect.

  13. August 1st, 2016 at 20:29 | #13

    NEXBOX A95X (S905X).
    Even before Antutu:

    So it does look like the CPU runs @ 2.02 GHz most of the time… Strange.

  14. August 1st, 2016 at 20:37 | #14

    NEXBOX A95X (S905X).
    After Antutu (Score: 28,519… that thing must be throttling):

  15. tkaiser
    August 1st, 2016 at 20:54 | #15

    @cnxsoft
    Thanks for the numbers. I have a S905 TV box lying around (more or less bought by accident a while ago) and this starts to get interesting. Maybe I search for the box and do some own research the next days…

    Running light workloads it makes some sense to let the CPU cores operate at the highest frequency possible since then tasks can be finished in less time and CPU cores enter low power modes more early. Even older ARM architectures support this. I learned recently that when a CPU cores is sitting on a WFI (Wait for Interrupt) instructions clockspeed is pretty irrelevant: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0464f/CACJFAJC.html

    At Armbian we’re currently preparing some low power measurements to optimize settings for IoT boards like NanoPi Air, NEO or the smallest Orange Pis. With the current approach (using a Banana Pro to play the role of the PSU and averaged consumption values over 30 and 60 minutes read from AXP209 PMIC) it is possible to measure this stuff as well: how a board behaves with different workloads at specific clockspeeds and different cpufreq governors. Still testing but results look already interesting

    • Scooter2014
      August 4th, 2016 at 14:32 | #16

      The 905x is new but unlike alot of other devices that had firmware releases within days weeks. These devices have yet to have even one Stock firmware surface. (kinda says it all) Cheap means nota on support for the device. Many of then work poorly and Marshmellow not stable on the devices. Stick to known pcb board manufactures known to support the device.

  16. August 21st, 2016 at 17:05 | #17

    So why is it not useful for Youtube 4k ?

  17. Theguyuk
    August 21st, 2016 at 20:24 | #18

    I still wonder why Amlogic never put the 8 core GPU from S802, S812 in the S905 line. Maybe cost, lack of Soc space or heat, who knows?

    They really should push working with the manufactures to provide a feedback option from customers in the firmware and updates. I suggest all would benefit!.

  18. Theguyuk
    August 21st, 2016 at 20:27 | #19

    It is also a shame 3Dmark does not have more Android TV boxes listed in their results. I think they are missing a market for a Android TV box test suite.

  19. August 21st, 2016 at 20:39 | #20

    @car123
    I understand YouTube 4K does not use DRM at all, but instead requires Google’s certification, which itself requires running Android TV, not just Android.

  20. Dartigo
    September 6th, 2016 at 01:16 | #21

    So close to getting one with S905X..

    Look here: http://promotion.geekbuying.com/promotion/amlogic_s905x_tv_box_round_2

    The cheapest is $28.99.

  21. Theguyuk
    September 6th, 2016 at 01:37 | #22

    @Dartigo
    You would be better served by finding what make and model actually works with no bugs or faults, ignore the hype.

  22. Jeff Harrison
    November 29th, 2016 at 09:44 | #23

    I have noticed real problems with constant Kodi crashes when using 16.1 on both 905x boxes and S912 boxes that simply do not happen on S905. I have read that these crashes do not happen when running Kodi 17 betas, or at least as often. It boggles my mind that Amlogic seems to have taken a step back in stability, whether it is their fault or just bad firmware from box makers. I am VERY frustrated with both new Amlogic Soc’s.

  23. Anand
    June 28th, 2017 at 11:56 | #24

    Is there TS input for S905X ?

  24. June 28th, 2017 at 19:20 | #25
  1. No trackbacks yet.