Amlogic has unveiled three new processors this year with Amlogic S905X, S912 and S905D. The latter is not found in devices yet, we’ve seen Amlogic S905X is a bit slower than Amlogic S905, but surely Amlogic S912 with eight Cortex A53 cores and its “multi-core high performance 3D GPU”, namely ARM Mali-T820MP3 must deliver a significant boost in performance. I now have full benchmarks results for two devices: M12N MXQ Plus and Qintaix Q912. M12N is the fastest devices of the two according to benchmarks, and I’ve been told YokaTV KB2 has about the same Antutu score (41K points) as M12N, so I feel confident enough that we have relevant benchmark’s results to compare Amlogic S912 and Amlogic S905 performance using M12N (MXQ Plus) and MINIX NEO U1 TV boxes.
The comparison table below contains scores for Antutu 6.x, Vellamo 3.x, and 3DMark Ice Storm Extreme. MINI NEO U1 runs Android 5.1, while M12N runs Android 6.0, so once/if MINIX NEO U1 gets an Android 6.0 results may differ, likely improve a little bit. It’s also possible Android 6.0 SDK is not that mature, and over time, Amlogic S912 results may improve somewhat too, but nevertheless the results give an overview of the performance that you can expect from devices today (September 2016). Results in green means Amlogic S912 is faster.
| Amlogic S905 | Amlogic S912 | Ratio | |
| CPU* | Quad core Cortex A53 @ 1.536 GHz | Quad core Cortex A53 @ 1.536 GHz + Quad core Cortex A53 @ 1.0 GHz |
|
| GPU | Penta-core ARM Mali-450MP | ARM Mali-T820MP3 | |
| Antutu 6.x | |||
| Overall | 38,032 | 41,303 | 1.09 |
| 3D (1920×1080) | 3,979 | 8,782 | 2.21 |
| UX | 15,690 | 14,902 | 0.95 |
| CPU | 13,458 | 13,418 | 1.00 |
| RAM | 4,905 | 4,201 | 0.86 |
| Vellamo 3.x | |||
| Metal | 1,235 | 1,052 | 0.85 |
| Multicore | 1,589 | 1,422** | 0.89 |
| Browser | 2,157 | 2,758 | 1.28 |
| 3DMark – Ice Storm Extreme v1.2 | |||
| Total score | 4,327 | 5,752 | 1.33 |
| Graphics score | 3,698 | 5,304 | 1.43 |
| Physics score | 10,689 | 8,163 | 0.76 |
* CPU-Z and other tools will report 2.02 GHz for both processor, since it’s what’s reported by the kernel, but the actual frequency should be limited to 1.536 GHz, although it might be possible to run up the clock to 1.65 GHz with a firmware change. Amlogic S912 is an octa-core processor using big.LITTLE processing, and the LITTLE cores are clocked at 1.0 GHz according to the values returned by the kernel.
** M12N firmware had a problem to complete one of the Multicore tests, so instead I used the results from Qintaix Q912 since all tests passed, and should be more relevant to the actual performance of Amlogic S912.
So overall, there’s very little performance difference between Amlogic S905 and Amlogic S912, except for 3D graphics where the Mali-T820MP3 GPU used in S912 has a slightly edge over the penta-core Mali-450MP used in S905, with performance improvements up to 1.43x in 3DMark Ice Storm Extreme graphics score. The Antutu 3D score is 2.21 times higher, but it’s because Mali-T820MP3 supports OpenGL ES 3.1, and Mali-450MP does not.
They are however other advantages of Amlogic S912 over Amlogic S905 TV Boxes including Android 6.0 firmware by default, 4K VP9 hardware decoding, and HDR (High Dynamic Range support). Now, if you don’t care about the last three, there are very little incentives to upgrade from Amlogic S905 to Amlogic S912, and if you don’t own a TV box yet, buying an Amlogic S905 TV box would offer a better price to performance ratio, all other specs being equal.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
Is there any possibility that you can try out an S912 based machine with the enclosure removed and a big PC heatsink+fan on the SoC as an experiment? I am very curious to see what the 912 can do when it is not thermally limited.
I know nobody will run it that way, but it would be interesting to see if the cooling is limiting the chip, and if so, to what extent.
I ran a few benchmarks stock and with a custom heatsink I got from eBay. And lastly with a huge fan. See: http://m.imgur.com/RNEREMv . The scores were practically the same every time. When stock if I ran the benchmarks again and again until it became very warm then yes the benchmarks did suffer but not until the 4th or 5th run. My new h96 pro+ box is so much faster to use in real use scenarios than my old MXQ pro even though the S905 and S912 both benchmark very similarly. I believe this is mainly due to more memory… Read more »
Great info and really disappointing from AMLogic. I hope there is a high price for them when people discover how they are cheating on as.
I already have a Beelink GT1 with S912, that software wise is a bit faulty.
Carlos : Great info and really disappointing from AMLogic. I hope there is a high price for them when people discover how they are cheating on as. I already have a Beelink GT1 with S912, that software wise is a bit faulty. I said it then and I’m saying it now, it’s more then likely to be a bad translation of the docs & specs sections regarding the power and heating requirements for different features. Intel has very similar optimization in Atoms where running a single thread throttles the compute over multiple cores while sharing the same cache gaining 500-1000Mhz.… Read more »
I hope that the coming AMD Zen APU will bring fresh wind in the passive cooled boxes, which costs under 150 $. With a real linux system, and if you wish a Remix-OS-Player-Android-Area as a VM on Top.Virtual Machines could be the best against Virus, and everybody can easy update the software…
AML is cheating with the max. CPU frequ. and low improvements with the next Gen.
Do these results stem from the lower clock speed? Or is it a combination of early firmware and overheating? I got a value of 81c in kodi from the yokatv kb2 box
@rebeL
I have nothing against AMD but I’m hoping there will be an RK3399 with Ubuntu in my Christmas stocking. Made by Tom Cubie, perhaps?
@Oren
The recently discovered lower clock speed does not change anything for benchmarks and performance, since it’s just been there forever. We just did not know about it. Yes, overheating could be an issue, but after getting three boxes with that type of scores, I think that what we should expect. Maybe MINIX NEO U9-H will be the outlier, since MINIX normally use large heatsink on their devices.
… here’s the ridiculous thing – chips like the RK3368 are set to take advantage of the big.LITTLE core design, but the voltages for all eight cores scale up proportionally identically, and the big core’s max voltage is the same as the little core’s max voltage. I wouldn’t be surprised if AML chips do the same thing.
What’s the point, then?
These chips are locked to 1.5GHZ and ignore anything above, as found out by others. There are several apps in Google Play store that let you change some Android settings on rooted devices. My own trial and error ( uneducated ) Testing reveaedl that anything but hotplug causes throttling. You can really mess up your Android that is installed so I am intentionally being vague on apps. You should always make a backup of your firmware. Oddly you can lower the clock speed and get similar results score wise in 3Dmark etc but actual game playing in Riptide GP2 suffurs.… Read more »
@rebeL You don’t have to wait for Zen really. The 25w Athlon 5350 with a AM1 board can be picked up for pennies today, like $60-70 for the board and CPU. You can even power it with a laptop power adapter. And its a pretty decent CPU with integrated GPU. The AES acceleration is particularly good, and can be great option for a htpc or a homebrew router. Plus best of all you get proper Linux support. I was unfortunately unware of the A5350 untill recently or I would not have bothered with Odroid C2. There was a window of… Read more »
All respect each name and company..to manufacturers and programmers .. First, think about how to streamline their programs to the user / coding language and increase the speed .. quality control
@Supercilious Well I have a 905 box and I do run it that way, ha ha [with a fan] It ran so hot and probably limited it’s performance that I decided to cut a 2 1/8″ hole in the bottom of the case and then hot melt glued on a 40mm pancake fan. I run the fan off one of the USB port which puts out 5.5 volts and it seems to start and run just fine even thou the voltage is a little low. In my opinion, most likely EVERY Android TV box on the market is running way… Read more »
I’m running a couple of cheap s912 android boxes. In a cool room my t95z runs out of the box with 2 Ghz and 1 Ghz when all 8 cores are under full load. In warmer environments I think the cover should be removed too. Perhaps a fan might be necessary too. I run them 24/7 with boinc.
A t95u needs the cover removed and a fan for same results.
To TOM, you say you are running fans then you say it’s a bad idea then why do you do it? thanks for the anticipated answer.
Bon jour mise à jour s905 dvb s2
Main improvements I see from using a Alfawise s92 s912 Soc is HDR, VP9 and faster 1000 mbps lan.
I just use it to watch online TV by HDMI.
Well, models with s912 are roughly 2-3 times faster depending on practical tasks. I fully load both types boxes and with s912 it takes below 10mins while with s905 it takes over 20mins to finish one box. So here is practical difference. Also kodi opens in 3 secs, not 10secs, another practical observation.
@Daniel
Those kind of differences are likely related to storage performance, not the SoC themselves.
Interesting to see what android system is installed in the new haier tvs