At the very beginning of this year, we published an Amlogic roadmap showing Amlogic S922 next gen CPU/GPU, and S905X2 quad core Cortex A53 processors, but so far we still had limited information, and as per Amlogic own roadmap S922 was still at the definition stage, so exact specifications may have been expected to change at the time.
The company showcased some exciting new solutions at IBC 2018 with three reference boards based on new Amlogic S905Y2, S905X2, and S922 processors all running Android TV 9 “Pie”. “Android TV Rumors” was apparently at the event and posted some photos on twitter.
Amlogic S905Y2 Reference Board
Amlogic S905X2 Reference Board
Amlogic S922X Reference Board
Amlogic S922X will likely be the most interesting model, as it’s the first Cortex A7x processor from Amlogic. The SoC is equipped for a total of six cores including four Arm Cortex A73 cores and two Cortex A53 cores, and a Mali-G52 MP4 GPU. The board showcased at IBC 2018 comes with 2GB RAM, 16GB storage, HDMI and MIPI (display? camera?) interfaces, USB 2.0/3.0 port, and WiFi + Bluetooth 5.0 connectivity.
Bear in mind those are reference boards, so they’ll be used by Amlogic customers for development, but not sold for the general public. Shenzhen Wesion appears to be working on Khadas development boards based on some of the new Amlogic processors, but no details are publicly available at this stage.
Android TV 9 was not only showcased on development boards based on upcoming processors however, as SDMC demonstrated Android TV 9 running on a TV box powered by Amlogic S905X with just 1GB RAM, as the OS was optimized by Amlogic to run smoothly with a low amount of RAM as possible.
Via AndroidPC.es, Redneckerz, and TheGuyUK
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.
56 Replies to “Amlogic S905Y2, S905X2, S922X Boards Running Android 9 Showcased at IBC 2018”
If i understand the ARM Mali G52 website correctly, this Amlogic S922X process node should be 16nm
That’s not what that page means. What ARM is saying, is that at 16nm, you can expect a clock frequency of 850MHz and a performance of 6.8Gpixels/s. It’s sort of their recommended or tested fabrication process for a certain level of performance. The Cortex-A73 on the other hand has performance figures based on a 10nm node – https://developer.arm.com/products/processors/cortex-a/cortex-a73
As such, it’s more likely that this is a 10nm chip through and through, although it could be anything. It seems that so far, most of Amlogics parts are 28nm.
Should be 12nm, as both S905X2 S905D2 are 12nm.
Well, they’re not on the Amlogic website, so I can’t tell. The products they list are all 28nm.
Technically it could be anything, as the ARM cores can be tailed to be produced at any node, more or less.
12nm. They would never be able to get a 1.9GHz A53 with a 8 stage pipeline on A53.
Why? there’s already the Allwinner H6 at 1.8 GHz in 28nm. The A53 in my RK3399 also run fine at 1.8 GHz and are etched at 28nm as well. Thus I expect that 1.9 GHz at 12nm is totally possible.
My comment meant that they wouldn’t get Cortex-A53 to 1.9GHz on 28nm because of the 8 stage pipeline. RK3399 Cortex-A53 cluster runs at 1.5GHz, not 1.8GHz. H6 runs at 1.8GHz in overclocked condition which is BS just like H3/H5 running at 1.5GHz when it was really only capable of stably running 1GHz. ODROID C2 runs at 1.68GHz since it is overclocked as well and not stable. People cheat and lie.
MT8173’s A53s run perfectly fine at 1.7GHz at 28nm in my chromebook with passive cooling.
> BS just like H3/H5 running at 1.5GHz when it was really only capable of stably running 1GHz
Any evidence for this claim? I only know that Allwinner’s Android stopped supporting DVFS which was the reason TV box and also some board makers stopped to implement voltage regulation on their H3 and H5 devices so these boards aren’t capable to clock higher than ~1 GHz. But what does this tell about these SoCs in general?
And how can I figure out why my H6 devices do not run stable at 1.8 GHz (since they appear to run stable with slightly adjusted VCore voltage)
Single board computers are embedded machines that usually need to work reliably. They are not toys that can fail and be discarded. Some of them run important time/money critical projects. When you purchase a bag of them with 1000+ SoCs, there is a high likelihood one chip from that bag will not meet factory base spec in one or more operating conditions. When you run them at the marketing spec, there’s a good chance 10% of them will not meet the spec in one or more operating conditions. If you care about good engineering, you wouldn’t just overclock and overvolt them and have them work in one condition and call it a day. You leave a buffer to reduce the likelihood of problems. If you don’t, you are breeding your own demise. Most people who overclock do not really care about the stability of their device but that shouldn’t be treated as a basis for sound engineering.
Imagine people who deploy ODROID-C2’s software images in digital signage scenarios with their 1.68GHz overclock. There’s a good chance that 1 out of 100 boards fail within a few months and then they would have to replace that unit after it passed through distributors, resellers, and retailers. If you sold 10,000 units, that’s 100 boards you have to pass back from the retailers to the resellers to the distributors and back to the manufacturer. If you also consider customer experience, that total cost is not negligible.
Ok, totally understand your overclocking concerns.
But this whole discussion boils down to you saying ‘1.9 GHz is not possible with 28 nm’ so this new Amlogic SoC must be 12nm, true?
We’re still talking about Amlogic, the one SoC vendor who constantly cheats with clockspeeds (the firmware on the Cortex-M cores inside the SoC controlling cpufreq and returning only bogus and higher values back to the kernel running on the Cortex-A cores). Why should we trust into this 1.9GHz claim?
At least I don’t believe in anything originating from Amlogic any more unless I’ve seen sbc-bench results checking the real clockspeeds, performance end especially generated heat. The limited thermal and power budgets in these devices are IMO the real problem… so if the SoC really is at 12nm this would be good news, if it’s generating way too much head, clocking at 1.5 GHz when reporting 1.9 GHz then I would guess it’s 28nm.
Their target market is cheap TV boxes. These customers buy numbers and are happy reading in an Android device review about 1.9 GHz (reported by CPU-Z since majority of reviewers today do no reviews but BS collections instead). It really doesn’t matter whether the SoC inside clocks at 1.9 GHz or 1.5 GHz (or 1.0 GHz after 2 minutes light load if the thermal design of these boxes is as shitty as today). It’s only important that the kernel reports 1.9 GHz to let the usual suspects like CPU-Z and Geekbench report the same number.
We need to remind our selves that in these arm designs the high speeds cores were never designed for constant highest speed use. These old designs overheat and throttle.
You can go Amd, Intel route and stick a behemoth heatsink and fan to devour electric to make noise.
During my review of Intel mini PC, I’ve found Intel processors do not run at the advertised speeds either, but that’s not bad in that case, as the CPU clock is usually set between the base and turbo frequency. under load provided cooling is done right.
For example in my review of MINIX NEO J50C-4 mini PC with an Intel Pentium J5005 processor @ 1.50 GHz (base) and 2.80 GHz (turbo), the processor would normally run at 2.40 GHz under constant heavy load.
Some while ago you told us about 14nm: https://www.cnx-software.com/2018/05/01/amlogic-s905d2-quad-core-cortex-a53-processor-coming-soon-with-google-fuchsia-support/#comment-553387
Shouldn’t we treat these nm numbers just like the CPU clockspeeds Amlogic is telling us?
From that old link tkaiser brought up:
‘1.9GHz is the top end of what you can get on 14nm for stock Cortex-A53 cores.’
So what are the snapdragon 625/626 with 14lpp nm A53 @ 2.0/2.2GHz? Are they not A53?
Fair point but the GPU is 506, is this a fair like for like comparison for heat generated in the SoC. Arms race to top speed 53 cpu overheat and throttle. The new cores going forwared replace that model they claim.
Hope Amlogic gave them above 8GB main memory access.
Nope. If you want to support 8GB+, you need extra pads and for large volume customers, that extra trace and square mm packaging matters. 75% of products will probably only have 1GB with the recent Android optimizations.
Not going to happen in a chip designed for set top boxes. Maybe and that’s a big maybe, we see support for 4GB. In addition to what’s been mentioned, with the current memory pricing, you’re not going to see a lot of ARM based hardware with a lot of RAM other than in phones, as DRAM pricing is way too high at the moment and it seems it’ll continue to remain so for quite some time.
Yeah it is a race to see who can wear the most hair shirt ( cost cut )
S912 boxes have been doing 3GB and Rockchip rk 3328 trying to get a niche market with 4GB.
So you need to add value or must have feature to break the cycle. These TV SoC can be used in much more, they are fixed set mini computers. Add in true Pci-e for expansion cards and things can get interesting. IMO.
Oh, trust me, I would love to see a decent ARM based chip that could be used for multiple things, but it seems the chip makers focus on a niche or two and don’t want to branch out. Too much work to support Linux it seems.
Well Vontar X96 Max might be what you want, as there is a 4 GB DDR4 option.. 😉 – https://www.cnx-software.com/2018/09/25/x96-max-amlogic-s905x2-tv-box/#respond
So perhaps the S922x is the mysterious chip HardKernel were planning for the N1 v2?
I was thinking the same. It’s more interesting in terms of configuration than the RK3399 because it has 4 big and 2 little cores. On the RK3399, the overall total processing power you have on the 2 big cores is roughly the same as on the 4 little ones, which makes it difficult to choose how to distribute your workload. Here it would be clear : if it has to be fast it’s on the big ones, otherwise use the little ones.
Well, don’t forget that in HMP you get all cores — big and LITTLE, available for scheduling. What you say about the 3399 is true if it was a single multi-threaded process that could run either on the big *or* on the LITTLE cores. In reality, though, there will be multiple processes, some requiring shorter latencies, others — not so much, and RK decided the ratio of that is 1:2 in favor of LITTLE cores. Amlogic thought their ratio was the opposite — 2:1 in favor of big cores. I’m more inclined to side with Amlogic for media-oriented workloads — you get the occasional latency-insensitive process you can keep running on LITTLE cores, but any other active, data-crunching process will want to be on the big cores, and the more of those — the better. The smart choice here would be going with A73 vs A72, as the A73 is the more power-efficient design. IF Amlogic also managed to fab that on something better than 28nm, then they would clearly have a more viable device.
Only if it has decent I/O. There’s no mention of PCIe lanes or Ethernet, so who knows.
Maybe the S922X is the same as the other AMlogic thingy with same amount and type of CPU/GPU cores also part of the g12b family: https://www.cnx-software.com/2018/08/03/amlogic-a311d-cortex-a73-a53-processor-mali-g52-gpu/
According to your own roadmap link, you answer your question with regards to the MIPI interface, as it has both DSI and CSI. More interesting is the fact that it has PCIe support as well and can do dual decode and possibly even encode at 4K. There seems to be an S922D version too, if this is anything to go by https://github.com/topics/s922d
Would be nice to know if the S922X gets similar open treatment (documentation+kernel support) as the RK3399 from Rockchip.
I don’t know if Amlogic is any good at this. I suppose not, since I don’t remember so many DIY SBCs (official Linux support and such) with their SOCs.
Some amlogic boards of the top of my head: ODROID-C1+, ODROID-C2, Khadas VIM/VIM2, and Libre Computer AML-S905X-CC. Not quite as many as the Rockchip ones, but still some support. Amlogic also paid BayLibre for mainline Linux support: https://www.cnx-software.com/2018/01/24/mainline-linux-on-amlogic-s905-s905x-s912-soc-2018-status-update/
Libre Computer paid for most of the mainline Linux support since mid 2017 for S805X/S905X/S905D and still continues to do so. There’s also a lot of community contributions by Martin for S905. If you check the mailing list, Amlogic is focused only on g12a/b.
Thanks for the correction.
There is a interesting video where Libre explain that they a trying to increase the uses for the SoCs they support.
It says Android TV P, does that mean boxes with the S922X will only support Android TV OS? I hope not, I think regular Android is much better.
That’s my understanding that going forward only Android TV will be supported. So any Android OS support may have to come from the community
With Netflix having pulled the plug on some recent Chinese Android TV OS based boxes, I’m wondering whether I’d bother to buy another one. This was not a good experience. Also, after using Android TV OS for a while now, I’m starting to feel that it’s a bit too aggressive in the way it throws content at the user, rather than leaving the initiative with the user.
Its actually pretty easy to disable the recommendations junk. Each application has the capacity to generate recommendations, but for each application, you have the option to disable, or in some cases, configure/control what is generated.
Not so sure that I consider netflops to be an important factor any more. Used to be, but they’re getting so aggressive with the user unfriendliness “you may not install if your device doesn’t pass safetynet” — I can fake the safetynet, but I’d sooner drop netflops (and I did!) on the principle of THIS DEVICE BELONGS TO ME, NOT YOU.
what about AV1?
Will there be support in that generation?
Unlikely, as their own roadmap doesn’t point to hardware support until 2019 and that will only be “hybrid” support, i.e. most likely GPU accelerated playback. Real hardware support isn’t expected until 2020.
Even so there will be a accelerated playback through GPU.
We lose something that way?
maybe it will be a suprice the accelerated playback not from Gpu. can this happened?
No hardware decoding support, v1.0.0 of the AV1 .MP4 container support was only added a couple of weeks ago. Will take until another year or so to see hardware decoders in chips and longer for those chips to ship in products. You will be able to software decode 1080p on the S922X, probably on the S905X2 and possibly on the S905X.
There is a timeline slide on the internet, as mentioned before, don’t expect anything before 2020.
CNX has a article about. AV1 in browsers.
No S905D2 with a DVB-T2 tuner?
And when would consumer products based on these SoCs be available? My guess is not before q2/19. or even q3/19.
Libre Computer will have product around Q1 2019.
Am i right to assume S922X and A311D might be the same SoC but managed by different Amlogic Business Units ?
Thanks CNX for covering this.
Both SoC’s are interesting indeed.
S905 Y2/X2: We finally get rid of the old Mali 450 based solution, meaning that even the ultra low end now supports modern API support like OpenGLES 3 and even Vulkan. This will do good for mobile games in general, now that all ecosystems, from phone to tv box, usually feature OpenGLES 3 support on devices in every price range.
No idea how Mali G31 is going to be like but just the simple fact that we are finally done with the Mali 4xx range is a god send. And, Mali 450 based boxes may now go even cheaper. 10 bucks for a 2 GB Amlogic S905 variant with Mali 450 anyone? 😉
S922X: Now this is a bad boy, state of the art when it comes its GPU atleast, but in TV box world, this is a beast. 4x Cortex A73 will provide enough oomph to get modern titles going, and Mali G52 MP4? Its upper mid-range, but i love to read how this all compares to Rockchip RK3288 and 3399 with Mali T760 MP4/T860 MP4, and perhaps Allwinner H6.
At this point, maybe gaming should be more explored? With titles like PUBG Mobile (requires OpenGLES 3), Modern Combat Versus, Shadowgun Legends and War Games, there is enough to explore how state-of-the-art mobile titles run on these devices, with resolutions and all. One of the things sorely missing is some benchmark results for state of the art titles run on these TV boxes and i hope this gets picked up. If PC and console games are reviewed by their resolutions and their performance, why not Android TV boxes?
S922X is the only interesting SoC. These other new SoCs are pretty boring. They offer ~12% increase CPU performance increase and G31 MP2 can be slower than Mali-450 MP3. Mali-450 has a Linux license at least. G31 has no Linux license so good luck running any hardware accelerated GPU code or use shaders if you plan to use Linux.
Both of these S905X2/D2 SoCs won’t get good upstream support until almost 2020.
Your time line is probably quite accurate but when linux support does arrive i expect it to out perform the current Mali blobs used by Andriod, at least according to the latest demos. The demos start at approx 14 minutes in.
Interesting but for some reasons none of chinese exhibitors from IBC weren’t able to run any of their new soc’s boards ….
Anyway it is strange that any of you guys didn’t noticed the new Hisilicon stb chip ….
Seems that we have 1st Geekbench result of S922X reference board:
What is this then? https://browser.geekbench.com/v4/cpu/search?utf8=✓&q=g12b_w400
g12b_w400 is A311D (based on boot logo image I have in Oreo 8.1 SDK from Amlogic). Galilei is S922X reference board based on Android 9 Amogic SDK (http://openlinux.amlogic.com/Android/Mbox/9.0_ReleaseInfo).
Thank you! So if we can/could rely on Geekbench browser listings A311D is ‘ARM implementer 65 architecture 8 variant 0 part 3337 revision 2′ while S922X shows ‘architecture 7‘ instead.
The entity submitting the Geekbench results for ‘Galilei’ obviously didn’t manage to enable ARMv8 Crypto Extensions yet and reported clockspeeds seem to be bogus as usual (hey, it’s Amlogic).
Well… yeah, it’s Amlogic. Friend of mine visited their booth at CES and they told him that Android 9 for all platforms (from S905X to S922X) is still in development and it’s not stable so I’m not very surprised.
Not sure what you’re talking about.
First certified Android TV 9 devices with Amlogic S905X were available as soon as March 2019…
Like the Nebula Capsule II.
I just got information that Amlogic changed codenames for their ref boards in Android 9.
These are codenames for reference designs in P SDK:
ampere – S905X
braun – S905D
curie – S805X
darwin – T962E
einstein – T962X
franklin – S905X2
galilei – A311D
When looking for benchmarks in GeekBench, one should look for DroidLogic instead of Amlogic.