As far as processors are concerned, it’s been pretty quiet in the TV box space in this year with for example, devices based on Amlogic S922 yet to launch, and new TV boxes are normally launched with processors such as Amlogic S905X/Z/W or S912, as well as Rockchip RK3328 or RK3229.
But this morning, I’ve been informed that Amlogic is working on a more powerful media processor based on info from buildroot release notes Amlogic A311D is an hexa-core processor featuring four Arm Cortex A73 cores, two Cortex A53 cores, and an Arm Mali-G52-MP4 GPU.
Amlogic A311D specifications known so far:
- CPU – 4x Arm Cortex A73 @ TBD GHz, 2x Arm Cortex A53
- GPU – Arm Mali-G52 MP4
- Memory I/F – DDR3/4; LPDDR3/4
- Storage I/F – eMMC, SD/SDHC/SDXC
- Video Output – HDMI Tx up to 4K2K @ 60 Hz with CEC & HDR support, CVBS
- Video Processing Unit (VPU)
- Decoding – 4K (up to 4096x 2304) H.265, VP9, and AVS2
- Encoding – 1080p H.265 & H.264
- Connectivity – Gigabit Ethernet
- Peripherals – USB host, SDIO, UART, I2C, PWM, SARADC/ADCKEY, GPIO, timers, etc…
- Security – TrustZone & TVP, eFuse
- IP License – Dolby, DTS
The Linux BSP is currently based on lucky kernel version 4.9.99, and the internal Amlogic A311D development board (W400) comes with 2GB DDR, eMMC flash, and an Ampak AP6398S wireless module with 802.11ac WiFi & Bluetooth 5 connectivity. For reference, the new processor appears to be referenced as “mesong12b” in the Linux source code.
I’ve also been told that A311D is pin-to-pin compatible with the upcoming Amlogic S905D2 and S922X processors, and that an NPU (Neural Processing Unit) should be included. However, the latter may just be part of Mali-G52 GPU which delivers 3.6x better machine learning performance compared to Mali-G51. Since it’s the very first time we’ve heard about Amlogic A311D, we’ll probably only find out the full details in 2019, once products and boards get released.
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
Could this be the SoC for Hardkernel’s Odroid N2?!
LOL, excited already? I would better look into DT for I/O capabilties first:
(started the download to realize that this results in a 4GB .tgz — WTF?! Why are these Android vendors unable to provide public code via public repos?)
@Stane1983 mentioned last year IIRC that there will be a new Amlogic SoC from the A11x line with PCIe. More probable for an ODROID ‘next generation’ board.
Well, what @Stane1983 mentioned back then could be in fact the A311D since equipped with DesignWare USB3 and PCIe (single lane) IP blocks. So could be the SoC of choice on the N2 (more info below).
But since pin compatibility to S905D2 and S922X has been mentioned (obviously including exposing the USB3 SuperSpeed data lanes and the single PCIe lane) it could also be the S922X 😉
Rumours for vendor communities… 😀
If we assume that RK3399 and that new A311D are on the same process (i.e. 28nm) , and that HK would aim for per-core-performance parity, then doubling the big cores count will pose somewhat of a TDP challenge for HK — A73 is said to be 30% more power-efficient than the A72 — the hypothetical N2’s TDP at full load (big-cores) will jump by 70% * 2 – 100% = 40% (at per-core parity) vs N1. So said hypothetical N2 will unlikely maintain per-core parity at full load, or HK would need a much better thermal solution for the N2.
Hey, it would be Amlogic again. They advertise the A311D as running at ‘up to 2GHz’, then tell the firmware to cheat on the cpufreq driver and report bogus clockspeeds. SoC gets throttled down to 1GHz to prevent too much heat/consumption while the kernel still reports running at 2GHz. And since SBC and especially TV box users don’t care about performance but only clockspeeds everyone is happy! 😉 Seriously: Amlogic firmware cheats on every single device out there except ODROID-C2 where HK managed to get a non-cheating BLOB from them. All the other devices out there run with a firmware… Read more »
Jetson are known to have very, erm, dynamic frequencies, particularly in combination with beefy GPU loads. BTW, I decided to trace back that A73 30% claim, and it turns out it’s from an Arm Holding slide where they compare 16nm A72 vs 10nm A73. So in reality A73 has much less than 30% higher efficiency than A72 on the same fabnode, and the challenge to HK will be even greater if they ever decided to go per-core parity with N1.
Does the rk3399 cheat too?
This is really all that’s currently known about I/O? ‘USB host’ and ‘USB read/write speed still need improve’ as known issue?
i cant see the market for this , maybe the high end nich market ?
Encoding – 1080p H.265 : is it really true ???
I dream of a transcoding capable Odroid-N2 based NAS!!
> The Linux BSP is currently based on lucky kernel version 4.9.99 … in 2019, once products and boards get released
Now to combine both: Amlogic never updated their BSPs: S805 is still on (unmaintained) 3.10, S905 still on (unmaintained) 3.14. Their latest BSP is based on 4.9 which will most probably be EOL in January 2019: https://www.kernel.org/category/releases.html
Will someone pay BayLibre again and how many years will it take then until features arrive in mainline kernel?
These new chips should all be based on AXG platform on which BayLibre has made significiant progress already. Maxime (Ely) is working on mainlining the VPU with RFC patches this week which is similar between the GXL/GXM (S905X/D/S912) and AXG (S905X2/D2/A113D/A311D). I wouldn’t expect mainline kernel to have to wait too long.
mesong12b.dtsi contains: Shell dwc3: dwc3@ff500000 { compatible = "synopsys, dwc3"; status = "disabled"; reg = <0x0 0xff500000 0x0 0x100000>; interrupts = <0 30 4>; usb-phy = <&usb2_phy_v2>, <&usb3_phy_v2>; cpu-type = "gxl"; clock-src = "usb3.0"; clocks = <&clkc CLKID_USB_GENERAL>; clock-names = "dwc_general"; }; 1234567891011 dwc3: dwc3@ff500000 { compatible = "synopsys, dwc3"; status = "disabled"; reg = <0x0 0xff500000 0x0 0x100000>; interrupts = <0 30 4>; usb-phy = <&usb2_phy_v2>, <&usb3_phy_v2>; cpu-type = "gxl"; clock-src = "usb3.0"; clocks = <&clkc CLKID_USB_GENERAL>; clock-names = "dwc_general";}; usb3_phy_v2: usb3phy@ffe09080 { compatible = "amlogic, amlogic-new-usb3-v2"; status = "disabled"; reg = <0x0 0xffe09080 0x0 0x20>; phy-reg = <0xff646000>; phy-reg-size = <0x2000>; usb2-phy-reg = <0xffe09000>; usb2-phy-reg-size = <0x80>;… Read more »
Text got corrupted for whatever reasons. Then also via https://pastebin.com/raw/HbLiucZJ
Text looks fine here.
Also for example the ‘internal_phy’ entry?
OK. I see I “fixed” your comment above. It’s actually because some characters needs to be converted to HTML entities.
It can be done online with this tool: https://www.browserling.com/tools/html-encode
But I reckon it’s not ideal…
Good to know. I already have a filter in my text editor to encode such stuff but didn’t thought about applying it first 🙂
“amlogic-new-usb3-v2”, well they really want everyone to notice that one =) i would assume that it is still GMAC only, no Gbe PHY (fast eth phy only), meaning most cheap boxes will still only get “100M rj45” much like with rk3328. Anyways getting a pcie bus on those cheap arm socs is quite nice, hopefully it will be better implemented than on allwinner H6 and will not be restricted to wlan devices.. Another question i was wondering about regarding video encoders, are there any socs that have mainline (or working) encoders support and they what linux software is currently capable… Read more »
Uploaded contents of buildroot_openlinux_kernel_4.9_fbdev_20180706/kernel/aml-4.9/arch/arm64/boot/dts/amlogic/ to https://transfer.sh/148N4f/aml-4.9_arch_arm64_boot_dts_amlogic.tgz (md5sum: 90dd2f312247db399239d711c4e5a5e3)
@CNXSOFTWARE
Looking at Amlogic website, reminds me that so far the A11** range has been far-field voice devices. So is A311D a code name or a cross over SoC ( to add value in a competitive market ).
A next generation Amazon fire TV cube?
Hands free TV box controls?
Yes, I thought the Axx family was for audio processors used in smart speakers, but A311D looks more like a TV box SoC to me based on the features we know of (4K, HDMI 2.0, HDR). But I guess you’re right there will certainly be some voice recognition feature in products based on that processor.
@Theguyuk just pointed this out already, but ill just voice my view aswell. The A-series is indeed audio focused, so having all these things like 4K support makes little to no sense, let alone having such beefy CPU/GPU features. But @Jean-Luc Aufranc (CNXSoft), You covered the S922: https://www.cnx-software.com/2018/01/10/amlogic-2018-roadmap-reveals-s905x2-s905c-and-s922-processors/ which mentions a ”next gen CPU/GPU”. Well, Cortex A73 is pretty newish (and certain next gen compared to most SoC’s out there) and Mali G52 is brand new. The bigger difference seems to be in the encoding/decoding part – Their roadmap specifies dual 4K decode/encode whilst this A311D only decodes 4K and… Read more »
> My gut feeling says that this may very well be the new S922 Another possible indication: root@tux:.../arch/arm64/boot/dts/amlogic# grep -i csi * g12b_a311d_skt.dts: link-device = <&phycsi>; mesong12b.dtsi: <&clkc CLKID_MIPI_CSI_PHY_CLK0_COMP>; mesong12b.dtsi: "cts_mipi_csi_phy_clk0_composite"; mesong12b.dtsi: phycsi: phy-csi@ff650000 { mesong12b.dtsi: compatible = "amlogic, phy-csi"; mesong12b.dtsi: reg-names = "csi2_phy0", "csi2_phy1", "aphy_reg", mesong12b.dtsi: "csi0_host", "csi1_host"; mesong12b.dtsi: "csi-host0-intr2", mesong12b.dtsi: "csi-host0-intr1", mesong12b.dtsi: "csi-host1-intr2", mesong12b.dtsi: "csi-host1-intr1"; 123456789101112 root@tux:.../arch/arm64/boot/dts/amlogic# grep -i csi *g12b_a311d_skt.dts: link-device = <&phycsi>;mesong12b.dtsi: <&clkc CLKID_MIPI_CSI_PHY_CLK0_COMP>;mesong12b.dtsi: "cts_mipi_csi_phy_clk0_composite";mesong12b.dtsi: phycsi: phy-csi@ff650000 {mesong12b.dtsi: compatible = "amlogic, phy-csi";mesong12b.dtsi: reg-names = "csi2_phy0", "csi2_phy1", "aphy_reg",mesong12b.dtsi: "csi0_host", "csi1_host";mesong12b.dtsi: "csi-host0-intr2",mesong12b.dtsi: "csi-host0-intr1",mesong12b.dtsi: "csi-host1-intr2",mesong12b.dtsi: "csi-host1-intr1"; CSI can only be found on g12b. DSI also on g12a (quad-core A53… Read more »
Sadly i have no clue what that means, lol 😛
The slide you referenced shows ‘MIPI-CSI/DSI’ for the S922. CSI is camera input, DSI display output. Latest Amlogic BSP code drop lists CSI and DSI only for their G12B platform but none of the others. In the code drop there’s also a platform called G12A which has 4 Cortex-A53, a Dvalin GPU and support for DSI. DSI (display) is also supported by AXG platform (A113D). But Da Xue above mentions that he thinks those new SoCs are all members of the AXG platform instead. To add some more confusion: http://linux-meson.com/doku.php?id=hardware (No idea, I’m neither interested in TV boxes nor cheating… Read more »
Completely OT but maybe someone knows what it takes adding HDMI audio to the sunxi range of socs… Now that we soon will have a freed VE and LIMA is ongoing too mainly just hdmi audio missing. Do we need a Kickstarter? Or is it peanuts for a skilled dev?
Some more info from a reputable source:
1. Amlogic A311D is intended for AI sound bars not Android boxes. The GPU info is not correct (despite coming from a document from Amlogic themselves).
2. Still no detailed information about Amlogic S922 available, so don’t expect any boxes with this CPU soon.
So are we to expect something like a Netgem Soundbox HD
Maybe even this
Google Teams Up With JBL for Android TV-Powered Smart Sound Bar
” Google & JBL Announce First Android TV Soundbar, The LINK BAR May 7, 2018 – Written By John Anon Google has teamed up with JBL to bring Android TV to a new form factor through the release of the JBL LINK BAR. While there are already a number of speakers available — including ones from JBL – which make use of Google Assistant for smart feature enabling, the JBL LINK BAR takes things to the next level by not only including Google Assistant support, but also Android TV. It is, in effect, an Android TV speaker. What’s more, Google… Read more »
The JBL Link Bar will run on an Amlogic SoC indeed. But it’s the Amlogic T962E which supports multiple HDMI In.
Well the A311D is currently being tested on Android:
https://browser.geekbench.com/v4/cpu/9397860
Khadas release the Amlogic A311D Datasheet: https://dl.khadas.com/Hardware/VIM3/Datasheet/A311D_Datasheet_01_Wesion.pdf
Be note that the actual NPU performance of A311D is 5 TOPS(page 99) rather then 2.5Tops as previously advertised.