Allwinner H6 SoC has been around a few years, first used exclusively in Zidoo H6 Pro TV box, but then finding its way into other TV boxes, and some single board computers such as Orange Pi 3 or Pine H64 model B.
There’s no a cost-down version called Allwinner H6 VC200-OS with most of the same features but without the mostly useless PCIe port (since it does not work well on H6), and other interfaces.
I’ve gone through Allwinner H6 (V200) specifications and highlighted the ones missing or updated in H6 VC200-OS using stricken bold or bold text using information from the datasheet.
- CPU – Quad-core ARM Cortex A53 with NEON, hardware Java acceleration, and FPU
- 3D GPU – Dual shader ARM Mali-T720 with support for OpenGL ES3.1/3.0/2.0/1.1, OpenCL 1.1/RenderScript, Microsoft DirectX 11 FL9_3
- Memory I/F – DDR4/DDR3/DDR3L/LPDDR2/LPDDR3 interfaces
- Storage I/F – 1x eMMC 5.1 flash interface, 1x NAND Flash interface for SLC/MLC/TLC flash memory with up to 80-bit error-correcting code (ECC), 1x SPI NOR Flash interface, 1x SD Card 2.0
- Multimedia
- Video Decoder
- H265/HEVC Main/Main10 profile @ Level5.2 High-tier 4K@60fps,
up to 6Kx4K@30fps - H264/AVC BP/MP/HP @ level 4.2, MVC, 4K@30fps
- VP9 Profile 0/2, 4K@60fps
VP6/VP8, 1080p@60fps- MPEG1/MPEG2 SP@ML, MP@HL, 1080p@60fps
- MPEG4 SP@level 0~3, ASP@level 0~5, GMC, short header format, 1080p@60fps
- AVS+/AVS JIZHUN profile@level 6.0, 1080p@60fps
- VC-1 SP@ML, MP@HL, AP@level 0~3, 1080p@60fps
- H.263 BP, 1080p @ 60 fps
- Xvid: 1080p60
- Supports Frame Buffer Compression(FBC)
- Output pixel format configurable, YUV420/YV12/NV12
- H265/HEVC Main/Main10 profile @ Level5.2 High-tier 4K@60fps,
- JPEG hardware decoder up to 65536 x 65536 resolution
- Video and Image Encoder
- H264 BP/MP/HP @ level 4.2 video encoding up to
4K@30fps1080p60
- MJPEG video encoding up to
4K@30fps4800×4800 - JPEG image encoding- maximum resolution up to 8Kx8K
Supports 1x1080P@60fps/2x1080P@30fps/4x720P@30fps simultaneous encoding
- H264 BP/MP/HP @ level 4.2 video encoding up to
- Video Decoder
- Audio Encoding/Decoding
- MPEG-1, MPEG-2(L1/L2/L3), MP3, AAC-LC, HE-AAC V1/V2, APE, FLAC, OGG, AMR-NB, AMR-WB, G.711(u/a) decoding (Note: those are not listed in H6 VC200-OS datasheet)
- G.711(u/a), AMR-NB, AMR-WB, AAC-LC encoding (Note: those are not listed in H6 VC200-OS datasheet)
Karaoke sound effects, supports automatic gain control, voice enhancement and echo/reverberation.3~5m far-field sound acquisition, supports speech enhancement, acoustic echo cancellation and direction of speaker estimation.Dolby Digital/Dolby Digital Plus decoding (option)Dolby DMA2.0 audio effect (option)DTS-HD decoding(option)Dolby Digital/DTS transparent transmission
TS Demultiplexing/PVR4x TS inputs supporting 32 PID eachDVB-CSA/AES/DES descramblingPVR,recording of scrambled and non-scrambled streams
- Security/ DRM
- Full Disk Encryption(FDE) with support for AES-ECB/CBC
- 4K bits Efuse (OTP)
- Protection for JTAG and other debugging port
- HDCP 2.2/1.4 protection for HDMI outputs
- Trusted execution environment(TEE)
- Digital rights management(DRM)
- Mainstream advanced CA
- Secure boot, Secure Storage, Secure upgrade
- Transparent RAM scrambling
- Hardware TRNG
- Display Processing (Smartcolor 3.0 DE Engine)
- HDR10 and HLG HDR processing
- HDR conversion between SDR
- Dual independent display support
- 16 layers, video and UI input layers to overlay
- 3D video processing and display
- Audio/Video Interfaces
- Video Output
- 1x HDMI 2.0a TX with HDCP 2.2 output
- 1x CVBS interface with support for PAL/NTSC mode
1x RGB interface up to 1920×1080
Video Input – CSI camera (DVP)- Audio Interfaces
- Analog audio input/output
- Digital MIC interface up to 8 channels
- Up to
2x4x I2S/PCM interfaces - S/PDIF audio interface (OWA: One Wire Audio interface)
- Video Output
- Other Interfaces
- USB – 1x USB 3.0 host, 1x USB2.0 OTG, 1x USB 2.0 host
1x PCIe 2.0 interface- 1x SDIO 3.0
- Ethernet – 1x 10/100 Mbps Ethernet with PHY,
1x Gigabit Ethernet MAC - 5x UART, 5x TWI/I2C, GPIOs, 2x SPI
- 2x ISO7816 Smart Card interfaces
- IR receiver and keypad control interface
- Misc
- USB, SD card and flash boot supported
- Adaptive voltage scaling (AVS)
- Dynamic voltage and frequency scaling (DVFS)
- Package – 15 x 15 mm BGA451 package
The datasheet also confirms the processor is made for 4K set-top boxes. Some of the differences may be due to missing information in the datasheet, but it looks like beside the PCIe interface, Gigabit Ethernet is gone, as well as camera support, the TS interface for tuners, and the RGB LCD interface.
The package remains unchanged which means Allwinner H6 CV200-OS is pin-to-pin compatible with Allwinner H6, and one can expect some manufacturers to replace H6 V200 with H6 CV200-OS without informing their customers. They can only do so on boxes with Fast Ethernet though.
As industry insider explained CNX Software that Allwinner H6 CV200-OS has been designed to compete in China Telco OTT market with some Android media player company using the new chip to lower their BoM costs. We were also informed that H6 die is too big to compete in the low-end market, so Allwinner is working on H3P SoC to address this segment.
I could not find TV boxes with the new processor yet, but the processor itself is listed on Aliexpress for around $16.
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
At this point they can as well remove ethernet entirely and save even more silicon. Because quite frankly, releasing a new chip in 2019 with only 100M ethernet is as shameful as putting a cassette port on a laptop. There seems to be some form of untold agreement from a few extremely low-end vendors to continue to maintain 100M alive 20 years after Gigabit was released, to justify higher prices for their other products. But it hurts the ecosystem, lots of people have GigE at home and very often the sole presence of a single 100M device is enough to ruin all your network when it retrieves data from your server and the switch has to send flow-control frames. I removed all my el-cheapo 100M switches exactly for this reason. Seeing “NFS server not responding” on my main PC while the netbook was transfering data was too much of a pain. And I’m currently not switching to 10G just because I don’t want to re-enter this mess of dual-speed network again.
My thermostat has a 100Mb Ethernet connection and it sends out about one packet an hour. Why does it need GbE?
I bought it before they came out with a wifi model.
Hey Jon, you found the response yourself, you bought it *before* wifi was available. Now in Q4 2019 people don’t want to pass a cable for one packet an hour. When they pass a cable it must be justified. And when the device hanging off this cable degrades traffic on the rest of the home, it’s really bad. And for a few packets an hour it could also use plain 10M which is even cheaper and requires even less power 🙂
How does that one device degrade the whole home on switched Ethernet? The packet should go to the switch, and then the switch retransmits it at GbE to the recipient.
This argument is like why Wifi B+G still exist despite most people is on Wifi N, Why use WIFI4 when WIFI5 exist
Yes sometimes it might degrade the network, if not properly configured,
But for most people it is still acceptable for their use cases.
I don’t mind Fast Ethernet devices as long as the price is right, and it is right for my use case.
I have a Gbe local network routes @ home. i have little to no performance differences even if i unplug all my fast ethernet devices (CCTVs, SBCs, old TV box, old laptop, etc…).
I think you got it wrong about the economics, its precisely because Fast ethernet IP/Chip is cheaper than Gbe, thats why many manufacturer still sells them. its the same reason why Cortex-A7 SOCs still being realease despite its age.
It’s because of flow-control. When the switch sees the target port’s buffer is full, it sends pause frames back to the sender. Pause frames are targetted at a port or a MAC address. They instruct the sending port to stop sending for a while (a multiple of 512 on-wire bits). You can disable flow-control on manageable switches or most ethernet NICs but then you get a huge a mount of losses on the GbE to 100M path, to the point that any UDP-based or TSO/GSO transfer fails. For UDP it’s very common to see fragmentation and the stack cannot retransmit fragments, only full datagrams. For GSO/TSO, the interval between packets is too short for the switch’s buffers and results in only the 1st or 2nd segments to be received and the rest to be retransmitted, resulting in terrible performance. Thus it’s still better to keep fctl on and slow down the whole sender. The best option is to ensure that slow speed devices never talk with shared high speed ones.
I agree with this, but simply having a 100Mb device on the network has no impact on devices it is not talking to. So if you have a GbE NAS talking to a GbE workstation, having a 100Mb device elsewhere in the network has no impact on the two GbE devices.
Of course you wouldn’t want a NAS running on 100Mb, but there is no reason not to have 100Mb or even 10Mb devices elsewhere on the network.
The other issue here is that you should not be using UDP over a speed mismatch. Use TCP, it will do just fine crossing the mismatch. Why are you sending high volumes of data with UDP? I also don’t run with jumbo frames, the performance impact is negligible.
If you are going to blast UDP jumbo frames at a server, they will have to be speed matched and have a reliable network connection.
I’m not talking about the device’s presence, but its traffic. Your thermostat isn’t going to pull lots of traffic. The video player will. If it pulls data from your GbE server and you do not disable flow-control on the port, this last one sends pause frames that forcefully limit the server’s port ability to send anything faster. All non-manageable switches have fctl enabled. You could disable it on the server, but if you have two such switches in series you’re doomed again because the network is fucked on the link between them.
And no, TCP does absolutely not help here, because pauses are very short and do not cause losses nor retransmits. That’s precisely why they are good for a healthy network in general as they avoid losses by queuing in the sender instead. TCP only helps over UDP when you can disable flow control and prefer to rely on losses and TCP’s ability to retransmit. But even then, just look at the number of losses and your CPU usage when pulling from a GbE port to a 100 Mbps port through a switch. You’ll notice the window cannot grow and your sender is woken up every two frames or so, resulting in interrupt storms and higher CPU cycles used than when transmitting over GbE.
Because this isn’t for a thermostat? I mean, for crying out load, it’s a quad core Cortex-A53, not a single core Cortex-M3. You’re comparing apples with blueberries at this point…
On top of that, you can’t stream high-quality 4K content over 100Mbps, which makes this a useless device for at least a large part of its target applications. A GMAC can’t have costed that much silicon space to implement in this product.
>A GMAC can’t have costed that much silicon space to implement in this product.
Making this a cost down option means integrating a PHY so the BoM cost is less. I don’t think integrating a gigabit phy is easy or cheap and that’s why the integrated PHYs are always 100mb.
I think it’s instead due to slightly incompatible process being used for GbE PHYs which require fast analog signal processing, which is why you usually still have the GMAC inside the SoC but not the PHY.
That’s basically what I said. If it’s not because adding a gigabit phy would involve a second die in the package or a more expensive process for the main chip then it’s because AllWinner think limiting it to fast ethernet will stop it denting sales of their more expensive chips.
None of their chips have an integrated Gigabit PHY, only MAC, but whatever…
>None of their chips have an integrated Gigabit PHY, only MAC, but whatever…
Who said they did? Integrating components that would otherwise add to the BoM cost is another side of “cost down” parts.
You did.
Trust be told im still puzzled why allwinner still make this kind limited use case of SOC.
They provide 4K decoding VPU but network bandwidth dont support it.
Or in H6 case provide PCI-E, but unusable.
I miss the good old days when they produce a better SOC than the competition, with a clear use cases.
>Trust be told im still puzzled why allwinner still make this kind limited use case of SOC.
Because they have already developed/licensed the IP and they can apparently still sell parts based on them.
Personally I don’t mind much. Even old cores like the A7 are useful for things I want to do and companies like AllWinner rehashing that old tech over and over means we now have some very cheap A7s out there. /me is hoping for an S3 like design with the PMIC die integrated as well.
Willy NanoPi R1S Look up.
Ah interesting, these are an improvement over the R1. And they finally do it with an H5 too. I know someone who could be interested. It’s fun to notice that their enclosure looks exactly like the GL-iNet but flipped left-right. Thanks for the tip!
Neat. I asked FriendlyArm for something like this for a router.. too bad the LAN port is using a USB 2.0 interface so its pointless.
An AtomicPi + USB 3.0 Gigabit adapter is a few dollars more, has full Gigabit performance and full x86 software support.
I think they should use a switch chip on the board to provide 2 GbE ports from a single one. At least it would have GbE routing speed in one direction (which is what most people are looking for). It would be slightly more expensive but could be worth it since we always expect to pay a little bit more for multi-ports.
“There’s no a cost-down version called”. hey jean..i think “now” is misspelled somehow.
they should have naming scheme like Allwinner H2 & H3
If history repeats this will only cause trouble.
Rockchip released the the rk3188 @ 1.6 later quietly later releasing a slower rk3188T @ 1.4. Companies sold the slower SoC in devices while claiming rk3188 @ 1.6 spec ( they lied ).
Now we have Allwinner H6, H603 ( slower cheaper version ) and already some sellers list H603 with H6 spec.
Add now the H6 vc200 os. What could possibly go wrong ! ?.
Looks like a version of the h6 is used in the Mavic Air 2 drone