XuanTie C906 based Allwinner RISC-V processor to power $12+ Linux SBC’s

Allwinner XuanTie C906 RISC-V Processor

Alibaba unveiled Xuantie-910 RISC-V core (aka XT910) in 2019 for powerful SoC with up to 16 cores, but an update in 2020 revealed the company planned to have a complete RISC-V core family for a wide range of application from low-power microcontrollers to server SoCs.

At the time, I just assumed the company planned to keep their cores to themselves, but time proved me wrong as T-Head, the Alibaba subsidiary in charge of developing RISC-V cores, started to cooperate with Allwinner to develop open-source processors, which should lead to low-cost Linux capable RISC-V SBC very soon according to a tweet from Sipeed.

As far as I understand companies are still under NDA with Allwinner, so they can divulge too much. But here’s what the $12.5 Sipeed Linux RISC-V board specifications should look like based on public information available at this stage:

  • SoC – Unnamed Allwinner single-core XuanTie C906 64-bit RISC-V (RV64GCV) processor @ up to 1 GHz; 22nm manufacturing process
  • GPU – 2D accelerator only (similar to what is found in Allwinner V3s)
  • VPU – H.265/H.264
  • System Memory – 64MB to 256MB DDR3 (built-into processor); external memory version may also be available
  • Storage – MicroSD card socket
  • Video Output / Display I/F – HDMI, RGB LCD
  • Camera I/F – DVP and MIPI CSI
  • Networking – GMAC (Gigabit Ethernet MAC), optional WiFi and Bluetooth module
  • USB – USB host and OTG

I don’t have renders for the SBC, but the company will also make an Allwinner RISC-V system-on-module for other projects.

Upcoming Allwinner RISC-V SoM from Sipeed

Allwinner RISC-V processor will run the Debian Linux operating system, and the $12.5 price tag for Sipeed SBC makes it the cheapest RISC-V Linux board with MMU by far. You can already run Linux on RISC-V using Kendryte K210 boards, but those are not really practical due to the lack of MMU and low-memory, or on HiFive Unmatched offered for $665 and allows you to build a complete RISC-V PC.

The upcoming Allwinner powered RISC-V board from Sipeed does not come with a 3D GPU, and memory is limited so based on the information currently available, it may end being comparable to a Raspberry Pi Model A+ minus the ability to run any application relying on 3D graphics.

There’s still a lot of work to bring RISC-V ecosystem to the level of Arm, but it’s clearly a step in the right direction. Sipeed says the board will be available in one or two months, so let’s make that Q1 2021 to take into account potential delays.

Share this:

Support CNX Software! Donate via PayPal or cryptocurrencies, become a Patron on Patreon, or buy review samples

36 Replies to “XuanTie C906 based Allwinner RISC-V processor to power $12+ Linux SBC’s”

    1. China is going all-in on RISC-V because Trump’s unrestricted trade war made them belatedly realize how vulnerable they are to CPU architectures subject to US sanctions.

      1. not just China. Outside of the US, I would be wary of being overly dependent on US tech owners. With Arm being part of NVidia, any non-US country could get locked out.

  1. It could explain Allwinner’s tardiness in migrating from Cortex A53 to newer cores if they were considering divesting from ARM.

    No Mali = headless. For other use cases, expect them to pair with PowerVR, c.f. Allwinner A100

    1. They probably have the same limitation for PowerVR, and hope to have a more open GPU, there is no open source driver for PowerVR, only some blob and only for Android version of Linux. I hope we will see a RISC-V GPU emerging soon ?

      Nvidia spoke about replacing the microcontroler of the GPGPU from their own to a RISC-V version, but they could finally move to ARM microcontroller now they own it, depending on performances in both case ? Or they will be using ARM only for main processor ?? That’s the case of European processor for supercomputers and automotive, ARM processor and RISC-V accelerator.

      Rockchip has also delayed the new generation, from about one year. They probably working actively on RISC-V too ?

        1. Not a RISC-V anymore. The “developer” misunderstood what Open Source actually means in the large and decided to move to OpenPower instead. The stuff’s there, plausibly doable. But since he didn’t attempt to make anything WITH it, he was disappointed when they didn’t kow-tow and jump, johnny on the spot and integrate his ideas into the official frameworks. Doesn’t work that way. He may be onto something and you might be able to implement to his ideas…but he’s not even trying to work within the community. Kind of like NVidia or a few others.

          1. A fair description of the LIbre-SOC nee Libre-RISCV hybrid GPU project.

            However, the Think Silicon GPU referred to above is completely different. They are an existing GPU company and took their existing proprietary ISA and replaced it with RISC-V plus about half a dozen custom instructions. They were showing off the actual thing running FPGA at the RISC-V Summit in December 2019. I talked to them, volunteered what their custom instructions probably were (inverse square root approximation, texture interpolation, …) and they smiled and showed me a disassembly of some of their graphics code … and yup. They said they spent six weeks from start of the project to having a working prototype.

            Very VERY different to lkcl and crew.

        1. How? Mips never has been known as a super performer, at least not after the sgi workstation times (but that was before my interest in IT spawned, so can’t tell)

          1. Heh. The odds are solid you have a MIPS in your router. “Never been known as a super performer…” X-D

  2. There is one detail here, that wasn’t mentioned anywhere else before and makes a world of difference: PoP DDR3 memory. Without it, this upcoming board may be a good alternative to RPi Zero, as many said so before me, but with it, even at 64MB, it might dethrone V3s for projects that require a bit more processing power (C906 seems to have V extension, which is so much better than NEON, that ARM made SVE in its image), rather than video processing.

    1. I’m just been told there may be a version that supports external memory too, but not sure.
      It also looks like the initial version was designed with camera interface, but I’ve been told the final chip may not have it.

    2. >it might dethrone V3s for projects that require a bit more processing power

      Is there anyone using the V3s in projects? AFAIK it was discontinued (there are tons available though) and the replacements are all BGA packaged.

      If this is QFN or QFP packaged it will be interesting. Especially if it comes in below my current record for a ~1GHz Linux capable chip with memory at $1.6 for a single core or $3 for a dual core.

        1. Sigmastar SSC325 is a single core A7 + 64MB of memory. ~$1.6
          There are dual core and 128MB versions but they don’t seem to be so easy to get.

          Taobao link:
          https://item.taobao.com/item.htm?spm=a230r.1.14.42.4c8977969osNUw&id=610502496993&ns=1&abbucket=5#detail

          Product brief is here:
          https://github.com/fifteenhex/linux-chenxing/blob/master/infinity6/ssc335_pb_v03.pdf

          Sigmastar SSD201 is a dual core A7 + 64MB of memory, the SSD202 is slightly more expensive but has 128MB of memory. ~$3

          Taobao link:
          https://item.taobao.com/item.htm?spm=a230r.1.14.27.390b4a8apLIZnI&id=629244393880&ns=1&abbucket=5#detail

          Product brief is here:

          https://github.com/fifteenhex/linux-chenxing/blob/master/infinity2/SSD201_pb_S_v01.pdf

        2. F1C200s, for often less than a single dollar you get latest chip with ARM9 and 64MB of RAM, previous 32MB variant was featured in first version of Linux-running business card (https://www.cnx-software.com/2019/12/27/this-business-card-is-a-3-linux-computer-powered-by-allwinner-f1c100s-soc/). I think that once someone from Cedrus project (open video codec drivers for Allwinner chips) decides to work on variant for these, they may become an interesting alternative to beefier microcontrollers and current V3s/S3 projects.

          1. >F1C200s, for often less than a single dollar you get latest chip with ARM9

            Personally I would avoid putting ARM9 into anything new.

          2. I see nothing wrong with the ARM9 if you don’t need an FPU or high performance. It will work much better for running Linux than an MMU-less Cortex-M, and at least as well as the non-ARM chips in the <$5 price range.

            It seems that Allwinner agrees with you though, if the new chip hits the price point of the F1C200S, it won’t just upend the market for ARM9 but also MIPS32 (Ingenic, Mediatek) and single-core ARMv7 by being faster and cheaper.

          3. >I see nothing wrong with the ARM9 if you don’t need an
            >FPU or high performance.

            Do you think it’ll last much longer in the kernel? I can see support for Cortex A stuff hanging around for a while because the big players are still shipping those.

            >it won’t just upend the market for ARM9 but also
            >MIPS32 (Ingenic, Mediatek) and single-core ARMv7
            >by being faster and cheaper.

            I think if there is a QFN version then it will be a killer product.

          4. We added SAMA9X60 and F1C100s support last year and now SD5203. There is a large installed base of AST2400, imx23/28 and LPC32xx that will keep needing updates, as well as hobbyists running modern kernels on e.g. Kirkwood and DaVinci chips.

            I would hope to retire StrongARM and Xscale eventually, but there are still enough users that these hang around for now. I did queue the patch to remove ebsa110 for linux-5.11, and RiscPC will die as soon as the kernel requires gcc-9 or higher.

            SAMA9X60 is probably the most important ARM9 at the moment, and I’d expect at least 10 years before we can consider phasing it out of mainline. The timing will depend a lot on how Microchip keep marketing it, and how quickly the extreme low-end market gets absorbed by RISC-V.

            I don’t expect to see a lot new ARM9 chips getting released, but we might still add support for older Nuvoton or Mediatek chips if someone does the work.

            For phasing out ARMv7, my best guess is that this happens on a similar schedule around ten years later (Cortex-A8 came 10 years after ARM920T, Cortex-A7 came 11 years after ARM926E), so maybe another eight years until the last ARMv7 chip gets released, plus ten more years of products getting sold based on them. Less if RISC-V takes over that entire market, more if Arm comes out with newer 32-bit cores.

  3. For that price I’ll for sure buy it.
    I don’t think RISC-V will quickly be able to compete with ARM. I think it will take a long time.
    Maybe if NVIDIA has taken over ARM and destroyed everything. RISC-V will be ready to compete. ARM is also still evolving quickly, so to catch up it’ll be very hard and would need a lot of investments.
    I wonder how the performance per Watt will be. That’s the most important thing for the future.

    1. If they’re in keeping with the domains that VexRiscv and Rocket cores are, it’s 15% faster, 15% smaller, and 15% lower power than the comparable Cortexes in each respective class.

    1. The Pi Zero is not really $5. Try ordering 100 of them at that price. You can’t. It’s a loss leader, one per customer per order with $20 shipping.

  4. Single-core CPU, and 64MB to 256MB DDR3 (built-into processor) … hmmm, very, very embedded, IMHO.
    ONly if it indeed runs Debian, and has a network interface, then I want one just to see how usable it is.

Leave a Reply

Your email address will not be published.

Advertisement
Advertisement