i-Pi SMARC 1200 (MediaTek Genio 1200) devkit tested with a Yocto Linux image

Last weekend I received ADLINK’s i-Pi SMARC 1200 development kit powered by MediaTek Genio 1200 Octa-core Cortex-A78/A55 AIoT processor, checked out the hardware and wanted to install the Yocto Linux image but stopped in my tracks because it looked like I had to install Ubuntu 18.04 first in a Virtual Machine or another computer. But finally, the documentation has been updated to clarify “Ubuntu 18.04 or greater” is required, and I had no problem flashing the image from a Ubuntu 22.04 laptop after installing dependencies and tools as follows:

That’s it for the tools. Eventually, the development kit will support three images: Yocto Linux, Android 13 (July 2023), and Ubuntu 20.04 (Q3 2023). So that means only the Yocto Linux image is available from the download page at this time, and that’s what I’ll be using today. We’ll need to connect the micro USB to USB cable between the host and the board, as well as the 12V power supply, before running the command to flash the image:

Output from the command:

Once this is done I can connect the board to an HDMI monitor, an Ethernet cable, and RF dongles for the mouse and keyboard to access the Yocto Desktop.

i-Pi SMARC 1200 Yocto Linux glmark2 benchmark

The desktop is basically empty, but we can start a terminal from the desktop environment or SSH (root without password) to run some benchmarks:

Those include CPU benchmarks like Coremark and Dhrystone, memory bandwidth benchmarks (memcpy/memset), graphics benchmarks like glmark2, and some AI benchmarks (MobileNet V2) running on the CPU and GPU, but apparently not on the AI accelerator (TBC). Here’s the output from glmark2:

It’s difficult to make an accurate comparison since we don’t know all parameters used during the benchmarks, but for reference, a Coremark score of 102310.51 iter/sec is between the score of an AMD FX-8350 and an Intel Core-i7 2600 launched in 2011 as shown in the table below.

CoreMark Comparison

A Rockchip RK3588 system got 73178.23 iter/sec in CoreMark (Phoronix).

Let’s look at some system information:

We can see the root partition is only 3.9GB, but the module comes with a 64GB UFS flash.

So I tried to resize the root partition with fdisk, leaving some unused sectors at the end, just in case:

before running resize2fs:

The root partition looks to be resized fine, but this does not survive a reboot…, and I have to run aiot-flash to restore the image. I tried several times with different parameters, but no luck.

Let’s check Gigabit Ethernet networking performance with iperf3:




Everything looks good on this side.

I wanted to test the UFS performance with iozone, but the gcc toolchain has not been built into the Yocto Linux image. But I’ve noticed fio is installed in the image.

Sequential read speed

Sequential write speed:

542MB/s read speed is way faster than any eMMC flash I’ve tested, and 169MB/s write speed is also on the high side.

The i-Pi SMARC 1200 development kit also includes M.2 (PCIe) socket, so I inserted an M.2 2242 NVMe SSD into the Key-B socket for further testing.

i-Pi SMARC 1200 NVMe SSD

But the drive does not show up, and the command lspci only lists two PCIe devices:

It’s unclear what needs to be done to enable the SSD here. I can also see one person who got the kit with an M.2 WiFi module has some issues enabling it.

ADLINK i.Pi SMARC 1200 looks like a promising development kit with a powerful Cortex-A78/A55 processor, fast UFS storage, a 4.8 TOPS AI accelerator, and a good range of interfaces. But software support and documentation still need a lot of work, and hopefully, it will be worked on in the next few months. In the meantime, you can create your own custom Yocto Linux image for the i-Pi SMARC 1200 development board using the meta-layer on GitHub. The i-PI SMARC 1200 is listed for $379 on the i-Pi shop, but is currently out of stock.

Share this:
FacebookTwitterHacker NewsSlashdotRedditLinkedInPinterestFlipboardMeWeLineEmailShare

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

ROCK 5 ITX RK3588 mini-ITX motherboard

9 Replies to “i-Pi SMARC 1200 (MediaTek Genio 1200) devkit tested with a Yocto Linux image”

  1. Ah yes, just what you want, another half baked development “kit”. Why are all companies so shit at this and can’t supply software that actually makes the hardware useful? Or do these companies, including the original hardware vendors actually expect their customers to develop all the drivers they need before they can start using the hardware?

    1. Seriously this thing is almost $400 and broadly comparable to the rock 5b, which is less than half the price. All so you get the same half finished software?

      Don’t get me wrong I do like the hardware, but the price they ask is too high. They need to drop the price or rapidly upstream this board. The latter should not be terribly hard, as many of the drivers should be directly compatible with the Chromebook mt8195 drivers. But I’m not going to pay extra to work for free.

  2. I like hw on MediaTek genio 1200 (old of 1 year yet) but I prefer see it on rock pi 6 or raspberry pi 5.
    -better sf support
    -can oc via rasp os
    -lower price
    -bigger community /YT reviews
    -hw is good enough to support even Windows 10
    -you can install others os like pop!

    In short I hope in these two teams for a decent sbc with even updated hw like 4x A78 2.8Ghz 4xA55 or A57 2.5Ghz (g610 mp6 gpu) with lpddr4x 4266Mhz 128bit now that we are in 2024 I suppose that is the righ time…

      1. Saw it. But still on Nvidia handheld.
        A510/515/520 won’t see on next 3 years in sbc. Maybe RISC-V versions of old cores.

        1. And Nvidia handheld(i.e. Switch) is pretty old. A57 is an old big core, you will never see it paired with A78s.

          I don’t see why we wouldn’t see A510 appearing soon-ish on SBCs. It has already been 2 years since ARM announced it. Who knows, maybe the Raspberry Pi 5 will have a couple of them.

          RP 3 did launch about 2~3 years after the announcement of A53s too.

          1. > RP 3 did launch about 2~3 years after the announcement of A53s too.

            More like 4 years in fact (2012 for A53, 2016 for rpi3). But yeah, that’s not that much. A72 was released in 2016 and rpi4 in 2019, which is more like 3 years. It’s been 3.5 years without an rpi, which is the longest to date (OK, covid was there). In terms of dates and maturity, and based on the fact that each chosen SoC uses the CPU issue the year the previous board was released, A77 (2019) or A78 (2020) would seem a more likely candidate than A510/A710 (2021). We’ll see.

          2. I think the main thing stopping the development of new ARM SoCs that aren’t made for Android phones is ARM has jacked up the licensing fees: https://arstechnica.com/gadgets/2023/03/risc-y-business-arm-wants-to-charge-dramatically-more-for-chip-licenses/

            If you look at all the chip providers outside MediaTek/Qualcomm, there’s been virtually no exciting new chips since the RK3588 was announced two years ago (2021).

            If you look at the other SoC makers, there’s been virtually no progress in the last few years:

            • Allwinner (still only making A53 chips)
            • NXP (only released a few A55 chips in the past few years)
            • Amlogic (haven’t released anything more powerful than the S922X-B/A311D back in 2019)

            Looking at product releases in the last year or so, the A53 (now 11 years old) seems more popular than ever, which sucks as most of us want a more powerful A7X-based SoC.

            The only real jump I see in the community is if/when oDroid releases an RK3588 board which will help speed up the software/firmware side. Which it did for the RK3326, and then Ambernic and 20 other copycat companies started pushing out clones based on oDroid’s software dev.

            So TLDR; don’t keep your hopes high in the short term.

          3. Thanks for this article, this plus the $1.4 billion justice action from ARM against Qualcomm+Nuvia has good chances to boost the RISC-V adoption by many manufacturers.

            Replacing an unstable license model with an open-source one is quite a good thing (as far as very skilled people add interesting things to this standard).

Leave a Reply

Your email address will not be published. Required fields are marked *

Khadas VIM4 SBC
Khadas VIM4 SBC