Getting started with Ubuntu on MediaTek Genio

Ubuntu runs on MediaTek Genio through an official Canonical and MediaTek collaboration, but the documentation is spread across half a dozen pages and the board support is uneven. This guide pulls the path into one place: which boards are supported, how to get to a first boot, how to add the hardware BSP for video and NPU acceleration, and where Ubuntu differs from the IoT Yocto image. Every step links back to the official MediaTek and Canonical docs.

Which Genio boards run Ubuntu?

Ubuntu on Genio is a joint effort: Canonical maintains the kernel, the images, and the Ubuntu archive; MediaTek maintains the boot firmware and the hardware support packages. Deploying a board needs two pieces — a Canonical disk image and MediaTek’s board-specific boot firmware.

Board	Ubuntu status	NPU on Ubuntu
Genio 1200 EVK	Certified (22.04), image (24.04)	Full
Genio 700 EVK	Certified (22.04), image (24.04)	Full
Genio 510 EVK	Certified (22.04), image (24.04)	Partial (firmware + runtime)
Genio 350 EVK	Certified (22.04), image (24.04, server)	None (no NPU hardware)
Genio 720 EVK	Early Access (24.04)	Not yet in the BSP package set
Genio 520 EVK	Early Access (24.04)	Not yet in the BSP package set

The practical takeaway: for a production Ubuntu design today, target the 1200, 700, 510, or 350. The 720 and 520 are Early Access — the disk images exist, but the documented hardware BSP packages (video, NPU) cover the 350/510/700/1200 family. If you specifically need the Genio 720, the IoT Yocto image is the more complete path right now. For the broader platform trade-off, see our Genio vs Jetson comparison.

How Ubuntu on Genio differs from the IoT Yocto image

If you have worked with the Yocto IoT image, two differences matter on day one:

• Boot flow. Ubuntu uses the EBBR (Embedded Base Boot Requirements) U-Boot flow with the standard BL2 → BL31 → BL32 → BL33 stages. The IoT Yocto image uses FIT boot. The bootloader implementation is shared, but the path differs and is switchable through U-Boot environment variables.
• Software model. On Ubuntu you start from a Canonical-maintained kernel and archive and add hardware support from an apt PPA. On Yocto you build the whole image from the meta-mediatek BSP layers. Ubuntu is faster to stand up and easier to maintain with apt; Yocto gives you deeper control and, as we will see, a fuller accelerator stack.

For the full decision between the two, see Ubuntu vs Yocto on MediaTek Genio.

First boot and login

Flash the Canonical image and MediaTek boot firmware to the board’s eMMC (a single eMMC device is shared between the bootloader and OS partitions), then power on. On the first boot, Cloud-init creates the account and initializes the system:

# Wait for Cloud-init to finish before doing anything else
cloud-init status --wait

Default credentials are ubuntu / ubuntu. Console access is over the UART (ttyS0). If the serial console looks misaligned, fix the terminal geometry:

sudo apt install xterm
resize

Then refresh apt before installing anything:

sudo apt update

One display caveat to plan around: native X11 and GLX are not available — X11 applications run through XWayland on the Wayland session. If you are porting an X11 app, expect to go through XWayland rather than a native X server.

Installing the MediaTek hardware BSP

Out of the box, the Ubuntu image boots and runs, but the hardware video codec and NPU are not active until you add MediaTek’s BSP from its public PPA:

sudo apt-add-repository ppa:mediatek-genio/genio-public
sudo apt update

Press Enter when prompted to confirm the repository. The PPA carries two families of packages — video hardware support and NeuroPilot NPU support — maintained by MediaTek under proprietary licenses. Internet access is required during install. For testing unreleased fixes, MediaTek also publishes a ppa:mediatek-genio/genio-proposed channel; keep that off production images.

Hardware video codec on Ubuntu

The video stack is a user-space daemon plus patched GStreamer plugins. Install the daemon for your board, then the patched plugins:

# User-space VPU daemon (pick your board)
sudo apt install mediatek-vpud-genio1200    # or -genio700 / -genio510 / -genio350

# Patched GStreamer plugins — install the versions carrying "+genio23"
sudo apt install gstreamer1.0-plugins-base gstreamer1.0-plugins-good gstreamer1.0-plugins-bad

Unlike the Yocto image’s element naming, the Ubuntu BSP exposes the codec through standard V4L2 GStreamer elements:

• Decoders: v4l2h264dec, v4l2h265dec, v4l2mpeg4dec, v4l2vp8dec, v4l2vp9dec
• Encoders: v4l2h264enc, v4l2h265enc
• Conversion: v4l2convert (the hardware decoder emits the NV12_16L32S format, so a convert stage is usually needed)

A hardware H.264 encode from the test source:

gst-launch-1.0 -v videotestsrc num-buffers=300 ! \
  "video/x-raw,format=NV12,width=720,height=480,framerate=30/1" ! \
  v4l2h264enc extra-controls="cid,video_gop_size=30,video_bitrate=512000" ! \
  "video/x-h264,level=(string)4.1,profile=main" ! h264parse ! mp4mux ! \
  filesink location=/home/ubuntu/test-h264.mp4

Hardware decode to a Wayland desktop:

gst-launch-1.0 -v filesrc location=/home/ubuntu/test-h264.mp4 ! parsebin ! \
  v4l2h264dec ! v4l2convert output-io-mode=5 ! \
  video/x-raw,width=720,height=480 ! waylandsink

On a server (headless) image, swap waylandsink for kmssink connector-id=<id> driver-name="mediatek". For the deeper GStreamer treatment — the VPUD daemon, the 4K encoded-buffer-size gotcha, and pipeline patterns — see real-time video on MediaTek Genio.

NPU and NeuroPilot on Ubuntu

This is where Ubuntu and Yocto diverge most. The Ubuntu BSP gives you the firmware and the Neuron runtime, but a narrower toolchain than the Yocto/RITY image.

Install the APUSys firmware for your board, then the runtime:

# Firmware (pick your board)
sudo apt install mediatek-apusys-firmware-genio700   # or -genio510 / -genio1200

# Runtime + dev libraries (all NPU-capable boards)
sudo apt install mediatek-libneuron mediatek-neuron-utils mediatek-libneuron-dev

What you get: the TFLite Neuron Delegate, the Neuron Runtime, development headers, and benchmark models under /usr/share/benchmark_dla/. Verify the accelerator is alive:

sudo vpu5_test -a ks -l 10                       # should report PASS
sudo python3 /usr/share/benchmark_dla/benchmark.py --auto

Two things to plan around:

1. Board coverage is uneven. NPU support is full on Genio 700 and 1200, partial on Genio 510 (firmware and runtime only), and absent on Genio 350. Match the firmware package to the board exactly — MediaTek’s docs warn that the wrong firmware can break the video codec drivers.
2. The toolchain is narrower than Yocto. MediaTek’s Ubuntu NPU docs cover the TFLite delegate path, but do not document the ONNX Runtime NeuronExecutionProvider or the offline ncc-tflite compiler. If your inference stack is built on ONNX-on-NPU, the Yocto image is the documented path — see ONNX Runtime on the Genio NPU and on-device AI on Genio.

Customizing the image

For anything beyond the stock image, MediaTek documents customization of the kernel, device tree and overlays, bootloader, and partition layout, plus single-partition flashing so you can update one partition without re-flashing the whole eMMC. These are the entry points for bringing up a custom carrier board or module on Ubuntu rather than the reference EVK.

Where to start

If you are evaluating Ubuntu on Genio: pick a certified board (1200, 700, 510, or 350), flash the 22.04 or 24.04 image, add the genio-public PPA, and install the video and NPU packages for your part. Treat the 720 and 520 as Early Access. If you need the 720 specifically, or the ONNX/ncc-tflite NPU toolchain, the IoT Yocto image is the more complete route today.

Bringing up Genio on Ubuntu or Yocto for a product, and want it done on a fixed timeline? See our MediaTek Genio support services.

Sources: Ubuntu on Genio (Canonical) and MediaTek’s Ubuntu on Genio documentation.