Last month, Collabora exhibited at Embedded World 2025 in Nürnberg, Germany. As with previous years, we brought along a substantial number of demos to showcase our work.

These demos highlighted our projects with customers, such as:

• Platform enablement, including Bosch Power Tools
• GStreamer integration, with Audinate and ActionStreamer
• Machine learning demos running on the Renesas RZ/G2L, MediaTek Genio 700, ST STM32MP2, and Radxa Orion 06
• Wayland-rendered HDR10 video
• Open Source XR remote-rendering with ElectricMaple
• Our work to push forward the Open Source Vulkan NVIDIA driver, NVK]

One demo that we have shown and evolved over the last few years is our Board Farm. It allows us to feature not only the work on the platforms that we choose to include in the farm, but also the work that we do to grow and improve the automated testing available to the open source community. This includes our efforts to improve the reliability of testing in the Linux Kernel and bringing the power of pre-merge testing to Mesa CI. We hope this work will help drive further confidence in the reliability of the open source stack.

Controlling the farm

As with last year, the farm showcases several open source tools we use to connect the devices under test, orchestrate the testing process, and enable the continuous integration process. Collabora either contributes to or in some cases actively develops and even maintains these tools, including:

• LAVA automated testing framework: LAVA manages the testing process; schedules test jobs on devices; manages the artifacts required to perform the tests, and collects, stores, and displays test results and logs.
• Boardswarm: An increasingly mature and featureful abstraction layer for interfaces and equipment used to control development boards. Collabora is developing Boardswarm as a simplified means for integrating boards into systems like LAVA, but also to provide a convenient way for developers to access and control boards without having them on their desks or even in their vicinity.
• PDUDaemon: A daemon used to interface with various power control devices, such as PDUs (Power Distribution Units), smart switches, and relay boards. We utilise this from Boardswarm.

Devices under test

The number of boards in the Board Farm increased a little from last year. A few boards or SoCs (System on Chips) have returned for the second, or even third year. This highlights our continued involvement with these platforms through active upstream enablement and/or helping clients build their solutions around them with the upstream support.

• BeagleBoard.org BeaglePlay (TI AM625 SoC): We continue to support a number of customers developing products using the AM625
• Ezurio Nitrogen8M Plus ENC (NXP i.MX8MP SoC): We also continue to support a number of customers using the various i.MX8M platforms
• Radxa NIO 12L (MediaTek Genio 1200 SoC): Emphasizing our work to improve upstream support for the Genio SoCs with MediaTek
• Radxa Rock 4D (Rockchip RK3576 SoC): We have been working on upstream support for this relatively new SoC
• Radxa Rock 5B (Rockchip RK3588 SoC): Collabora has continued its effort to mainline support for the RK3588 SoC
• ST STM32MP157C-DK2 (ST STM32MP157C SoC): Highlighting our long running work with the Etnaviv driver on this and other platforms.

The booted software

We use a Board Farm like this to enable testing of software, be it the latest changes to the upstream kernel, Mesa, or other projects such as Apertis.

This year, devices in this demo were running the v6.13 or newer kernels. The majority of any patches applied to these kernels are part of our development work to upstream support for the relevant platforms. We look forward to these patches being part of future upstream kernel releases. The user space is built directly from Debian unstable (Sid) using debos. These images use the upstream Mesa graphics stack, with a Weston graphical environment. The majority of the boards were running glmark2 (Wayland ES variant), with the one board currently without an upstream working 3D GPU running a demo based on weston-flower. This last board stands as a reminder that whilst big steps have been made in recent years in the development of open GPU drivers, there are still platforms out there that don't yet have open drivers. Thus care is still required when choosing hardware if upstream support is something that you care about.

Whilst clearly a bit contrived, this simple test does go a little way to show that we are thankfully reaching a point on many SoCs where we can run a far more open software stack. This helps to ensure that its users will be able to easily update their software to meet their obligations under the various right to repair and cybersecurity laws that are now coming into force as and when required.

Other notable changes

Last year the Board Farm transitioned from sitting on a cabinet to being wall mounted on an IKEA SKÅDIS pegboard. This increased its visibility and allowed for a more visually appealing layout. However we did not make great use of the mounting options afforded to us by this option of board. This year we decided to rectify that by 3D printing brackets for each of the elements mounted on the board. The designs for these brackets we have are now available in our Embedded World 2025 GitLab repo.

This approach had a number of benefits:

• Mounting the devices and other equipment was quick, with each element being able to be easily positioned (and in fact repositioned as we evaluated different layouts).
• As we completed wiring and debugged a few integration bugs (as this was the first time that the full demo had been together in one location), the mounts enabled devices to be easily lifted off to access them better.
• The pegboard was in good condition once the show was over and we hope to use it again in the future, saving a little more from going to waste.

It was interesting to see that we were not the only ones using such a pegboard. Our friends at Ideas on Board also chose to use SKÅDIS pegboards with 3D-printed brackets as part of their display. This gave us a chance to compare our implementations and discuss the design requirements, which intriguingly led us to create quite different solutions.

In the past we've switched the devices on and off with a smart power strip, switching the mains supply to their power supplies. One downside to this approach is that some devices can take a long time to pull all the charge held in the capacitors in the power supply, thus requiring a long time to truly turn off. The explosion in consumer grade smart kits has also seen an increase in the accessibility of smart power bars, however the market is now full of Wi-Fi enabled options. Whilst relatively easy for the average consumer to use, these generally aren't as reliable as the USB or Ethernet-controlled bars which preceded them. This loss of reliability is especially likely to be true in an environment such as Embedded World, with the presence of large numbers of Wi-Fi enabled devices.

To combat this, we used a few higher-power industrial-grade power supplies and switched the low-voltage supply to the devices instead, using an off-the-shelf relay control board. This worked reasonably well, though a noisy board and one which ideally needed a USB PD power supply complicated the situation a little. Thankfully we were able to resolve that this year with a 12V supply and a couple of USB car chargers for these boards.

We already have some interesting ideas in the pipeline that may provide a better solution for this, as well as some other challenges that we see when wiring up boards for a Board Farm. Hopefully these will come to fruition in the next year and we'll be able to demo them at Embedded World 2026!