Right now AI support on AMD is officially only on specific models. But they are working hard to turn this around to have broader support. And making progress.
Vulkan compute is also getting some good press as a local llm platform (at least on the linux side), will be interesting to see which crosses the line to "can ship production quality apps on this" first.
I got to see this in person at pacificon a few weeks ago. Also the creator is my friend from UIUC who I consider a brilliant rf/DSP engineer.
The demo was able to show and end to end tx chain from gnuradio to a receiver.
Really excited to see this! As there are a myriad of other things that this hardware can be used for as well.
We’re starting with the “Quad” tile — a 4 Tx × 4 Rx SDR designed for arraying — and expect to ship the first units toward the end of this year. They're actually quite capable as a standalone SDR. A Quad can interface directly with a Raspberry Pi 5, and we’ve built a combined enclosure for the SDR + Pi setup. You can run SDR software locally on the Pi or stream IQ samples over gigabit Ethernet to a remote PC.
Software support includes GNU Radio, Pothos SDR, and just about any tool compatible with SoapySDR. We’re also doing some fun demos, like visualizing Wi-Fi signal sources in real time ("Wi-Fi camera") and performing mm-scale 3D localization—a prerequisite for the automatic array calibration.
Larger arrays are assembled by simply tiling these Quads into an aluminum/PCB lattice framework, enabling anything from compact 4-antenna MIMO nodes up to 240-element lunar-bounce arrays. The goal is to have full phased-array capability by March 2026.
The broader vision behind open.space
is to make advanced RF and space-communications hardware open and accessible—so anyone can experiment with technologies once limited to national labs: moon-bounce (EME) links, satellite reception, terrestrial RF imaging.
Happy to answer questions here.
One thing I'm excited about getting working is mobile moon bounce!
Great question, the latest version has Tx RHCP, and then Rx either LHCP or RHCP controlled with RF switches (in each antenna). This allows point to point links (where the Tx pol and Rx pol should be the same), or "bounce links" where the circular polarization flips with the bounce. I should note RHCP Rx has a bit worse noise figure (LNA is different) but good enough for any line of sight.
There are ten 640 MSPS ADCs (I+Q per channel and a cal path, per 4-antenna PCB tile). These are custom MASH ΣΔ designs built from discrete diff pair transistors (cost about $0.08 each) and do noise shaping/decimation to get a clean 50 MHz of baseband bandwidth. The 8x DACs are also ΣΔ, using the LVDS pins of the FPGA and some modulating DSP. Mixers are MAX2850/1, LNA are custom design based on Infineon transistor, and RFPA is a Skyworks part meant for WiFi (have iterated on a few model numbers).
Copilot supports it.
And on windows you can try AMD tools like https://github.com/lemonade-sdk/lemonade to run arbitrary LLMs. AMD is actively developing support for the NPU.
I took my 7 and 6 year old and they loved it! We spent all our time looking at projects and booths. They have no interest in the talks.
With that said, there are not that many kids below highschool age that were in attendance. It is more geared to older kids/adults. But I do think there are lots of things to see there for kids. Probably the higher entrance fee reduces this a bit as well.
With that said, you want to pay close attention to your small children, as some of the exhibits are not super kid safe. But that is part of the fun!
There is a lot of variability in what you see as well. Some tables have incredible cutting edge projects, and other are exciting for a highschooler. Some are amazing highschool/middle school projects that the builders are really passionate about, but might not wow you technically.
Every year I find an amazing creator there, where I bring in their work to our house for the family to build/play with.
Last year we found https://www.trackstacker.net/ which has provided hours of fun over the last 12 months.
And this year we found https://professorboots.com 3d printed construction equipment.
As a hardware developer lvfs and fwupd is amazing. It has support for all kinds of standard update protocols. So if you build a product and use a standardized update mechanism, it is super easy to get updates on lvfs.
The other thing that is great is the testing and validation groups allows you to setup target groups so you can fully validate your updates internally before staging rollouts.
Also @hughsient is really responsive at fixing issues.
We have LVFS updates in lvfs testing but there is a bios capsule update bug that wipes the boot entries on update. Feedback from 3.07 was that some distros fail to boot after the update because they do not put a efi loader in the fallback path of the ESP partition. So we are hesitant to enable LVFS on 11th gen due to this issue.
The code that handles this runs from the previous bios, so we cannot fix this in the capsule.
We do offer a EFI update which you can use without booting into windows. https://knowledgebase.frame.work/en_us/framework-laptop-bios...
Right now AI support on AMD is officially only on specific models. But they are working hard to turn this around to have broader support. And making progress.