Hmm, is it like, DP over the unused OCuLink pins?
This is the project page of my board btw GitHub - Terrails/FW16-OCuLink · GitHub.
I need to get around to pushing the latest revision that was shipped out to people, but the one currently on there just works as well.
no you just inject DP. Framework has a DP link on the … expansion bay interface.
You solder a DP connector so you can plug your gpu into the framework. Let the DP mux do the work.
That monstrosity I posted above seem to indicate.
Ah you meant just a DP port. I did think about that, but I’ll have to switch over to 6 or maybe 8 layers.
The only issue is how the DP mux would actually work. I am pretty sure it still needs OS support and can’t just be simply switched by injecting a signal.
So I’m in the process of moving which makes finding time for things hard, but things should ease up soon.
The 8i board with my RX 9070 XT is magic. Still some more benchmarking methods to figure out but everything is working great with the SFX PSU setup so far.
This shot is again with my eSSD installed.
The GPU stays nice and cool. Just one more side to create models for.
On another note: latest Linux kernel/Proton updates are amazing. I switched away from Windows when I first got my FW16 and I miss it less and less almost every day.
Enclosure is looking awesome my guy. I am jealous. And I hear you about proton. I’ve been daily driving CachyOS for a few months now and I love it. Rarely go back to Win11 now. Only keep the dual boot if there is a picky dev who doesn’t support linux. Them bastards 
Can’t wait to be back in Germany in September to print this thing out. Will just have to see where to get the two side panels.
@Van_F @Pierre_Cruse That’s the same issue I was having. Make sure you are using the 3.07 bios and not upgraded to 4.
Thanks!. I was on bios 3.05. I updated to 3.06 as i cannot see a 3.07 version from this link Framework Laptop 16 AMD Ryzen™ AI 300 Series — BIOS & Drivers | Resources
Finally moved and got some things up and running. Currently running a bunch of benchmarks on the 8i board. Since the enclosure is working, I am not focusing on the final 3D prints for it. I want to try to get through the benchmarks as quickly as I can.
Anyone try the newest 4.x BIOS on the 7840/7940 board to see if it works with the 8i board?
Not sure if you still are having issues but this batch script fixed it for me, you can download it here
unfortunately its not an error 43. I wish it were that simple. I appreciate the link tho,
I am on 4.05 with a 7840HS and it has been working the past months at x8 flawlessly. I did have to disable SmartShift with Smokeless UMAF or my CPU would get locked to .55 GHZ if i tried x8 (although this did happen on 3.xx bios). Other than that its been perfect.
That’s exciting! Once I wrap up my benchmarks, I will try updating to 4.05 as well.
That’s giving me some hope. I don’t have my eGPU setup with me here, but can’t wait to test the BIOS in a month to see if it resolves the issue for me as well.
Benchmarking is going slower than planned. It’s not an easy thing, and kudos to the folks that do this for a living.
My initial results comparing the 4i to 8i configurations were just too far apart. In some cases, the 8i was up to 26% faster. I reran the numbers repeatedly, and finally realized what was happening. I had updated Ubuntu from 24.04 to 26.04 and had not reinstalled all-ways-egpu.
This meant that, even though I had a monitor connected directly to my eGPU, the frames get rendered on the eGPU, passed back to the iGPU, then back to the eGPU. That is a lot of data going back and forth, and obviously the 8i setup is going to handle it better. Even in the 8i setup, that’s not great. There were also some really weird results where the 8i actually performed worse than the 4i. Not sure how that’s possible, but I ran the tests repeatedly and double-checked settings.
After running all-ways-egpu, the 4i numbers jumped up quite a bit (as expected). I’ve finished running some of the benchmarks on 4i again, but will need to re-run everything on the 8i setup as well. At least things are going more smoothly now that I’ve got things figured out.
I’ll be keeping all of the data from all of the runs, including running on the internal display, external display without all-ways-egpu, and external data with all-ways-egpu. Even data that isn’t meaningful can still be interesting. 
It is a tedious process you certainly don’t want to rush. Have to make sure all variables remain exactly the same (excluding the 4i vs 8i of course). It took me a while when I was benching between the two on my end as well, so I feel your pain. Now imagine to be a content creator who makes videos out of it all 
This may finally be getting fixed somewhat soon but for your current testing definitely run the compositor on the egpu for best numbers. With occulink there isn’t really any benefit to running the compositor on the igpu as you can’t hotplug anyway.
