Yeah, sorry about that. I’ll try to get the labels printed this weekend and then they should go out next week. I’ve just been waiting on feedback from the few EU people I sent the packages to first.
Oh no worries on your end! My remark was more about the state of the world when it comes to shipping - especially with the US. 
Now it’s a race to see if I can get a beta of the enclosure ready in time, hah!
1 Like
Well my feedback is not super great: I can run my laptop now with the board installed, but OcuLink does not work. I’m running a 7840HS with BIOS 4.03. I didn’t yet try Downgrading the BIOS or going x4, there is just too much life happening right now to tinker.
Ended up getting this one.
It is THE most expensive passive Oculink adapter, but it does go in effectively flawless. Top slot m.2. You might have to remove the m.2 carrier first, install this, and reinstall the m.2 carrier. It does fit, but getting it in there is a bit of a journey.
My work has been done by someone with complete lack of knowledge that it fit in the Framework 16.
A Foxconn connector, too. even though the PCB feel like $0.1 scraps off the table. Havent got to testing it yet. At some point, perhaps.
Oh and as a bonus it is upside down, so the clip is now on the bottom. Its the simple fact that you have to bottom-mount the Oculink connector. Not much you can do about it, and I dont think they make a reverse one.
https://www.digikey.com/en/products/filter/pluggable-connector-assemblies/443?s=N4IgTCBcDaIPYGMCuAbAlgOwNYgLoF8g
This looks almost identical to Kyle Tucks design but I thought
Hello everyone, been watching the topic for a couple of months and ordered the 8i board from Filip. I got a bit scared seeing people having trouble, so here are my findings:
First and most importantly, the board works! I’m running the Ryzen 7 7840HS main board, bought the AliExpress PCIe adapter and cable from the thread, and I’m using an AMD RX 9070 XT from XFX.
I’m running BIOS version 4.03.
Blender Cycles has no issues using the GPU, Unity’s lightmapper works too.
For real-world games, I only had time to play Final Fantasy XIV, which is not the toughest game on the GPU (it’s mostly CPU-bound), but I can report similar framerates to those on desktop in small areas (managed to get 240+ FPS in inn rooms).
PCI speeds, however, were reporting funny. I’m running Fedora 43, and both lspci and GPU-T (a Linux-compatible clone of GPU-Z) report the graphics card running x16 5.0@x16 5.0, which is probably not right 
I used djanice1980/GPU-PCIe-Test to get some real results and got… 3.5GB/s, which would indicate 4x speeds on 3.0…
Now, this could be a tool issue since it’s a fork of a Windows-only tool. I spent some time installing Windows 11 to try and run the test again, as well as using the real GPU-Z, and got more “in-line” results: 6.5GB/s and x4 4.0 reported.
Now, either boards or the cable might be at fault, the latter being the most likely imo. I might buy a more reputable cable and run the tests again but for now it’ll do.
1 Like
Hey, did you attempt to do the battery disconnect within BIOS? You just do that, unplug the power adapter, save settings and then the laptop turns off.
You then hold the power button for a few seconds to drain all power and leave it like that for a minute or two. Afterwards you can plug it back in and turn it on.
I had the issue of the board working at slower speeds previously. Usually happened when I was using x4 config and then went to x8.
its also almost identical to my layout, too. I havent finished routing yet, because i was trying to cram it all into a 2-layer, which simply isnt working.
It look crazy, because pretty much half of it is ground. I must ensure proper signal shielding. But that look pretty ugly. I dont think it will work.
And did I mention this is before length tuning?
Well, today I tried unplugging and plugging back in, battery disconnects, BIOS resets, changing switch configurations, and some combination of these changes to no avail unfortunately.
I did some digging on that AliExpress OCuLink cable and grew more and more suspicious of it for three reasons:
- The length: It’s 75cm long, which is a bit longer than the 50cm recommended for 4.0 links with a passive cable, or so I’ve read.
- Googling the serial number shows it is a spare part of an old storage server (circa 2017-2018? considering it’s using 1st/2nd gen Epyc) used to link the front drive bays to the main board, and the spec sheet only specifies Gen 3x8 speeds (Gen 4 spec was very new back then).
- Deep in the AliExpress description of the product, it does say something about some AMD processors only being able to run the board at 3.0 4x speeds because of signal attenuation, something something, hmm.
Apparently, the cable is supposed to be the same between an OCuLink 1.0 (3.0 speeds) and 2.0 (4.0 speeds); only the tolerances are tighter and max lengths shorter, meaning any cable should work, but not reliably.
Now I would think a poor cable would make the connection drop to 3.0x8 since all the electrical connections are there, just the speed can’t be hit due to too weak of a signal, but I’m way out of my domain there
I have yet to try rolling back to an older BIOS, but I’m kind of done debugging for the day I think haha.
Yes, 50cm maximum length is recommended for EGPU setup via OCuLink, according to all feedback in different fora I have read so far. Also, cables can indeed have compatibility issues, so you may need to try another one.
Hello, just to give some updates. I received today the board from @Filip . Fw16g2 hx370 and rx9070xt.
Mounted without issues on the clamshell, no problems with “Expansion Bay cable improperly seated”. Before doing it I’ve done the battery disconnect, just in case.
The only thing I can say is that boot time is slower, I mean just the init part, probably is the check in the bios for the module.
Tested on Windows11, GPU-z pcie4x8, 3dmark nomad stress test, furmark benchmark, no issues.
Tested also on Fedora 43, Mission Center pcie4x8, where using thunderbolt I had pcie1x1.
so for now it’s 
, just waiting to print the box from @Andrew_Shaffer when it’s ready.
4 Likes
@Andrew_Shaffer is it possible you could upload the 4slot gpu/psu print to your link? I’m less concerned about a full enclosure at present and would just like the assembly to hold everything in place. Assuming that part is complete?
Are you using @Filip 's OcuLink module? I just learned or its existence. My understanding is that at the 128Gpbs such a connection affords there shouldn’t be a huge difference between an eGPU and one installed directly.
Very few such devices so of course also very few benchmarks.
There are benchmarks above.
Yes, I am using his “prototype” he designed before making his revision and it works without issues. And as mentioned:
A few of us did reviews on this.
Performance on the 8i while minor, still shows improvement over 4i. The biggest benefit I see is I don’t have the bottleneck I did before (getting 99%-100% gpu utilization now instead of ~80% as on 4i). Benefits of egpu is more vram as the current availability of 7700/5070 are 8GB whereas I have 16GB with 9070XT.
1 Like
Also want to add @Andrew_Shaffer that the test print of the new fitment works on my version of the board. So good to go there 