OcuLink eGPU works with the Dual M.2 expansion bay module

that is mostly due to the problem of PCIe integrity. amd gpu are not great handling heavy PCIe error. how did your m2 adapter look like?. did you bent the cable? if so the signal integrity will be shit and for amd GPU that is not very good.

I tought about it! Same behaviour with 2 brand new M.2 → OcuLink adapters, and with two different 4.0 OcuLink cables (75 and 50cm).
The benchmarks are perfect, the bandwidth is 6.6GB/s, but I crash in fullscreen. Every game in windowed can run high fos for hours, but fullscreen fucks me. If it were PCIe degradation errors I wouldn’t have high frames and good benchmarks… Or would I? :frowning:

1 Like

if you have nvidia gpu lying around, try this in CMD: nvidia-smi dmon -s et -d 10 -o DT and check for errors in benchmark
If AMD, I honestly don’t know.
The 3dmark PCIE test is not accurate.

I ran a PCIE test with my 4.0x4 oculink, and the results are funny.

3 Likes

O.O quite an overkill config you got there!
Thanks, I’ll try to borrow a friend’s 3060Ti!

Hi all, I just wanted to let you know that this option arrived and I tried to use it, cannot really recommend. The reason is that it does not at all fold as flexibly as the picture suggests. It is quite rigid and you need to fold it twice to make it work, and the folds cannot be too tight for signal integrity. I think it makes it quite bulky, and also the folds cannot be too good for long-term signal integrity. Final point is that the two holes at the end are smaller than the other option from the start of this thread. So for now I believe the first option from the start of this thread is the best solution.

1 Like

That’s the one I use! I managed to fold it in such a way that it fits flawlessly. On my setup it is held down by a flexible clear plastic sheet screwed down on the M.2 board, with the port sticking out of a 3D printed adapter I designed on the original I/O cover.
I would share the details, but I need to get my setup working!
PS: I bought 2 of those flexible ones, for testing i’m using the non-bent one to avoid adding variables. Everything is running fine BUT games crash in fullscreen. Even on the latest 25.2.1. I’m running off the internal display. Please help!

1 Like

Hey, I don’t have too much experience with AMD GPUs, have been on Nvidia for a long time. I honestly suggest to post your experience in this forum here. PS: And looking forward to seeing pictures/3D printing files if possible!

1 Like

Hello, I installed an oculink nvme adapter to the dual m.2 expansion bay module and I thought I could share the results.



I made another Version of the io panel for the oculink connector that has prongs


It snaps on and holds onto the IO shield very tightly.

they both slide into the FW16 nicely.


The design is for following oculink adapter.
Amazon.com: NFHK 2280 NGFF PCI-E4.0 M.2 M-Key to Oculink SFF-8612 SFF-8611 Host Cable 15cm for NVME U.2 SSD Egpu : Electronics

Here are the stl files.
ioshield for oculink with prongs.stl (620.1 KB)
ioshield for oculink.stl (565.9 KB)
plugs_final.stl (204.6 KB)

25 Likes

This looks absolutly magnificent, thank you!

Very clean, well done. This looks exactly what I thought I would have done before my temporary solution became permanent…

2 Likes

Absolute perfection. This is why I have been checking this thread about once per day over the last months, and this is what I also wanted to do myself long-term. Already ordered the 3D print - thank you so much!

Now looking forward to OCuLink 8i solution at some point too. By the way, I found this OCuLink 8i EGPU on ebay, seems to be a good option (once we can use it).

PS: And here some cable suggestions: https://www.aliexpress.com/item/1005007939562515.html , https://www.aliexpress.com/item/1005006057139461.html

EDIT: And here another cable suggestion, this time OCuLink 8i to 2x OCuLink 4i (in case this becomes relevant for one of the boards being designd at a later stage): https://a.aliexpress.com/_EQ1wwcY

6 Likes

This is amazing! But…

I ordered from the linked Amazon product and received:

I still wanted to test it out, but hit a (literal) snag and once it was in my FW it didn’t want to come out. Pretty sure I damaged the cable now.

Sigh… I’ll update the model to accommodate the 4-hole part but won’t print anything until I figure out what the replacement part looks like.

2 Likes

Has anyone found a panel-mount Male to Female Oculink cable? I want to build one of those PCB adapter boards into a case, but none of them have the Oculink port pointing in any kind of reasonable direction to integrate it into a product.

2 Likes

I too had issues with this supplier, as I purchased two identical adapters yet I received different connectors on each.

I thought they had a minor mishap at the warehouse where they added two different products on accident, but now they seem to be shipping whatever happens to be in reach.

If your cable gets caught in the fw16 I recommend you tape it to the m.2 board and install it using a regular m.2 screw. As the one mount screw in the m.2 module is too small.

untitled

Bending the board shouldn’t be an issues since it doesn’t have any components on it.

Also when 3d printing I used PETG filament. It is more flexible than PLA so it doesn’t break when bending it into the slot.

7 Likes

I would like to mention some points regarding this solution for anyone else wanting to try this.
The Adapter ordered through Amazon was the right one for me. I printed the shield with ABS but I recommend printing more than one since on one the little tab at the edge broke while installing the connector. Be very careful and try not to use too much force.
Make sure to push the plastic prongs all the way on, I managed to cut the top off of one while installing the expansion bay.
I recommend removing the midplate, it made inserting the expansion bay way easier.
The Oculink → PCIe adpater linked at the beginning of the thread did not work reliably for me, lagging often and sometimes not working at all. Switching to a DEG1 everything worked flawlessly.
Thank you for your work with this!

3 Likes

And just one addition if I may as disclaimer, if you use the DEG1 note that it only works with the oculink cable that minisforum sends with the EGPU - so if you use this EGPU, you will not have any issues as it works pretty well, but from what I have heard so far from others is that only the minisforum cable works with it. So not possible to go for a longer one, for example.

I have had similar experiences with the Aoostar AG02. The cable that came with it is the only one that functions - two other vendors with longer cables were nonfunctional.

Just in case you get the adapter with the four holes.

Framework 16 Oculink IO Cover.stl (110.5 KB)

Sadly, this didn’t work for me, but I’m not sure what part is the actual problem. I might pick up a DEG1 to see if it works. I just need to narrow down where the problem is.

4 Likes

I had problems with the DEG1 but ONLY when using an AMD GPU (a 100% fine 6900xt. Heck, I even borrowed another 6900xt from a friend, same results). Had crashes, problems when using fullscreen, AFMF not working with several driver versions.
Using an RTX 2060, no issues.
Scrapped the project.

I’ll toss in another data point. I was really curious to try this out, so I picked up the adapter and a DEG1. I don’t have the m.2 expansion bay yet, so I tested it using the 2280 slot on the board with a 5070ti. I hope to test some more games and applications tonight, but so far it works just fantastically! Pretty great timespy GPU score and I played a round of BF2042. It honestly felt like I was playing on a desktop. Pushing a very stable 4k/120 with a mix of settings and DLSS set to balanced.

Update. I tested a few rounds of Overwatch and Helldivers 2 and the performance was fantastic.


6 Likes