You can get the GPU module. But also order the empty expansion bay shell as the card Josh is making will be installed in the empty shell. The configurator has the option to order both when ordering the laptop.
If you only order the empty expansion bay shell (no dGPU module), you would need to order the interposer which comes with the dGPU in addition to the laptop as the empty expansion bay shell comes with a different, wrong interposer.
Thanks for that info.
I was going to get a Minisforum DEG1 eGPU Dock because I only had plans to use it with oculink, but I’ll need to get something else that works with oculink and thunderbolt so that I can use it now.
Any suggestions?
Oh interesting. I’m surprised no one has come out with a dock that has both USB4 and Oculink
This is one AOOSTAR AG01/ AG02 EGPU Dock with OCulink/USB4 Port，Built-in Huntkey 3 – AOOSTAR Mini PC
That one is nice. Sold out though
You can actually just build one yourself, and it doesn’t need to cost much at all.
https://www.reddit.com/r/eGPU/comments/1djnvcx/modular_inexpensive_usb4_or_oculink_egpu_its_alive/
I am going to be making an update to that post soon, but I do want to point out that my 3D Mark PCIe benchmark result is significantly faster than 19Gbps (as someone else mentioned in this thread). 3D Mark is reporting ~3.95 GBps (GB, not Gb). The ASM2464PD USB4 chip provides way more bandwidth that Intel’s TB3 chips.
This is my V1 eGPU setup. This is an OcuLink PCIe board that connects to a USB4 NVMe enclosure that uses the ASM2464PD controller.
The orange colours in the FW spacers and the 3D printed parts of the dock obviously don’t match, but they are also not as different in person as they appear in this photo.
Despite the PD in the chip name, it does not actually do power delivery. I did experiment adding a Thunderbolt 4 hub that delivers 96W to try and get a single cable solution. Surprisingly, it worked. Unsurprisingly, my bandwidth dropped from 3.95 GBps to about 2.95 GBps. That is just too much of a drop to justify the single cable convenience.
The great thing about this setup (other than how relatively cheap it was) is that if I do switch to an all OcuLink connector in the future, I’ll still have a very fast USB4 NVMe enclosure.
I’ve been experimenting with the M.2 socket my Pi5 going to Oculink, and haven’t been able to get that working, nor the oculink sockets on my desktop. I’m hopeful this dock will work better. I suspect my cables isn’t working right, do you suggest any cables in particular that you know work well? And for that matter Oculink->PCIe docks that you’ve tested and recommend? I’ll admit so far I’ve been bargain-bin shopping for my parts which doesn’t always lead to good quality parts. If I can get this all working, I can share my GPU between desktop, laptop and Pi … yes I’m a huge nerd 
I’m not suppose to install the graphics interposer until I get the 8x oculink module, correct?
If you do not have a GPU module, then the graphics interposer can not be installed, will not fit.
Right, I got this interposer cause that is what is needed to get the oculink module working.
So it has to fit with something.
One side of the interposer will fit on the mainboard, but the other side will not fit unless you have a GPU module, or the yet-to-be-released oculink module. So you can’t install it unless you have one of those.
I’m having flashbacks
Hah, that was my exact mindset when I was putting together this laptop.
I’m pretty sure that the graphics interposer won’t work until you have the Oculink Module to connect to. As I understand (correct me if I’m wrong) it will have the same connection on it as the graphics expansion would. So the graphics interposer will connect the main board to the Oculink Module’s board. The interposer that comes with the fan-bay only has the connections needed to connect to the electronics specifically in the fan bay.
Yep. That’s correct
Wait, OcuLink is only x4? So this can’t take advantage of the x8 that the Expansion Bay can offer? That’s unfortunate 
Oh, hm, but looks like it can, Amazon.com: Cablecc Oculink 8X SFF-8612 8611 to PCI-Express PCIE 16x Adapter with ATX 24pin Power Port for Mainboard Graphics Card : Electronics. Interesting. So yes, looks like there is 8 lane OcuLink.
So without me needing to read through the entire thread, the one being sold here uses the standard SFF8611 8i OcuLink type cable? Obviously cable and PCIe slot board quality will vary.
In doing some more reading, including PCIe For Hackers: External PCIe And OCuLink | Hackaday and When to use a PCIe retimer vs. redriver | Video | TI.com, does not sure if the retimer/redriver should be on the source or external board, but does this Expansion Bay have one?
I presume I need to modify the existing Expansion Bay shell? And of course buy the GPU interposer.
So costs are the ~$90 for the OcuLink Framework side, ~$40 for the interposer, tools & time to modify the expansion bay shell, and whatever it costs for cabling and external slot & PSU, so probably roughly $200-400 depending on specifics. Assuming the Amazon board & 8i cable I linked, that’d be roughly ~$80, plus PSU for whatever GPU you’re running (call it a 500W SFX, perhaps $50-100). And then of course 3D printed (I do already have one) mount/holder/case/etc for it all.
Does that all line up, generally, with needs/costs? Not including shipping/taxes.
Without reading through your whole post, Josh sells both an 4i and 8i version.
The ASM2464PD is also featured in the ADT-Link UT3G. I am running this with my RX6800xt on my Fw16. Why are you running this Adapter–>Adapter–>Adapter Setup? The UT3G is just a single USB4 40gbit out of the PCIEx16 Board
