Hey there!
Some of the lessons you learned were already shared in this thread. 
Are you leaving the ribbon cable hanging out the back? If you check the first post, you will find links to some cleaner solutions.
The early AG02 models had 400W power supplies. The newer ones have 800W. I would be surprised if 800W wasnât enough.
Hi!
Thank you for your comment, even if shared, as a total beginner I think itâs hard to figure out just watching the thread, I had to do my own tests to realize and understand how to proceed exactly.
About the AOOSTAR, the one I bought is the latest from Amazon:
ăAOOSTAR AG02 EGPU Dock ă The AOOSTAR AG02 external GPU docking station has a built-in 800W power supply. The maximum power consumption of the GPU is 600W, so it basically meets the needs of most existing graphics cards. AG02 eGPU dock adopts an open metal body design with aluminum profile center frame, size 22.5 x 11 x 6cm, weight 1.5KG, and unlimited graphics card length.
Iâm going to return it because it doesnât work, and Iâve also noticed that itâs a bit less stable than the DEG1 configuration, if I understood correctly.
Now that my setup is working well and I understand much better how everything works, Iâm going to review the other cable solutions in detail and choose a cleaner option if possible 
Thanks!
Yes, have a look at the 3D printing solution by @Morkale (see overview right at the beginning of the thread). Your current m.2 connector is compatible.
Ah, and just some other tips for the DEG1 in case you would like to work on it a bit more. The following other âmodsâ exist:
I got this working with the short internal m.2 port using the coaxial adapter and the deg1. Works great but I havenât done any benchmarks yet.
This is interesting to see from a technical perspective, but what is the goal? It doesnât seem practical unless youâre considering permanently routing to one of the swappable ports. Maybe Iâm missing something?
Iâm actually in the process of making my own case from carbon fiber as part of my dual screen laptop project. I havenât posted about it yet but I was showing off a framework 13 based prototype at open sauce. Iâm currently redoing everything to use the framework 16 mainboard. Once I have more progress I will post about it.
Oh wow! Iâm excited to see/learn more!
AlrightâŚ, I am really sorry as this is not helping anyone, but I have to ask: which GPU is that? XD ;). And more seriously, looking forward too!
I know its pretty ridiculous having such a large egpu but its the one I already had. It is the Gigabyte 4090 GAMING OC.
Nah! Itâs good to see some people using these monster GPUs. Do you use it for gaming or productivity?
Yes, I remember, the Gigabyte 4090⌠I think they really outdid themselves there. It calls for a deshroud with T30s, just as a statement ;). And it is as @Kyle_Tuck said, no reason at all to apologize for a 4090 - extremely good GPU.
Could you specify how you got it working under linux (even if it was not working that great?) I am using the same setup and the eGPU is not detected at all. I might also give up and go for the DEG-1 later, but I want to see how far I can get using my current setup first.
So, being that the AI 9 HX 370 (from what I am reading) has 16 pci lanes (a loss of 4) compared to the 7940hs at 20, can anyone speak to what this may do to eGPU performance via oculink, if any? Sorry if itâs a silly question, but thatâs deeper knowledge than I have understanding. I ask because I plan to switch to the new mainboard when it ships in November.
There will still be just 8 lanes on the connector from mainboard to expansion bay. So no nothing new.
Thanks for clarification. Out of curiosity, does this mean those who were using the actual internal nvme slot directly on the mainboard for oculink gain any additional performance by seemingly bypassing the lane restriction of the expansion bay? Or is it 8 regardless, internally or externally connected? (I know, defeats the purpose of the laptop and makes it even more of a frankenstein as it is all disassembled but I saw photos of folks doing that and wondered if there was an advantage)
M.2 NVMe are 4 lanes (generalizing, because there are some exceptions out there), so anyone using any of the M.2 slots NVMe slots are limited to 4 lanes of PCIe 4.0. That doesnât change with the AI 9 HX 370.
Thanks Kyle. I was looking at it from the wrong perspective. All this oculink stuff is new to me as of a few months ago and I continue to learn. I love this community for it.
How much performance would we gain from upgrading the board in terms of fps for gaming via oculink, leaving all else unchanged?
None if you arenât Cpu limited.
OK, thanks! I thought somehow better CPU always means a couple of more frames.
Then I guess nothing for me to buy in this updateâŚ
