Might be stupid question but can eGPU be connected with Framework Desktop?
Yep, Iāve had no issues getting it working via USB4 with an AG03 dock, works the same as the FW16 and FW13 and shows up as rebar enabled too.
Yes, running this eGPU with a AMD 7900 xtx via USB4. It took a little bit of setup and had to get the right drivers installed, but once configured, it runs great. Amazon.com: Maskedfish eGPU Enclosure Thunderbolt 3/4 USB4 40Gbps PD 85W Charging External GPU Dock Compatible with NVIDIA/AMD Graphics Cards on Win 10/11 Linux System, ATX Power Supply (MK-L18) : Electronics
I got mine working with oculink and my rtx 3090. More info to come
Hi @Rehuel you may also like to view this thread too: Request: verify dGPU support
Just adding a tip for anyone setting up an AMD eGPU:
Windows might default to a generic āBasic Display Adapterā driver or a conflicting Windows Update version that prevents the card from working properly alongside the iGPU. If it shows up but doesnāt work (Code 43 or just invisible), use DDU (Display Driver Uninstaller) in Safe Mode to wipe everything, then manually install the official AMD Adrenalin drivers. That fixed detection instantly for me.
2 Likes
any luck with Linux (ubuntu) ? my eGPU with Nvidia doesnāt seem to be recognized, but it works on my Framework 16 running Windows
might be Linux vs Nvidia thing but itās kind of weird, the card doesnāt even show up as pixie device
Iām kind of a noob though
1 Like
Iām running AMD 9070 xt via oculink on Windows and having issues with the driver. Windows sees my 9070 xt, but wonāt use it and always default back to the igpu. Did you use DDU to remove Framework provided gpu driver and then installed 7900 xtx driver from AMD website?
i am also struggling to get my setup to work. so i got a minisforum deg1 dock hooked it all with to the framework desktop with an minisforum eop4a card(with redriver). the framework detects the rtx 5060 ti in device manager (all nvidia drivers all installed) but i cant get it to run when i try borderlands 2 (image attached for the message) for example it will not start. i also cant seem to run the free demo of 3d benchmark. any help or tips welcome
Are you on Windows? Iām dual booting on FW16 with Fedora on a 2280 and Windows on a 2230. Fedora doesnāt have any issue detecting the 9070 xt and it just works unlike Windows.
I follow this youtube video to removed my framework provided gpu driver with ddu and reinstalled with 9070 xt driver that I downloaded from the AMD website. Additionally, ddu removed my chipset driver which I had to also reinstalled. The reinstallation worked as intended and Windows see the 9070 xt and default to it as the high performance gpu.
You can check at Settings > System > Display > Graphics > Change default graphics settings and see which gpu is set at high performance.
Everything went well, but now I have ran into another issue with āAMD drivers timeout errorā and Iām again at a lost.
yes i am on windows but i have an nvidia rtx card not amd
I have an update on my setup. I have had my Framework desktop since last November. I originally connected with an eGPU and USB4 for my AMD 7900 xtx. Then I added a second eGPU for my AMD v620, which worked great- for a while. after a few months of this setup, the eGPU for the v620 fried, then not long after the one for the 7900 fried. I replaced one of them and that fried too. I donāt really know what is going on, but a few weeks ago, I switched to Oculink and have the 7900 connected with this one below. The v620 is not connected at the moment, but Iām trying to find a viable path to get it running again.
With the latest firmware update, I think the stability of the internal x4 pcie port has been resolved. Iām now able to run with 2 eGPUs. One is on an oculink connector on the internal x4 slot as described by someone else. The second is using an m.2 to PCIe x4 adapter on the rear m.2 slot. The power supply on the oculink station powers both cards so everything powers on and off together.
If thereās interest, I can post photos, parts and my system setup.
1 Like
Iām very interested in how this is setup.
Did you retain the original case?
Yes, Iām using the framework case and power supply. I havenāt had any issues other than the internal x4 slot having issues with the eGPU until recently. Iām attributing that change to the latest 3.06 firmware, but itās possible that the fix was in a kernel update. Iām running Fedora 44.
Hereās the hardware I added. Iām not recommending anything here, just noting that is working for me.
- JMT ADT-K993G-BK7 PCIE 5.0 X4 / M.2 NVMe to OCuLink SFF-8611 External Dock Adapter (K993G-BK7-R29Q-30cm)
- be quiet! Power Zone 2 1200W, ATX 3.1, PCIe 5.1, 80+ Platinum Power Supply
- XRIKUI NGFF M.2 M-Key to PCIe X4 Expansion Card Adapter
- Glotrends 300mm PCIe 4.0 X4 Riser Cable - 180 degree straight
- 2 AMD Radeon AI PRO R9700 Graphics Cards
I cut holes into the case with a Dremel. I didnāt really know where to cut so my cuts could have been cleaner. If anyone wants to follow, Iāve given sizes and locations in the images below that should allow for cleaner cuts.
The card is somewhat lose because itās only held in by the slot. Iāll probably 3d print a brace to hold it at some point. My thought is to attach it to the fan screw.
Images:
2 Likes
I love it!!
Itās just as jank as my setup!!
In fact iāve added another 2x 5090ās via the usb ports using the AORUS RTX 5090 AI BOX
The hole you cut in the side panel with the fan header/pci slot. i assume thatās connecting to the m.2 slot? if so why did you not go with the OCulink connection?
Yes, thatās the m.2 to PCIe adapter. Itās not a fan header though, thatās power for the slot. Really, I just didnāt see the point in another oculink connection. Because Iām running both cards from the single power supply, I thought that the straight PCIe adapter suited my purpose better. Youāre right that a second oculink connector would have been less janky though.
1 Like
Iām using an eGPU, Iāll make an update if that can be useful to someone.
Using Miniforum DEG1. Had trouble in the past to boot (probably related to using a AMD igpu and AMD dgpu together), I think updating the kernel fixed it. I have it boot normally with Linux 6.18.9+deb13-amd64 and Linux 7.0.13+deb13-amd64. It still hangup for a while before post but it boot fine. Iām still using bios 3.02.
Here is a little bench I just made with ROCm10.0
llama-bench -dev ROCm0/ROCm1 -ts 15/85 -fa 1 -sm layer -mg 0 -m /unsloth/Qwen3.8-Flash-Next-GGUF/UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf
ggml_cuda_init: found 2 ROCm devices (Total VRAM: 143280 MiB):
Device 0: AMD Radeon RX 9070 XT, gfx1201 (0x1201), VMM: no, Wave Size: 32, VRAM: 16304 MiB
Device 1: AMD Radeon 8060S Graphics, gfx1151 (0x1151), VMM: no, Wave Size: 32, VRAM: 126976 MiB
| model | size | params | backend | ngl | fa | dev | ts | test | t/s |
|---|---|---|---|---|---|---|---|---|---|
| qwen4exp A3B Q4_K - Medium | 103.68 GiB | 176.94 B | ROCm | -1 | 1 | ROCm0/ROCm1 | 15.00/85.00 | pp512 | 324.80 Ā± 19.52 |
| qwen4exp A3B Q4_K - Medium | 103.68 GiB | 176.94 B | ROCm | -1 | 1 | ROCm0/ROCm1 | 15.00/85.00 | tg128 | 21.13 Ā± 0.19 |
1 Like
Hi Martin,
Nice setup. Did you touch any of the switches in the R29Q? Did all of worked for you? Iām struggling since I can make my Aoostar AG01 work with an M.2-OCuLink adaptor, but not with the R29Q.
Would appreciate some help here. Thanks!
Not sure what GFX card you are trying to load with the Aoostar AG01 but i check their website and they flat out say "Noteļ¼Not compatible with rx9070xt gpu devices".
They give this reason:
Q:Please provide explanation on why it does not compatible with 9070xt, i had this egpu , the 800w version and was going to upgrade my current setup to 9070xt
A:The 9070ās instantaneous power consumption is too high.